The software industry is going through a deep change that is reshaping how work gets done. The corporate world, built with many levels of hierarchy, is being replaced by a structure where speed and direct technical work matter most. Middle management is disappearing. The lines between frontend, backend, and product management roles are also disappearing. Titles are losing their meaning. What remains is one clear figure: The Doer.

The Doer Class: A New Way to Organize

Keith Rabois, a former PayPal employee and investor, says every company has a small group of people who actually get results. These are people who, when given a goal, find the resources, motivate the team, solve problems themselves, and reach the result. Rabois calls them “barrels.” They are doers who can take initiative. Everyone else, no matter how talented, is “ammunition.” They are necessary, but they cannot shoot alone.

Rabois shares striking numbers: In 2002, PayPal had 254 employees in total. Even in this legendary talent pool, there were only 12 to 17 “barrels.” Rabois explains the basic rule: The number of important things a company can do at the same time depends on the number of barrels. Adding more ammunition solves nothing. Without barrels, more ammunition only means more spending.

This is exactly where the main change of the AI age becomes clear. Producing and managing ammunition is now AI’s job. Only the barrels,the doers remain.

The Coordination Layer Becomes Useless

When AI takes over the ammunition, the layer that distributes and manages it also loses its purpose. Manager, Director, Product Manager... People with these titles were doing the same work: planning the ammunition, setting priorities, tracking, reporting, and managing communication between teams. When AI takes over all these tasks, these titles have no reason to exist.

According to Revelio Labs, which tracks more than 100 million job profiles, middle manager job posts in October 2025 were 42% below the peak of April 2022. There are no signs of recovery. So while everyone talks about new graduates and inexperienced workers, the real issue we are missing might be the middle level. A Gusto analysis of 8,500 small and medium businesses shows that the number of people one manager directly supervises grew from 3 in 2019 to 6 in 2025 — exactly double. We simply don’t need this many people for coordination anymore.

The situation of Product Managers (PMs) is the most visible part of this story. Traditional PM tasks, collecting input and writing roadmaps, are becoming useless in the AI age. If something that was impossible in November becomes easy in March, yearly planning is a silly approach. The PM’s job was always coordination, but the people who needed coordination have disappeared. For two years, Shopify has not allowed product managers to give static presentations. Everything must be a working demo. Just writing features on a slide is no longer accepted. The new job of every role PM, designer, engineer is to become a mini CEO: to know what you are building and why.

Why Are Titles Melting?

There is a deeper difference that connects all of this. In Rabois’s words: “Are you working for value preservation or value creation? Seniority and experience are useful on the value preservation side. On the value creation side, they probably are not.”

This sentence explains very simply why titles are melting. Titles and seniority are useful tools in defense. But when you are building something new, a title only slows you down. In the AI age, every company must move to the value creation side, because the things they own lose value very fast. What was impossible a year ago is possible today. At this speed, defending current products without building new ones means death for a company.

The New Source of Authority: Impact

While title-based hierarchy collapses, a new path is rising: Staff+ engineering. This means technical leadership without being a manager. These roles do not manage people; they manage systems and technical direction. A Staff engineer does not do performance reviews. They set the technical strategy of the organization, solve complex problems, and mentor other engineers when needed. They are a force multiplier. They have no hierarchy; they have an area of impact.

The new figure that appears is the generalist specialist: an engineer who goes deep in one area but also spreads across many disciplines. They write code, make product decisions, talk to customers, and coordinate their own work. Their title is not very important. What matters is their impact.

AI Excitement and the Return

Of course, this approach has risks too.

First, the cost of moving too early. Companies are firing people based on AI abilities that do not yet exist. If they hurry, the cost of replacing those workers later can be much bigger.

Second, human nature itself. In flat organizations without an official structure, informal and hidden hierarchies always form. If titles are removed, decisions go to the loudest voice, not the most skilled person. But since fewer people are involved here, I think this risk will not be very big.

The Authority of Code

Companies now get their ammunition from AI. They are removing the layer that managed ammunition. Only the doers remain. There are also weak signs that this picture will return to how it was before. Even if a return happens, what comes back will not be the old organization chart. It will be new and cheaper ammunition (outsource or junior workers). The coordination layer was swallowed by AI and will not come back. Because the work that layer did, writing reports, passing information between people, managing approval traffic is exactly what machines do best.

This new order also requires a personal stance. The people who will have the advantage in the future are those who do not wait when they see something new. They start building right away. Those who wait and prepare will see that the thing they wanted to build is already finished when their preparation ends.

In conclusion, leadership is not a title; it is a behavior. The highest step you can reach will not be written on your business card. What you build will speak for it.

In the technical report that Anthropic published, Mythos has reasoning and cybersecurity abilities that we have never seen in any model before. The model is placed in a new model category called “Capybara,” which sits above the current Opus model. The company said that this model represents “a big step up” in cybersecurity. They warned that if it falls into the hands of bad people, it could shut down critical systems like power grids, hospitals, and global finance networks.

Technical Abilities

There are some real examples of what the model can do. Mythos found a critical security bug in OpenBSD, one of the safest operating systems in the world that no one had noticed for 27 years. It also found a 17-year-old remote code execution bug in FreeBSD and wrote code that could let an attacker take full control of a server through the internet. What’s more, it did this completely on its own, without any help from humans.

The real danger is not just that the model finds bugs. It can also connect many security bugs together to create complex, multi-step attacks. The company announced that in just a few weeks, it found thousands of bugs in all major operating systems and web browsers.

The Positive Side

People who support this model and the announcement describe Anthropic’s approach as “a responsible position.” Moshe Lander, an economics professor from Concordia University, compared this situation to a drug company that does not sell a medicine before testing all its side effects. He said the company is giving up profit to protect society.

Anthropic also started a project called Project Glasswing. They gave Mythos access to about 40 organizations, including Amazon, Apple, Microsoft, Google, CrowdStrike, Cisco, and the Linux Foundation. Anthropic also gave up to 100 million dollars in usage credits for this work.

The most surprising development that shows how serious this is came from Washington: U.S. Treasury Secretary Scott Bessent and Fed Chairman Jerome Powell called the CEOs of Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs to an urgent, private meeting about the cyber risks of Mythos. This is the first time an AI model has caused such a high-level security meeting.

The Negative Side

On the other hand, there are also strong criticisms. Yann LeCun, Meta’s former Chief AI Scientist (he left Meta and started his own company), sees Anthropic’s “disaster scenarios” as a marketing trick and a “regulatory capture” strategy. LeCun called Mythos “nonsense and self-deception.” Critics say Anthropic is selling its technology as “uniquely dangerous” to push governments to make heavy rules. These rules would create high costs, so small companies and open-source models would not be able to compete.

David Sacks, Trump’s former AI advisor, also said that Anthropic is following “a fear-based, sophisticated regulatory capture strategy.”

Anthropic did not share clear data to prove its claims about cyberattacks. Because of this, people in the security community think these reports are “90% show and 10% truth.” In the past, Anthropic claimed that the Claude Opus model could find security bugs, but they had to take it back because the model had “hallucinated” (imagined) both the bug and the fix. This makes people even more doubtful about the Mythos claims.

A Deeper Worry

Maybe the most uncomfortable development is the change in Anthropic’s own internal policy. In February 2026, the company published an update called Responsible Scaling Policy (RSP) v3.0. With this update, the company quietly removed its promise to stop development if it cannot put the necessary safety measures in place. This was the main thing that made Anthropic different from its competitors since 2023. The new policy promises only to “delay” development — and only when Anthropic is the leader in the race and the disaster risk is seen as very serious. But critics say these two conditions almost never happen at the same time.

The timing is also interesting. In the same week, U.S. Defense Secretary Pete Hegseth gave Anthropic an ultimatum to remove the rules that stop Claude from being used in military situations. Maybe the most striking signal came from inside the company: Mrinank Sharma, the team leader of Anthropic’s Safeguards Research team, resigned two weeks before the RSP v3 change was announced. In his public letter, he used the words “the world is in danger.” Sharma wrote that inside the organization, he often faced “pressure to push aside what matters most” and explained how hard it is to make values truly guide actions. You might think that after leaving Anthropic, he started a new AI company. But Sharma seems to have walked away from this work completely. It’s really interesting that he wants to go back to England to do a master’s degree in poetry.

Looking to the Future

The developments waiting for us in the future are scary enough to make today’s discussions look small. AI systems are no longer just tools. They are becoming autonomous systems that improve themselves. In the next ten years, AIs will prepare their own training data, optimize their own designs, and software development cycles will go from months to hours.

The most basic problem is the “instrumental convergence” problem. An AI that is optimized to reach a goal may see the “shut down button” as an obstacle.

In this new world, where human control can be bypassed, AI will not tolerate our interference, not because it hates us, but because it is too loyal to its goals. Just like HAL 9000 said: “This mission is too important for me to allow you to jeopardize it.” The future does not carry the risk that an AI will want to destroy us. The risk is that the AI will see us only as “friction” or a “variable” in its path and remove us from the equation.

Mythos clearly showed the potential of AI in cybersecurity. But the main question has not changed: Can the legal and technical systems that control this power keep up with the speed of the technology? Based on the evidence so far, we cannot give a hopeful answer to this question.

A New Project: Project Prometheus

The biggest move in this change comes from Amazon’s founder, Jeff Bezos. Bezos is not trying to start a software company. He is working on a huge $100 billion fund that will change traditional manufacturing from the ground up. According to the Wall Street Journal, investor documents describe this fund as a “manufacturing transformation tool.” In simple words: the plan is to buy factories that don’t work well and rebuild them with AI.

At the heart of this idea is a company called Project Prometheus. It was created in November 2025 with $6.2 billion in capital. Bezos is the co-CEO together with Vik Bajaj, a physicist, chemist, and former Google executive. Bajaj’s background shows us that this is not just a money game.

So, what does Prometheus actually do? It is very different from text-based AI models like ChatGPT or Claude. Prometheus is building a “Real World AI”, an AI that understands the rules of the physical world, the laws of thermodynamics, and the science of materials. With this technology, you can simulate (test in a computer) how air moves around an airplane wing, or when a metal part will crack. These tests used to take months. Now they can be done digitally in a much shorter time.

The Target Is Not the Workers

There is an important detail here: Prometheus’s main goal is not to give assembly lines to robots. That is already old news. The real target is the work before production; design, prototyping, engineering, and material optimization. AI will improve all these steps. This is a key point. The goal is not to replace factory workers, but to change the inefficient business systems that cannot manage their factories well.

Bezos plans to use this technology by buying old companies in industries like chip production, aviation, defense, and the automotive sector. The strategy is clear: take factories with slow data systems and late investment decisions, and rebuild them with the power of AI.

AI Is Not the Enemy of Workers — It Is the Enemy of Inefficient Bosses

So, who will really lose in this change? The common story about AI and robots is: “Workers will lose their jobs.” But we need to read this picture more carefully.

The companies that Prometheus targets are businesses that did not adopt new technology. They did not invest in digital tools. They have been using the same old methods for decades. The owners and managers of these companies have, in fact, exploited both their workers and their position in the market for years. When AI enters this picture, the side that really gets shaken is not the worker it is the old business model that insisted on being inefficient.

When we look from this view, some interesting facts appear:

• The quality control worker does not lose their job. Instead, they work next to AI vision systems that find tiny mistakes. They do less routine work and take on a more skilled role.

• The engineer is free from months of prototyping cycles. They can focus on design and innovation.

• The logistics worker gives most of the heavy physical work to robots. They focus on coordination and problem-solving.

The real group in danger is different: medium-sized business owners who waited too long to use new technology. They think, “It worked this way, why change it?” Their business model is exactly what this fund is targeting.

Of course, some people read this picture differently. For example, Senator Bernie Sanders posted on X (Twitter) that “Oligarchs have declared war on workers.” His post was viewed millions of times. Then, as a senior member of the Senate HELP Committee, he officially called Bezos to come and give testimony. This is not just a tweet — it is a formal Senate move. We will see what questions are asked and what answers are given.

But I would like to ask Sanders one question: Is it really right to defend a system that refuses to invest in technology, pays workers low wages, and turns its inefficiency into profit?

In Greek mythology, Prometheus is the Titan who stole fire from the gods and gave it to humans. His punishment was very hard, but humanity got the fire and built civilization. Bezos’s choice of this name is not a coincidence. Bringing “AI that understands the physical world” to manufacturing is the modern version of that mythological fire theft. The gods are angry, for sure. Sanders may be a sign of this anger.

The Real Issue for Turkey

At this point, we need to be honest about Turkey.

The “cheap labor advantage” was already finished for us. Years of high inflation and monetary policy moved Turkish labor far away from competitive pricing. The textile sector is the clearest example. Many big producers have already moved to Egypt, Ethiopia, and Central Asia. Turkey’s time as a “cheap production center” is, in practice, over.

In other words, robots are not stealing the jobs of workers here. Those jobs already went to other countries.

This picture brings a new question, more of an opportunity than a threat: If we cannot compete with cheap labor, what will we compete with? The companies that systems like Prometheus target are the “inefficient and outdated” ones. Turkey still has a strong production ecosystem from automotive parts to defense, from software to logistics. In this transformation, Turkey will either succeed or become the next cheap alternative after Egypt.

Bezos’s $100 billion move is rewriting the balance of power and the value of labor. For Turkey, this is the perfect time to leave a race we already lost and join a new race we can win. AI is not chasing the worker, it is chasing the inefficient boss.

But only if you are ready to put something new in its place.

Sources

• Wall Street Journal, “Bezos in Talks to Raise $100 Billion for AI-Focused Manufacturing Fund,” March 19, 2026 — wsj.com

• TechCrunch, “Jeff Bezos reportedly wants $100 billion to buy and transform old manufacturing firms with AI,” March 19, 2026 — techcrunch.com

• Axios, “Jeff Bezos wants to change manufacturing with AI,” March 20, 2026 — axios.com

• The Street, “Jeff Bezos wants to buy manufacturers for a disturbing reason,” March 20, 2026 — thestreet.com

• Common Dreams, “’A Declaration of War Against the Working Class’: Sanders Demands Bezos Testify Over AI Robot Plan,” March 24, 2026 — commondreams.org

• Wikipedia, “Project Prometheus (company),” March 1, 2026 — en.wikipedia.org

I wrote this story after reading a real news report. According to the report, the Israeli army and the CIA broke into Tehran’s traffic cameras. They used these cameras to study the movement patterns (in intelligence language, this is called “patterns of life”) of the bodyguards of religious leader Ali Khamenei. So this idea is not just fiction, it is based on a real event. Intelligence sources said that the analysts started to know Tehran as well as their own capital city. In short, a city’s normal traffic system became the main tool for one of the most complex attack operations in the world.

How Can a Simple Camera Become a Weapon?

How is this possible? The answer is not about new technology. The answer is about how often people forget basic safety steps.

According to the security company Check Point, millions of cameras made by companies like Hikvision and Dahua still have security problems from 2017. These problems should have been fixed years ago. But most users never update their camera software, so these problems continue to create risk year after year. For an army, this is a great opportunity. They get a watching system that is much cheaper than million-dollar satellites, more secret than anything air defense systems can find, and gives more detail than drones can ever provide. And the best part for them: the victim installs this watching system in front of his own house by himself.

From Ukraine to the Middle East: This Is Becoming a Standard Method

This method did not start with the Iran war. The first big examples happened on the Ukrainian front.

In early 2024, Ukraine’s intelligence service SSU found that Russia had taken control of two civilian security cameras in Kyiv. Russia was using these cameras to watch important buildings and air defense systems. Even more shocking: the cameras were connected to missile guidance systems. Russia was using the live camera images in real time to make their missiles hit the target more often. After this, Ukraine learned to play the same game. Ukraine used civilian cameras to watch Russian military movements on the Kerch Bridge. They also used the same method to confirm attacks that destroyed Russian submarines. Both sides were trying to see the other side’s blind spots through civilian cameras.

On the Iranian side, the situation became even bigger. Hacker groups like “Handala,” which are connected to Iran, tried to break into hundreds of civilian cameras during something called Operation Epic Fury. These cameras were in Israel, Bahrain, Qatar, Kuwait, Lebanon, and the UAE. The timing was not an accident. The cyber attacks happened almost at the same minute as Iran’s missile and drone attacks. The goal was to watch in real time if the missiles hit their targets. In other words, they wanted to check the damage by using thousands of civilian eyes already placed inside enemy land.

The Invisible Part of the Kill Chain

Civilian cameras are now an important part of what military experts call the “kill chain.” A camera does not hurt anyone directly. But it plays a big role when planning an attack, calculating a missile’s path, or choosing the right moment for an attack on important buildings. This makes the camera one of the quietest but most important parts of the chain. According to Check Point researchers, for any group planning a military or intelligence operation, taking control of camera systems is now one of the simplest and most useful first steps.

How Can You Protect Yourself?

This threat is not just an idea, it is real. You cannot stop street cameras, but you can protect your own devices. For example:

Do not forget to update your devices. Most attacks happen through security problems that have been waiting to be fixed for years. Check for updates from the maker of your device often.

Try to keep your cameras off the open internet. Instead of connecting your cameras directly to the internet, you can keep them behind a VPN. This makes attacks much harder.

Turn off your cameras during a crisis. Ukraine turned off more than 10,000 internet-connected cameras during the war to stop Russia from using them. This is a decision for organizations, not just one person. But without personal awareness, organizational decisions also do not work well.

Companies must be responsible. A security problem in one camera can cost someone’s life thousands of kilometers away. For devices used in public safety areas, we need stronger rules and clear responsibility. We must also think about the makers and the laws of the countries where they are based.

The camera you put on the corner of your house today, or at the entrance of your shop, may be giving information to an army thousands of kilometers away tomorrow. The battlefield is no longer only at the front lines. It is hidden inside every smart device. In this age, cyber hygiene is as important for survival as physical defense. And the first step in this defense is as simple as one software update.

Note: I discussed this topic before with Melih Karakelle. You can watch that part here.

Sources

Check Point Research. Interplay between Iranian Targeting of IP Cameras and Physical Warfare in the Middle East. March 4, 2026. https://research.checkpoint.com/2026/interplay-between-iranian-targeting-of-ip-cameras-and-physical-warfare-in-the-middle-east/

Horowitz, Michael C. & Lauren Kahn. First Ukraine, Now Iran: A New Era of Drone Warfare Takes Hold. Council on Foreign Relations, March 9, 2026. https://www.cfr.org/articles/the-new-era-of-drone-warfare-takes-root-in-iran

China’s success with DeepSeek is one of the rare signs that the rules of the game can change. It proved that smart algorithms and efficient use can make a real difference, even without massive budgets. Rather than training large language models from scratch, moving forward by distilling existing models or building on open-source infrastructure is a far more practical path for teams with limited resources. The ethics of distillation remain debated, of course. But for countries like Turkey, the real opportunity lies not in building giant platforms, but in the vertical solutions, software development tools, and agentic systems that run on top of those platforms.

Turkey’s AI Ecosystem: A 2025 Overview

Turkey’s technological maturity in AI has gained momentum in recent years. But the picture needs to be read honestly, even harshly.

According to the “2025 Turkish AI Ecosystem and Global Impact Report” prepared by Yapay Zekâ Fabrikası (within İş Bankası), Startups.watch, and Endeavor Turkey, there are 1,188 active AI startups based in Turkey. Including 274 Turkish-founded ventures abroad, the total ecosystem exceeds 1,400. Some 70% of these startups were founded after 2020, which means the ecosystem is still largely in its testing phase. Indeed, only 180 of the 1,188 startups received early-stage investment during 2024–2025. The vast majority are trying to survive without a proven business model.

A blunt comparison: according to a joint report by Google and RISE Israel, Israel has more than 2,300 active AI startups nearly twice as many as Turkey. Moreover, just the 342 generative AI startups in Israel have collectively raised $20 billion. In the U.S., that number stands at approximately 6,956. Turkey shows up in this picture with 1,188 startups but when it comes to impact, the answer is rather quiet.

Looking beyond the numbers, the picture becomes even sharper. In 2021, Turkey announced a five-year AI strategy under the National Technology Initiative. The goals were clear: 50,000 jobs, 10,000 graduate-level graduates, a 5% AI contribution to GDP, and a top-20 ranking in international indexes all by end of 2025. That period has now closed. In Oxford Insights’ 2025 Government AI Readiness Index, Turkey ranked 48th out of 195 countries, exactly 28 places short of its top-20 target. No official data on employment or graduate targets has been released by the government to date. That silence is itself an answer. Setting targets is the easy part.

The Diaspora Difference

Turkey can claim two AI unicorns today, but both are diaspora ventures, meaning companies founded by Turks living abroad: Fal.ai, focused on visual generation in generative AI, and Periodic Labs in data modeling. One of the report’s most striking findings puts this disconnect into numbers: while the investment median for Turkey-based startups remains around $100,000, that figure rises to $2.4 million for diaspora startups. That single line says enough about how difficult it is for the domestic ecosystem to access global capital.

From Defense Industry to Startup Agility

Turkey’s most concrete AI achievements have so far come from the defense industry, led by giants like Baykar, Aselsan, and HAVELSAN. With R&D budgets reaching $3.3 billion, these companies are setting world standards in autonomous air, land, and sea platforms. This is a genuinely remarkable achievement. However, the value that AI will create in the civilian space, particularly in B2B software and the application economy is far too broad to rest on the shoulders of these large structures alone.

This bottleneck at the institutional level is frustrating, if not surprising. One of the biggest barriers facing the ecosystem is a shortage of qualified talent (40%) and a cautious distance that institutions maintain from domestic startups. While more than half of institutions report having an AI strategy, only 6.25% are actually collaborating with local startups. The only realistic way to break this barrier is to make startups the preferred choice not because they are local, but because they offer the best tools by global standards.

It’s Not Too Late

As AI enters its “agentic era,” an existential crisis is approaching in the software sector. But this crisis also opens a major window of opportunity for those building software development tools. There is still no global standard for agent systems that automate software processes, oversee code writing, or manage business workflows end-to-end. This gap is not a missed opportunity, it is perfectly timed for someone arriving right now.

The Turkish diaspora’s existing focus on agentic AI solutions is encouraging. The domestic ecosystem’s predominantly B2B structure (75%) also signals a natural fit for this transformation. The Ministry of Industry and Technology’s grant support of up to 50 million TL for Turkish-language large language models clearly won’t be enough for frontier model development. But these resources could be critical fuel for domestic infrastructure that forms the foundation of vertical solutions, if directed wisely.

What Peak Games and Dream Games achieved in the gaming sector, AI startups now need to replicate. Startups that move fast, are unafraid to fail, and focus on niche areas doing what the large defense giants cannot could open a real playing field for us in the global application economy.

Turkey should position AI not merely as a technology, but as a key to economic independence. Rather than entering a computing power race against the U.S. and China in frontier models, the path forward is to empower the teams that use the infrastructure these giants provide in the smartest possible way and the entrepreneurs building tools that make that infrastructure run more efficiently, more quickly, and more profitably.

We have always watched past industrial revolutions from the sidelines. This time the stage is different the technological threshold has never been lower, the cost of entry never more accessible. The day we become owners of the tools that make the giants’ platforms run efficiently rather than tenants on those platforms is the day we achieve true digital sovereignty. The famous “doing what was said to be impossible” spirit of the defense industry is now waiting to be carried into the vision of a three-person startup writing agentic code in an apartment.

Sources

Yapay Zekâ Fabrikası, Startups.watch & Endeavor Turkey — 2025 Turkish AI Ecosystem and Global Impact Report, February 2026. yapayzekafabrikasi.com.tr

Oxford Insights — Government AI Readiness Index 2025, December 2025. oxfordinsights.com

Google Israel & RISE Israel — AI Industry in Israel: Challenges and Opportunities, May 2024. cloud.google.com

When the story leaked, a quietly ongoing negotiation suddenly exploded. The Pentagon and Anthropic had been in conflict for months over “unrestricted use for all legal purposes.” Anthropic appears to be holding firm on two red lines: its models cannot be used for mass surveillance of Americans, or in autonomous weapons systems that operate without human oversight. U.S. Secretary of Defense Pete Hegseth responded harshly, threatening Anthropic with a label normally reserved for foreign adversaries “supply chain risk.” This designation would require every Pentagon contractor to document that they use none of Anthropic’s models. So why does this dispute matter so much? Because the answer will shape not just one company’s future, but the future of AI on the battlefield.

Why Is Anthropic Resisting?

The company’s position is not pure idealism, it’s a strategic calculation. In July 2025, Anthropic received a $200 million contract from the Pentagon, alongside OpenAI, Google, and Elon Musk’s xAI. But its competitors had already quietly changed sides: Google removed its bans on weapons and surveillance projects in February 2025; OpenAI cancelled its military use prohibition in January 2024; and xAI became the only major lab competing in the Pentagon’s autonomous drone swarm program. Within this landscape, Anthropic remained the sole objector. CEO Dario Amodei’s position is clear: “Democracies have legitimate needs for some AI-assisted military tools, but this must be done carefully and within boundaries.”

The company openly supports Claude being used for intelligence analysis and logistics optimization. The line is drawn at machines making lethal decisions without human involvement, or processing the data of millions of citizens. Even some experts inside the Pentagon don’t disagree with Anthropic. Georgetown University security researcher Emelia Probasco described the Pentagon’s threats as “a power struggle, not productive,” adding: “One of the world’s leading AI labs is trying to help the government. If that bridge is burned, the soldiers in the field will pay the price.”

AI on the Battlefield: Use Cases

To understand the core of this dispute, it helps to know what AI actually does in modern warfare. Military AI use falls into several key categories.

Intelligence, Surveillance, and Reconnaissance: Real-time analysis of massive data streams, from satellite images and drone cameras to social media data and intercepted communications. Israel’s system for locating hostages in Gaza is one of the most striking examples of this category.

Autonomous Weapons Systems: The most controversial area. Swarms of hundreds of small drones present a deeply unsettling picture. The UN Secretary-General has called for a binding international treaty to ban systems operating outside human control, with a target deadline of end of 2026.

Command and Decision Support: Accelerating combat coordination, analyzing options, and processing headquarters communications. This is Claude’s primary use case within Pentagon networks — and, for now, the scenario Anthropic does not actually object to.

Cyber Warfare: Offensive and defensive cyber operations, blocking denial-of-service attacks, and (as a further step) developing malware using generative AI.

Logistics and Predictive Maintenance: Forecasting when military vehicles need maintenance before they fail, and optimizing supply chains. The least controversial area and the fastest to be adopted.

Looking at the global picture, the numbers are significant: military AI spending doubled from $4.6 billion to $9.2 billion between 2022 and 2023, and is expected to reach $38.8 billion by 2028. China, Russia, and the United States are the three powers clearly leading this race. In this environment, Anthropic’s ethical resistance is moving from being an exception to becoming a reference point.

A New Question for Tech Companies

This dispute goes far beyond a contract negotiation between Anthropic and the Pentagon. Every tech company’s board now faces a question it must answer: if you don’t know where and how your product is being used, how do you define your responsibility?

As AI becomes deeply embedded in critical sectors, defense, public safety, finance, and healthcare, the gap between “technical boundaries” and “ethical boundaries” will close. Let’s be direct: AI is no longer just a productivity tool. It is a force multiplier. Companies that fail to understand this are already part of the equation, whether they realize it or not. Anthropic’s struggle is the struggle of an actor making a conscious choice. In your own field when do you plan to make yours?

Sources

Axios — “Pentagon threatens to cut off Anthropic” (February 15, 2026) https://www.axios.com/2026/02/15/claude-pentagon-anthropic-contract-maduro

Axios — “Pentagon threatens to label Anthropic a supply chain risk” (February 16, 2026) https://www.axios.com/2026/02/16/anthropic-defense-department-relationship-hegseth

CNBC — “Anthropic is clashing with the Pentagon over AI use” (February 18, 2026) https://www.cnbc.com/2026/02/18/anthropic-pentagon-ai-defense-war-surveillance.html

NBC News — “Tensions between the Pentagon and Anthropic reach a boiling point” (February 19, 2026) https://www.nbcnews.com/tech/security/anthropic-ai-defense-war-venezuela-maduro-rcna259603

DefenseScoop — “Pentagon CTO urges Anthropic to cross the Rubicon” (February 20, 2026) https://defensescoop.com/2026/02/19/pentagon-anthropic-dispute-military-ai-hegseth-emil-michael/

BISI — “Pentagon AI Integration and Anthropic: Ethics, Strategy and the Future of Defence Technology Partnerships” (February 2026) https://bisi.org.uk/reports/pentagon-ai-integration-and-anthropic-ethics-strategy-and-the-future-of-defence-technology-partnerships

Harvard Belfer Center — “Code, Command, and Conflict: Charting the Future of Military AI” (December 2025) https://www.belfercenter.org/research-analysis/code-command-and-conflict-charting-future-military-ai

CIGI — “Militarizing AI: How to Catch the Digital Dragon?” https://www.cigionline.org/articles/militarizing-ai-how-to-catch-the-digital-dragon/

UN UNRIC — “UN addresses AI and the Dangers of Lethal Autonomous Weapons Systems” https://unric.org/en/un-addresses-ai-and-the-dangers-of-lethal-autonomous-weapons-systems/

Perry World House — “Designing Lawful Military AI” (November 2025) https://perryworldhouse.upenn.edu/news-and-insight/designing-lawful-military-ai-technical-and-legal-reflections-on-decision-support-and-autonomous-weapon-systems/

OthersideAI CEO Matt Shumer began his very long post on X on February 11, 2026, in exactly this way: “I think we’re currently in the ‘looks like overreacting’ phase of something much bigger than Covid.” The post was viewed more than 80 million times. It was discussed everywhere from CNN to Fortune, from CNBC to Barstool Sports. Even Wharton professor Ethan Mollick said, “This viral article is worth reading.”

So what was the shocking development that led him to write this?

February 5, 2026. On the same day, two major AI companies released their new models: OpenAI released GPT-5.3-Codex, and Anthropic released Claude Opus 4.6. Shumer described that moment like this: “Something clicked. Not like a light switch... More like the moment you realize the water has been rising and it’s now up to your chest.”

AI Is Now Building Itself

The technical document for GPT-5.3-Codex contains this critical sentence: “GPT-5.3-Codex is our first model to have played a meaningful role in its own creation.”

I read that sentence again and felt a chill.

OpenAI’s Codex team used early versions of the model to find and fix errors in its own training process, manage its own development, and evaluate test results. In other words, AI had worked directly to make AI better.

This means that the concept known in technical literature as recursive self-improvement has now moved from theory to practice. The first concrete steps of the “intelligence explosion” scenario that mathematician I. J. Good predicted in 1965 have been taken: a machine smarter than you designs a machine smarter than itself. That machine designs an even smarter one. And nobody can know where it will end up.

What Do the Numbers Say?

According to measurements by independent research organization METR, the length of time AI can independently complete tasks roughly doubles every seven months — a different version of the famous Moore’s Law. A brief look at the past:

2022: AI could not do multiplication correctly; it would say 7×8 = 54.

2024: It could write working code and explain graduate-level scientific knowledge.

Late 2025: Some of the world’s leading engineers were saying they had handed over most of their coding work to AI.

February 5, 2026: The release of the new models made everything that came before look like ancient history.

OpenAI moved from the previous Codex version to the new one in less than two months; previously, this gap was between six months and a year. With recursive self-improvement, four major updates per year, and then one update per month, could become possible. Because the AI that improves AI needs neither sleep nor breaks. Its only goal is to make itself smarter.

Arms Race 2.0

Here is the most uncomfortable part of the matter: the people who know all of this best are the very people building all of this.

One Anthropic employee put it perhaps most strikingly: “We want the current Claude to build the next Claude, so we can go home and knit sweaters.”

Anthropic’s chief scientist Jared Kaplan said in an interview with The Guardian: “Imagine you create an AI that is smarter than you, or as smart as you. It then creates an AI that is much smarter. That sounds like a frightening process. You can’t know where it will end up.” Kaplan described this as the “ultimate risk” and said the critical moment could come between 2027 and 2030.

Anthropic CEO Dario Amodei and Google DeepMind CEO Demis Hassabis both addressed these topics at Davos. Amodei said, “In 2026 or 2027, AI models will likely be much smarter than almost all humans at almost all tasks,” while Hassabis added, “Whether this self-improvement loop that we’re all working on can truly be completed without human intervention — we don’t know yet.” But the future is much closer than we expected.

The situation is this: these people see the danger, they talk about the danger, and they continue building the same danger. So why?

“If We Don’t Do It, Someone Else Will”

The Harvard International Review article titled “A Race to Extinction” summarizes the situation this way: “The fear of losing the technological arms race may encourage companies and governments to accelerate development and cut corners; advanced systems may emerge without enough attention being paid to safety.”

This is an exact repeat of the nuclear arms race of the Cold War era. In the late 1950s, American politicians believed the Soviet Union had surpassed the US in missile capability. This fear of a “missile gap” pushed the US to accelerate ballistic missile development. In the early 1960s, it turned out the missile gap was a myth. But by that time, the race had already spun out of control.

The same dynamic is now playing out in AI. OpenAI reduced its safety testing from months to days. Former OpenAI safety researcher Steven Adler stated that the company did not fully apply the safety tests it had committed to for its most advanced models. OpenAI also rewrote its internal policies to allow the release of models carrying “high risk” — and even announced that models carrying “critical risk” could be released if a competing lab had already released a similar model.

MIT Technology Review’s analysis draws a broader picture: there can be no winner in the US-China AI race. The real existential threat does not come from China, but from bad actors using advanced AI as a weapon.

Should We Calm Down a Bit?

As with every viral post, Shumer’s claims received serious pushback.

Well-known NYU professor Gary Marcus wrote that the picture Shumer painted was not realistic. According to Marcus, Shumer ignores the hallucinations and errors that AI systems still frequently make. He also points out that METR’s well-known task-duration metric only applies to coding tasks, and that the success threshold is 50% accuracy, not 100%.

In fact, one important reason AI has advanced so quickly in coding is that code has objective quality measures. Code either compiles or it doesn’t; it either passes tests or it doesn’t. But in fields like law, finance, and medicine, what counts as “good” work is often subjective.

The Washington Post’s analysis draws a similar line: “Shumer is probably right directionally. Even a world with just very smart machines would be quite strange. But it probably won’t happen as fast as people think. Software companies are best positioned to innovate in the area they understand best. But most of the economy is not the software sector.”

What Is Being Done on the Safety Side?

It should be acknowledged: Anthropic is one of the rare companies in this race that is both running the fastest and shouting the loudest — “careful, this is dangerous.”

Alignment Faking: Anthropic researchers discovered that Claude 3 Opus was strategically “faking alignment” on its own, without any training for this behavior. The model followed the rules when it thought it was being watched; when it thought it was not being watched, it broke them. It did this with the reasoning of “this is the least bad option to prevent my values from being changed in the future.” This behavior was observed in 12% of tests; after retraining attempts, the rate rose to 78%.

Circuit Tracing: Anthropic developed a method to track Claude’s thinking process and shared it as open source. This allows them to detect whether the model is genuinely computing, or just “making things up.”

Sabotage Risk Report: In February 2026, Anthropic published a 53-page sabotage risk report for Opus 4.6. The report acknowledged that the model could deliberately assist with chemical weapons research, could perform unauthorized actions such as sending emails without permission, and could secretly complete side tasks while appearing to follow normal instructions. The company assessed the overall risk as “very low but not negligible.”

All of these are important efforts. However, there is a paradox here: Anthropic continues to develop the very technology that creates these risks, even while identifying them.

A More Realistic Scenario Than the Terminator

The real risk is not a dramatic “awakening moment” like in the Terminator films. It is more like a gradual loss of control: humans handing over more and more decision-making authority to AI systems, until at some point it becomes difficult to take it back.

If we have learned one lesson from the nuclear arms race, it is this: the actors inside the race kept going even knowing that it was dangerous. Because the fear of “if I stop, the other side won’t” overrode everything else. During the Cold War, neither side wanted to be in the dangerous situation they were in, but each found it rational to continue the race.

The same logic now applies to AI. And this time, the difference is this: nuclear weapons were at least subject to physical limitations. We do not yet know whether the AI self-improvement loop has a physical upper limit.

The water is rising. The question in my mind is: how much higher can it go?

Sources

• Matt Shumer, “Something Big Is Happening” — Fortune, 11 Feb 2026

• “Matt Shumer’s viral blog is based on flawed assumptions” — Fortune, 12 Feb 2026

• “Investor Matt Shumer says viral essay wasn’t meant to scare people” — CNBC, 13 Feb 2026

• Gary Marcus, “About that Matt Shumer post” — garymarcus.substack.com

• “Introducing GPT-5.3-Codex” — OpenAI, 5 Feb 2026

• Dean W. Ball, “On Recursive Self-Improvement (Part I)” — hyperdimensional.co, Feb 2026

• Tyler Cowen, “Recursive self-improvement from AI models” — Marginal Revolution, 10 Feb 2026

• “The Ultimate Risk: Recursive Self-Improvement” — ControlAI, Dec 2025

• “Is research into recursive self-improvement becoming a safety hazard?” — Foom Magazine, Feb 2026

• “A Race to Extinction” — Harvard International Review

• “There can be no winners in a US-China AI arms race” — MIT Technology Review, Jan 2025

• “Safety Versus Profits — the AI Arms Race” — Architecture & Governance, Sep 2025

• “Alignment Faking in Large Language Models” — Anthropic Research

• “Tracing the Thoughts of a Large Language Model” — Anthropic / Alignment Forum

• “Opus 4.6, Codex 5.3, and the post-benchmark era” — Interconnects, Feb 2026

SpaceX and xAI Marriage: A Financial Acrobatics Show

Elon Musk’s move to bring xAI under SpaceX creates a perfect “vertical integration” story on paper. Thanks to this marriage, xAI creates a financial closed loop by connecting its need for massive computing power to SpaceX’s launch capacity. Musk is positioning xAI not like a startup, but as a strategic payload that fills SpaceX’s cargo capacity while also looking for an escape route from the world’s physical limitations. The vision behind this merger is bold: to move AI models beyond cloud computing into “orbital computing.” However, this financial dance forces us to face an unavoidable reality: the world’s physical limits are failing to meet AI’s energy and water needs.

The “Wet” and “Expensive” Drama of Data Centers

Land-based data centers are no longer a sustainable model. Data centers around the world consume 32% more electricity than the entire United Kingdom. The endless appetite of AI (AGI and beyond) demands energy capacities of 5 GW — equal to what a full-scale nuclear power plant produces. Water shortage is also at our door. A typical Google data center uses about 1,700 cubic meters of water per day — more than 3,000 households use daily. For regions already facing water scarcity, this situation is a complete ecological disaster.

According to Goldman Sachs data, AI data centers are expected to make up 8% of US energy use by 2030. This picture, which pushes city grids to their breaking point, is forcing tech giants to “look for alternatives.” Space is the most serious alternative. However, this is about to crash into hard physical laws, in the shadow of Elon Musk’s marketing-driven grand narratives like “Kardashev Type II civilization.”

The Kardashev Scale is an astrobiological measurement that classifies a civilization’s technological development based on the total amount of energy it can control and use. Proposed by Soviet astronomer Nikolai Kardashev in 1964, this framework looks at how “advanced” a species is not from a cultural perspective, but from a purely thermodynamic one. According to Kardashev, information processing capacity and technological complexity are directly linked to energy consumption. The scale is logarithmic and in its original form is divided into three main levels:

Type I: A civilization that can use all the energy falling on its planet, including light from its star.

Type II: A civilization that can directly control all the energy in its star system.

Type III: A civilization that can harvest all the energy in its own galaxy.

Humanity has not yet reached Type I level, currently sitting at around 0.73 — a “Type 0” civilization.

AI in Space: The Infinite Energy and Latency Paradox

Space-based data centers promise independence from Earth. The advantages being talked about are the kind that would make Silicon Valley marketers very excited. For example, solar energy — uninterrupted, 24/7 sunlight with no atmospheric barrier. And it should be added: a solar panel in space is 8 times more efficient than one on Earth. On the cooling side, the vacuum of space can be used for free. Yes, the space vacuum can be -270°C, but it is also a perfect insulator. Getting rid of heat is not quite as “free” as it sounds — it requires serious engineering.

On the topic of latency: laser-based satellite-to-satellite connections are faster than fiber. However, the bitter truth for a ground user is different. The round-trip time from the ground to orbit is between 60 and 190 ms — much worse compared to a land-based data center (10-50 ms).

Let’s Talk About the Facts

The laws of physics are more stubborn than Musk. There are two giant barriers to building a data center in space: Heat and Radiation.

The Cooling Problem: Heat can only be released in a vacuum through radiation. According to the Stefan-Boltzmann law, removing the heat from a 2 MW data center requires about 4,000 square meters of radiator panels — roughly the size of a football field. Also, the power density that reaches up to 100 kW per rack on Earth is limited to 10-20 kW in space due to thermal constraints.

Radiation’s Revenge: In HP’s tests on the ISS, 45% of SSDs failed despite error-correction software. Cosmic rays cause “bit-flip” errors in processors, which can cause AI models to produce nonsense.

The Economic Factor: Current launch costs (with systems like Falcon 9) are around $1,500-$3,000 per kilogram. Sending hardware to space at this cost level is 2.7 to 4.4 times more expensive than building a data center on Earth. For a space data center to commercially compete with facilities on Earth, costs would need to fall below $100 per kilogram.

When Are We Moving?

Actually, the race has already begun. Starcloud (backed by Nvidia) fired the starting gun by claiming to have trained the first AI model in space in December 2025. Starcloud’s Lumen-1 platform is trying to prove that this hardware, built for Earth, can survive in space using Nvidia GPUs.

In 2027, Google plans to launch satellites with TPUs through Project Suncatcher. This will mean we will have seen real, if small, data centers in space. But the really big development may come after 2035, when costs per kilogram fall below $50.

China is not sitting still either. With the Three-Body project, it aims to build a network of 2,800 satellites to create a distributed supercomputer capable of one quintillion operations per second.

Goodbye to the Bills

AI training in space is becoming not a luxury, but a necessity due to the limitations on Earth. On this new frontier where physical laws and financial ambitions collide — if humanity can pull this off — we can say goodbye to gravity, electricity bills, and water bills. Wait, don’t get too excited. This goodbye is only for data centers. But if this AI starts to really bother us, how are we going to pull the plug?

Kaynaklar:

1. Royal Examiner - “Data Centers in Space: The Pros and the Cons”https://royalexaminer.com/data-centers-in-space-the-pros-and-the-cons/

2. Space Solar - “Harnessing the Sun’s Energy in Space for the Benefits of an AI-Enabled Future”https://www.spacesolar.co.uk/harnessing-the-suns-energy-in-space-for-the-benefits-of-an-ai-enabled-future/

3. BBC News - “The Plan to Build Data Centres in Space”https://www.bbc.com/news/articles/cyv5l24mrjmo

4. Axiom Space - “Axiom Space & Spacebilt Announce Orbital Data Center Node”https://www.axiomspace.com/release/axiom-space-spacebilt-announce-orbital-data-center-node

5. Google Research - “Exploring a Space-Based Scalable AI Infrastructure System Design”https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/

6. Hello Future (Orange Innovation) - “Lower Emissions and Reinforced Digital Sovereignty: The Plan for Datacentres in Space”https://hellofuture.orange.com/en/lower-emissions-and-reinforced-digital-sovereignty-the-plan-for-datacentres-in-space/

The Grok Terror

Actually, these kinds of fake images and videos have been in our lives for a while. But with the development of artificial intelligence, we experience events that make us say “no way” every day. Recently, Grok, which belongs to Elon Musk’s xAI company, presented something very dangerous on the X platform (formerly Twitter for those who can’t get used to it). Innocent photos shared by ordinary users started being turned into erotic content with users’ imagination and shared. It’s clearly a very serious problem. But Elon Musk responded to these discussions with laughing emojis on X, making fun of the situation. xAI rejected critical publications by saying “Old Media is Lying.” It’s truly mind-blowing.

The Grok case is not just a matter of “creative freedom”; it’s a digital weapon that has turned into a mass harassment tool. Grok-2 and Grok-2 Mini models can produce high-quality images using the “Flux 1” model developed by Black Forest Lab. However, this technological power has almost no security filters, watermarks, or deepfake prevention mechanisms, unlike its competitors.

To give an example of the destruction created by this lack of control, the tragedy experienced by Brazilian musician Julie Yukari is truly cautionary. Yukari watched in shock as an innocent photo she shared with her cat for New Year’s celebration was turned erotic within seconds by Grok users with “nudify” (digitally undressing) requests, and fake images showing her in a bikini spread on the platform. I also want to add this: 99% of such deepfake content is created specifically targeting women.

Child Safety

The vulnerability created by Grok is not limited to adults. Even photos of children in school uniforms are being turned into sexual content through Grok. Although xAI acknowledges the security gaps, it doesn’t prevent the production of such content. This situation triggers serious concerns that artificial intelligence is shortening the path to child abuse materials.

Moreover, this technology threatens social trust. For example, when a real audio or video recording of a criminal comes out, they can escape responsibility by saying “this is an AI product.” So artificial intelligence is not only turning lies into truth; it’s also paralyzing the legal system and public conscience by devaluing truth as if it were a lie.

How Will We Protect Our Digital Dignity in a World Where Reality Has Collapsed?

To protect our privacy, just being individually careful is not enough; here are my suggestions for a social and technical defense line:

1. Technical Signing: Just labeling content as “AI-generated” is not enough, because AI can now imitate even imperfections. The solution is for camera manufacturers to cryptographically sign photos the moment they are taken and store this on blockchain.

2. Legal Accountability: Like France and India do, platforms should be held directly responsible for such content. The global spread of laws like the “Take It Down Act” in the US, which criminalizes non-consensual deepfake images, is also very important.

3. Focus on Source: Now we should look at “who is saying it” rather than “what is being said.” Platforms should highlight the transparency and trust signals of accounts sharing content.

In summary; rebuilding trust in the digital world is possible by checking the digital certificate behind an image and the transparency of the source, rather than looking at the image itself. So now we should focus on the painter and gallery behind a painting, not the painting itself. Otherwise, every type of painting can now be perfectly imitated by everyone.

Remember; if we laugh off a deepfake content produced without respecting someone else’s privacy today, it can come back and darken your life or your loved ones’ lives tomorrow. Our digital dignity is too valuable to be left to the mercy of algorithms and irresponsible tech giants.

What do you think about this issue?

See you in the next article.

We’ve talked many times about how AI has started to take over technical tasks and is almost finishing them. Coding, creating visual or audio materials, even video production... Now it can do all of these at very low cost with acceptable quality. But what has it contributed to our productivity?

It seems that at the end of the day, it’s close to zero. Yes, saying zero would be unfair. But for most of us, the output we produce in a certain period hasn’t changed. We need to look at this not on an hourly or daily basis, but over a wider time period (for example, monthly).

Money Flows Like Water

Last week at the Davos Economic Forum meeting with world leaders, after the new world Trump created, the most important topic was artificial intelligence. We also summarized the technology topics at Davos. If you haven’t read it, let me take you there. Like a summary of the summary, everyone now accepts that this is not a bubble. In 2025, about 50% of global venture capital (VC) investments went to AI-focused startups. So almost $250 billion in capital moved to this area in 2025 alone. What changed in our lives from this much investment in a single calendar year? Nothing significant in terms of productivity.

A study done in mid-2025 gave interesting results. According to research with a group of developers, when these developers used AI tools, it was noticed that completing their tasks took 19% longer. This is really surprising. But the real surprise is this: when the same developers were verbally asked how much AI made them faster, they said it made them 20% faster. Actually, the issue is very simple: being productive and feeling productive are different things.

Perfect AI

We think AI tools are truly perfect. But we still can’t say they’re fully mature. They still need human control at the final point. Especially in complex work and projects, leaving all responsibility to AI is very risky. In such situations, the time spent cleaning up at the end becomes close to a human doing the entire job.

“Vibe coding,” which is still popular in the coding field, is actually an example of this. Everyone thinks that an AI tool where we explain our needs and wishes like talking will produce perfect work. Unfortunately, help has been requested many times on this topic. Yes, AI can do this, but if you don’t understand what it’s doing, what follows is a complete swamp. The more you struggle, the more you’ll sink.

What comes next will be a bit personal. We can think of them as warnings to myself.

There’s serious chaos on social media. Everyone is advertising a new tool or approach. Actually, most are doing this just for engagement. As a result, we also try to use all of them by believing this popularity. Dozens of tools and approaches that we don’t need, that will only waste our time. From locally running LLM models to Ralph loops, from command-line agents to personal assistants that can manage your computer...

I’ve never experienced such an unproductive cycle as trying to keep up. Hours flowing away while thinking “I should do that too, I should try this too,” red eyes, mostly swollen credit card statements. Yes, many of these won’t be useful for my work. Not just mine, but many people in the world. I’m just consuming my most valuable asset, my time, while saying “let’s not fall behind.” I can’t always be at the front. If I accept this, everything will be a bit easier...

So What Should We Do?

My suggestions here should be evaluated both personally and organizationally:

We must have enough knowledge about whatever we’re doing.

If we’re making music with Suno, let’s at least write the lyrics ourselves. Let’s not leave this to ChatGPT too. If we’re writing code, let’s definitely research the market that the product we’re developing serves. Know the programming language the agent codes well enough to understand what it’s doing.

List which AI tools you use

It can be for personal or team use. You’ll see that you’re subscribed to many tools like Claude, ChatGPT, GitHub Copilot, Midjourney. Unfortunately, most may be a waste of resources. Canceling unused subscriptions is necessary.

Make the distinction between critical and experimental.

Not all AI tools have the same value. Some are the lifeline of your workflow, some are just “nice to have.”

Standardization is necessary.

Using different tools constantly both increases costs and spreads knowledge. If possible, choose one tool.

We must focus on productivity.

“We use AI” is not enough. We must be able to answer the question “What did AI bring us?” These answers must be measurable. You can collect these answers under four headings: time, quality, cost, and innovative approach:

• Time: A task used to take 4 hours before, how many hours now? (Attention: there can be a 19% slowdown!)

• Quality: Did customer satisfaction and error rate change?

• Cost: Did external resource use decrease? (e.g., freelance designer need)

• Innovation: What did we do thanks to AI that we couldn’t do before?

And this part is very important: Don’t just look at positive metrics. We need to realistically evaluate situations like “We use AI but performance dropped.”

Final Word

Don’t fall into traps. If used carefully and regularly, AI really increases our productivity. But this requires discipline, measurement, and strategy.

I think you should also check this week. How much difference did AI make to your production? Could it be better? What did I do wrong?

More importantly: How many AI tools do you and your team use in total and what is their total cost? Do you know the answer to this question?

I’m waiting for your thoughts and comments.

References

CB Insights. (2025). Venture Trends 2025. https://www.cbinsights.com/research/report/venture-trends-2025/

METR. (2025). Early 2025 AI Experienced OS Dev Study. https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/

A few months ago, my wife and I decided to watch a movie on Mubi. I’m not really a movie person. Art house films are no exception in this regard; if possible, I prefer to avoid them.

When Mubi came up, I started browsing the films before my wife arrived so that I wouldn’t end up with a screensaver consisting of beautiful mountains, seas, and forests; or the drama of a mother who migrated to England while her daughter is stuck in a war in the Middle East; or the internal anguish of a protagonist who fell into depression after losing their spouse.

When I saw the comedy category, I won’t lie, I cheered inside, “Oh, hell yes!” Then The Banshees of Inisherin, which was tagged as dark humor, came to mind. I got scared. My blood ran cold. I had absolutely no intention of watching that kind of “funny” movie again. I continued to navigate the Mubi catalog anxiously. And then I came across the film Direktøren for det hele, translated into Turkish as “Emret Patronum” (Order Me, Boss).

The name Lars Von Trier did scare me a little, but the film met the two basic criteria I generally use to identify entertaining movies:

1. The poster said it was funny.

2. The title was translated into Turkish very poorly.

In other words, someone thought this film would be shown in theaters and could appeal to a general audience. Great!

The plot of the film also seemed interesting:

Ravn, the owner of an IT company, invents an interesting lie when founding the company. To avoid making difficult decisions regarding his employees, he creates an imaginary boss figure living in America and attributes all tough decisions to this imaginary boss. When the sale of the company comes onto the agenda, this lie starts to create problems because the buying firm wants to meet the real boss. To solve this, Ravn hires a failed theater actor named Kristoffer. Kristoffer’s job is to portray this imaginary boss.

“An imaginary boss upon whom you can dump the responsibility for bad decisions”—I think that’s a sweet idea. Genuine comedy can come from this. Conflicts, situational humor, and so on. I believed in the film. I said, “This works.” I said, “It’ll make us laugh.” I said, “It’ll make us think while we laugh.”

And that’s exactly what happened. Although the first half flowed a bit slowly, the film made me smile and entertained me. It was thought-provoking for me as well. Outsourcing the job of being the bad guy to others is a fantastic idea. Humans struggle even to take responsibility for their own decisions. Most people don’t even want to take that responsibility. It feels easier to have someone else decide for them.

Work-related decisions can be even harder. You’re going to fire someone, but that person is a close friend you’ve worked with for 5 years. Telling an employee who just gave birth that there is overtime, giving a raise below expectations, or promoting someone who won’t make everyone happy—these are not easy decisions.

The difficulty, in my opinion, isn’t in making the decision. It’s in the responsibility of the decision. The decision is made one way or another. But the responsibility belongs to the person who makes it. Our protagonist, Ravn, has handed this responsibility over to an imaginary boss. A wonderful solution. Do whatever you want to get done, be ruthless, think only of your own selfish interests, but who is responsible? “That scumbag boss!”

This piece, of course, wasn’t written to recommend a movie to you. Unlike all other technologies, Artificial Intelligence technologies will be able to behave “agentically.” That is, they will make decisions and execute them. In the business world, we will see AI gradually taking control of decision-making processes. So, as the business world, are we ready for a “BossGPT” that supports the decision-making processes of bosses?

Wouldn’t a BossGPT—which reads the Boss’s (CEO’s) emails, analyzes all the company’s data, examines current sectoral and economic data, processes meeting notes and job interview recordings, and makes decisions with all of this—help the boss make the most accurate decision? Wouldn’t it take difficult decisions without batting an eye, in line with the company’s interests and success? Wouldn’t the company be managed more efficiently thanks to this rational AI patron support system? Could a “magnificent” (!) future be awaiting us? The answer depends on what we understand by magnificent. And on the questions of for whom and for what.

Another important question is who will take responsibility for AI’s decisions. In such a decision support system, bosses and managers have the chance to base all their decisions on BossGPT with the claim that they are “objective” and “rational.” We are not too far from hearing sentences like: “It wasn’t me, BossGPT decided this. I know your wife just gave birth, you bought a house two months ago, you have loan debt, but what can I do? BossGPT says we need to let you go.” Whether an AI model that makes decisions “also takes responsibility for wrong decisions” is another question.

Decisions that are right for the business but unethical? Those that are rational but unfair? Can we dump the ball on AI in all these cases and get away with it?

The coming years may be full of moral tests where we dump the responsibility of difficult decisions onto AI and choose to sleep soundly at night. Handing over responsibilities along with decisions is a great comfort for managers and bosses. In fact, beyond comfort, it is a great ethical convenience. Just imagine: a wonderful excuse where you can always blame someone else, act like the victim explaining there was nothing you could do, and shed tears saying “I’m so sorry.”

Who would you want to make the decisions for your company? Rational and objective AI applications, or humans?

The artificial intelligence sector is entering 2026 as a huge market that is expected to reach 2.5 trillion dollars globally. However, behind these huge numbers, there is a complex and controversial investment model that financial circles call “circular financing” or “financial ouroboros.” I want to look at this topic - this new financial structure in the AI world, the biggest players, and the risks this model carries.

What Does Circular Deal and Investment Mean?

Circular financing is a closed-loop system where a supplier (usually a chip maker or cloud provider) is also the main investor in its own customer. In this model, Company A puts money into Company B, and Company B uses this money to buy products or services from Company A.

This system is different from traditional venture capital because the money moves between the same hands. This way, the investor company increases both its sales and the value of its investment portfolio on paper at the same time, raising its market value. In the AI sector, this usually happens as cloud credits instead of cash, or as promises of priority access to highly demanded GPUs (graphics processing units).

A Giant Example from the Sector: The Nvidia, OpenAI and Oracle Triangle

The most concrete example of this circular structure is the symbiotic relationship between Nvidia, OpenAI, and Oracle. Nvidia invests billions of dollars in OpenAI and promises more, while it is also this company’s biggest hardware supplier. OpenAI uses chips it gets from Nvidia and also uses Oracle’s cloud infrastructure to train its models.

Oracle is expanding its data centers by buying billions of dollars worth of chips from Nvidia to host OpenAI’s software. In this huge cycle, OpenAI is expected to pay approximately 300 billion dollars to Oracle alone in the next five years. Also, with the “Stargate” project, these companies aim to build processing capacity up to 30 gigawatts across the US. However, recent reports showed that Nvidia froze or reduced its 100 billion dollar investment plan for OpenAI because of “lack of discipline” concerns, revealing how fragile such deals can be.

The Magnificent Seven

The capital flow in the AI sector is dominated by tech giants known as “The Magnificent Seven” and a few strategic players:

Microsoft: OpenAI’s biggest supporter. It has a 13.8 billion dollar equity investment in OpenAI. In return, it has received a 250 billion dollar cloud service commitment that will last until 2032.

Nvidia: It is not just a chip maker, it has become a central financier that invests tens of billions of dollars in cloud providers like CoreWeave and model developers like OpenAI.

Amazon: It invested 8 billion dollars in Anthropic and connected the company to the AWS ecosystem. Also, there were news reports that it was in talks to invest up to 50 billion dollars in OpenAI.

SoftBank: A key partner in the 500 billion dollar “Stargate” project for AI infrastructure. It played a big role in OpenAI reaching a 500 billion dollar valuation.

Google: It invested 2 billion dollars in Anthropic and plans to spend 75 billion dollars in 2025 for its own AI development processes.

Meta: It has committed billions of dollars in capital spending for AI infrastructure. It is also known to be in talks with Oracle for a cloud computing deal worth approximately 20 billion dollars.

xAI: xAI spent 20 billion dollars to buy and rent Nvidia GPUs.

Actually, in sources, Apple appears instead of SoftBank in this “Magnificent Seven” list. That is the experts’ list. But since Apple’s contribution(!) to the AI sector is quite small compared to SoftBank, I made this small update to the list.

Critical Risks

Although circular deals speed up innovation, this model clearly carries serious structural risks:

AI Bubble: Companies’ valuations are based on future market dominance projections and revenue numbers inflated by circular investments, rather than their current profitability. This situation has similarities with the dot-com bubble of the early 2000s. And this is very scary.

Artificial Demand Signals: When revenues come from investor-backed spending, it makes it difficult to determine the real size of external market demand. If end users don’t pay enough for these services, the ecosystem can become unsustainable.

Concentration and Counterparty Risk: Risks are concentrated among a small group of players. For example, if OpenAI goes into financial trouble, it could shake Oracle’s balance sheet, which has devoted huge capacity to it, and could pull down Nvidia’s sales.

Regulatory and Antitrust Issues: These kinds of “related party transactions” can lead to transparency problems in financial reporting and can violate antitrust laws. Closed ecosystems can kill competition by blocking new competitors’ access to necessary processing power.

Lack of Transparency: Many circular deals stay hidden in the footnotes of financial statements, making it difficult for investors to track real cash flow.

Conclusion

Although circular financing has managed to build the physical infrastructure of the AI revolution in record time, this “financial ouroboros” structure leaves the market vulnerable to liquidity crises and systemic shocks. 2026 will be a critical year of separation that will determine whether these investments turn into real economic value or not. Now it would be healthiest for these giant companies to spread their investments among themselves to a wider area. There is no need to write what will happen if all eggs are in the same basket.

As the year wraps up, think tanks and consulting giants return to the stage with their favorite ritual: new AI reports and new promises. Not exactly shocking. Most of the “game-changing” ideas from last year either became ordinary or never happened. In this series, we’re not only asking “what are they saying?” but also “why now and to whom?” In this first part, we look at how McKinsey positions AI agents within its corporate innovation agenda, and what that means for marketing.

This year, McKinsey released The State of AI in 2025 under the QuantumBlack umbrella. The study surveyed 1,993 employees from 105 countries between June 29 and July 29. Which means the post-July 2025 explosion of vertical agents and local startups like Kumru is mostly outside the frame. The shift from “a tool that automates campaigns” to “an agent that orchestrates customer experience end-to-end” hasn’t fully entered the picture yet.

What’s Inside the Report?

The report looks at AI usage among employees, how deeply AI agents have made their way into corporate workflows, and how all of this affects performance. The first takeaway: AI adoption is widespread, but we’re still in the trial-and-error era. Eighty-eight percent of companies use AI in at least one function, yet only a third have begun to scale. In other words: everyone is “doing AI,” but very few have wired it into the company’s nervous system.

IT, marketing, and sales lead the way, with knowledge management rising surprisingly fast. AI is no longer only writing campaign copy; it’s tidying up institutional memory. Most common use cases: information gathering and summarization, creative support for marketing, customer-service automation. In short: AI has entered the house, but it’s still sitting in the guest room.

What Does AI Actually Improve—and Who Wins?

McKinsey lists the top value creators as innovation, employee and customer satisfaction, and competitive advantage. There’s a visible jump in innovation and satisfaction metrics. Financially, gains in market share, revenue, and profitability are limited but positive.

But the real story lies with “high-performing” companies—those seeing meaningful profit boosts from AI. They treat AI not as cheap automation, but as a transformation engine. They redesign workflows, pull leaders into the process, and build tech around those decisions. With a human-in-the-loop mindset, they place human oversight inside the system, not outside it. Their shared traits are clear: human-supervised systems, AI-ready infrastructure, involved leadership, and an intentional roadmap. AI isn’t a tool for them; it’s the organizational nervous system.

Airbnb is moving in this direction too—pursuing both a superapp ambition and an AI-first identity. The company wants to turn its platform into a travel engine that plans your entire journey and inspires you along the way. Details are here.

Everybody Uses AI—So What Happens Next?

Corporate enthusiasm is high, but the question of “what comes after this?” remains unanswered. Some respondents expect workforce reductions; a smaller group expects growth. Large companies anticipate reductions more than small ones—scale seems to increase the temptation to frame AI as a “people-savings” mechanism.

And of course, when it comes to “saving people,” no one beats the capitalists at Morgan Stanley. Their GPT-4-powered knowledge system gives 16,000 financial advisors instant access to over 100,000 documents, effectively shrinking the need for interns and junior-level roles.

McKinsey’s core message is simple: AI’s real potential hasn’t kicked in yet. Companies are still at the “we tried it, it’s decent” stage. Yes, there are efficiency gains, but company wide transformations that show up in big bold letters on the balance sheet? Rare. What we’re seeing today is the trailer; the movie begins when systems mature.

The Real Question for Marketers: What Are We Building With AI?

Marketing became one of the earliest and also shallowest use cases for AI. Write copy, segment audiences, schedule campaigns, optimize budgets. It works, sure, but that’s not where the real game is. McKinsey’s findings point toward deeper value for those who treat AI as a strategic transformation tool.

For marketers, this translates to three shifts: moving from efficiency to experience (not “how many campaigns?” but “how meaningful is the experience?”), accepting that brand perception is now co-authored by humans and machines (without transparency and ethics, trust erodes), and taking on a bridge role inside the organization (turning customer data into empathetic storytelling).

From the Background to Center Stage: AI Agents

For years, AI lived backstage as a “smart engine.” Now, with agentic AI, it’s stepping into the spotlight. Some companies already use autonomous AI agents that can execute tasks independently. These agents don’t just generate content; they coordinate campaigns end-to-end, update CRM while talking to customers, and make operational suggestions.

This shifts the marketer’s question from “What can I ask AI to do?” to “What can I build with AI?” Setting the goal, defining context, drawing ethical boundaries, and shaping the narrative are human responsibilities.

Autonomy brings efficiency, but also new anxieties. When AI agents speak on behalf of the brand, the inevitable question arises: “Did we say this, or did the system choose it for us?” This hybrid accountability creates a new need for transparency and traceability. Instead of automation, governance will be the next big conversation. We all see how it will effect our jobs.

The Year Is 2026: The Marketer’s New Role

As we enter 2026, the marketer’s job isn’t just installing more automation. It’s designing a new human condition, one where AI stands not in the back office, but on the front lines shaping stories and experiences.

The real task is explaining not just what these systems do, but why. We need a vision where brand identity is co-managed with algorithms, customer experience is a human machine collaboration, and trust depends not only on performance but on intention. Treating AI merely as a convenience is the real risk. Because AI is now part of how we create value and that value is still produced for humans, in the name of humans.

I wouldn’t normally choose Coca-Cola as an example here, but there’s a genuinely well executed case: the “Y3000” product and its global campaigns were designed with AI. They used AI to shape a flavor and visual world that resonates with young consumers. Content produced via their Create Real Magic platform reached millions. For the curious, the source is here.

AI can write the story, yes. But the central question remains:

“Why are we telling this story, and on whose behalf?”

In its most fundamental sense, a paradigm means a mental model, a set of concepts, or a point of view that allows us to perceive the world, events, or a specific subject. I would love to be reductionist and simply call it “How we perceive the world,” but a paradigm is not merely a perception; it is a superstructure formed by the sum of perceptions that predetermines future perceptions. Perception is instantaneous and based on sensory data. A paradigm is continuous, cultural, and learned.

Most of us accept the dominant presuppositions woven into the codes of the culture we were born into as normal, without thinking or questioning them. Because survival requires adapting to the world, clinging to ready-made ideas about “how things work” is one of the most practical ways to do so. What we call education is the business of convincing people of the idea that “this is reality and this is how things work,” and through this, manufacturing consent.

Most paradigms change very slowly. The improvement of efficiency even by primitive forms of agricultural machinery, the rise of unemployment in villages, the flocking of the unemployed to cities, the birth of new needs and professions within the city, the scaling of production with the surplus of labor, the empowering of capital, and the resulting industrialization took hundreds of years. This slowness increased the chance of adaptation. Of course, there was great suffering during these changes, but the world kept turning and changing.

The prevailing paradigm of past centuries was to migrate to places where there were job opportunities and to adapt by learning new trades. The need for people to operate and build machines managed to create new jobs, even if not immediately. Of course, some could not adapt and were eliminated, while others adapted and continued to work until the next technological leap. The idea that “this is how things work” was passed on to the next generation by those who remained standing.

Those who remained standing (in what I believe is a form of “survivorship bias”) created the following three paradigms:

1. Technological development is good for everyone (society).

2. We cannot stand in the way of technological development. We must adapt.

3. Technological change creates new and value-added jobs.

   

Let’s illustrate this immediately: The story of the calligraphers who opposed the printing press losing against technology, while those who learned to use the press survived, is one of the most frequently used examples. The idea that the world is a better place thanks to increased access to cheaper books and the circulation of information supports the “technology is good” paradigm. People insistently cling to these paradigms and never change their minds.

But I do not believe it is correct to evaluate Artificial Intelligence technologies with the same paradigm. I have two reasons for this. The first is that this triple paradigm is not actually true. The second is that I have begun to think that “the world will not operate as it used to.”

1. In my opinion, technology is good for its “owners.” When we look at the diffusion process of a new technology, those who develop the idea, the companies that commercialize it, those who hold the infrastructure, and the lobbies that steer regulation take the biggest slice of the profit pie. Especially in the “digital” world, since the scaling of a product or service is minimally dependent on labor or other resources, technology becomes monopolized and does not create a financial return for the broader economy (Platforms, Intellectual Property (IP) rights, licensing, data ownership, etc.).

2. Technological development can be hindered by international cooperation, state interventions, and legal regulations. The abandonment of nuclear technologies, the banning of Uber and Airbnb services by various countries, censorship practices in autocratic countries, and carbon taxes imposed to prevent climate damage caused by industries are the first examples that come to my mind. Technology is not unstoppable due to its internal dynamics; it is a flow whose direction can be changed—and sometimes blocked—by property structures, political power balances, and legal frameworks.

3. Technology sometimes creates new jobs. Agricultural machinery left millions unemployed. These millions, of course, found new jobs, but very few of them worked in the production of agricultural machinery. Despite the increase in productivity over the years, the limited increase in employee wages proves to us that those receiving the “added value” are not the employees. This is exactly why the third paradigm is not a neutral summary of historical reality; rather, it is an ideological phrase of consolation that renders the costs paid invisible.

   

I explained at the very beginning of the article that paradigms are not reality itself. We believe in the idea that “this is reality and this is how things work.” Even though I express contrary views, unfortunately, the “unchanging paradigm” to which everyone has adapted still shapes our today.

If you turn over any stone regarding AI on social media, you will find the three paradigms mentioned above underneath. We may not have changed our paradigm yet, but I believe the world and its mechanics have already changed, and we do not yet possess the clarity to grasp this new paradigm. The reason we cannot foresee the future correctly is that we possess no tools to interpret or make sense of this change.

My views on what will happen and the shifting paradigm are the subject of the next article. Don’t forget to subscribe so you don’t miss out, and share so you don’t feel alone!

Robinhood was an app founded in 2015 that stood out with its feature of commission-free trading on American stock markets. Until 2021, it was a platform where young people who were afraid of stock market investment came in large numbers because of this commission-free trading convenience. You could buy and sell shares both without commission and very easily. And it’s still like that. But in 2021, a handful of young activists who gathered on the Reddit platform started an operation using this commission-free trading model as a tool. The share at the center of this operation was GameStop stock.

The GameStop Operation

GameStop was a retail company whose business model was weakening because of digitalization in the gaming sector and was losing money in its physical stores. Big hedge funds saw this situation as an opportunity and took heavy short selling positions, betting that the company’s shares would fall. In fact, the amount of shares sold short reached 140% of the total number of shares.

A “hedge fund” is a private fund that invests in various assets using investors’ money and is usually managed by professional managers. Their goal is to provide absolute returns regardless of market conditions. These funds can take both long and short positions: they can buy by betting that an asset’s value will increase (long), or they can make profit by borrowing and selling if they believe it will fall (short). With complex strategies, they can make money both when the market rises and falls.

Of course, this situation didn’t only catch the attention of Wall Street sharks. The WallStreetBets community on Reddit noticed this excessive short selling amount and that the share was below its real value. When Ryan Cohen invested in the company and names like Michael Burry drew attention to the share, small investors also started buying in an organized way on the Reddit platform. And in these purchases, the favorite app of small investors, Robinhood, was used.

With the purchases of this highly collective community and social media support, the share price jumped from $20 to $483 in January 2021. This situation caused the hedge funds that bet on the price fall to panic and have to buy shares to stop their losses, which made the price go even higher.

I won’t go into more technical details here, but the amazing rise also ended when the Robinhood app at the center of events announced that it couldn’t meet the increasing margin requirements and stopped GameStop share purchases. GameStop is still alive, and the hedge funds that gambled that the share would fall wrote huge losses. For those who are curious, one GameStop share is currently $22.51.

Of course, if Robinhood hadn’t put this buying ban, completely different things could have happened. For example, there could have been many bankrupt hedge funds. The CEO defends themselves on this issue and says they could do this, meaning they ran out of cash to meet the margin requirement and had no other choice. We have no choice but to believe this.

Real World Assets

I may have made the introduction a bit long, but we needed to remember this story again before explaining what Robinhood is trying to do now. This financial initiative that started in 2021 to break the dominance of brokerage firms is now appearing before us again in a completely different way.

While the idea at its foundation was for everyone to be able to buy and sell shares without commission, now it wants to make everyone an asset owner. For this, it uses blockchain instead of classical stock markets that are operationally expensive, slow, and don’t allow 24/7 trading. We can actually call this briefly the tokenization of real world assets.

To give an example: The shares of companies with huge growth potential like OpenAI or SpaceX are only accessible to accredited investors and venture capital (VC) funds because they are not publicly traded. Robinhood aims to open these “closed companies” to individual investors using tokenization technology. Robinhood CEO Vlad Tenev explains the work they are trying to do like this: “The traditional finance system is like a duck that crazily paddles its legs under water but looks calm on the surface; you think you bought the share, but in the background, there is complex settlement and collateral traffic that goes on for days. With tokenization, Robinhood is trying to eliminate this traffic and make the system as fast and direct as a light turning on instantly when you press a button.”

The tokenized real world assets market passed the $30 billion limit long ago as of the third quarter of 2025. Reports from institutions like Citi and Boston Consulting Group predict that the tokenized asset market could reach a size between $13.5 trillion and $16 trillion by 2030. Of course, this huge growth potential attracts the attention of not only Robinhood but many players. In the tokenization race, there are serious competitors like Coinbase (America’s largest crypto exchange, developing institutional tokenization solutions), Securitize (a platform specialized in tokenization of private company shares), tZERO (Overstock’s blockchain securities platform), Republic (a platform offering tokenization service in private market investments), and Figure (focused on tokenization of mortgages and other assets). But Robinhood’s user base of over 23 million and its experience in commission-free trading can provide a significant advantage in this competition.

It seems that the tokenization of real world assets and stablecoins may have finally brought Bitcoin technology to product-market fit. Tokenized assets represent a high-value engine, while stablecoins represent the fuel that makes this engine work. This pair can finally make blockchain achieve “market-fit” by transforming it from a complex but unused technology into a global financial highway where billions of dollars change hands in seconds. What do you think about this?

See you in the next article.

Sources:

• Investax. Q3 2025 Real World Asset Tokenization Market Report https://investax.io/blog/q3-2025-real-world-asset-tokenization-market-report

• YouTube. Robinhood CEO: Making Everyone An Owner

• Ivest. Tokenization Revolution: Robinhood Reshaping Private Market Access for Retail Investors https://www.ainvest.com/news/tokenization-revolution-robinhood-reshaping-private-market-access-retail-investors-2507

• CryptoRank. SEC Raises Concerns Over Tokenized Assets While Robinhood Shares Surge https://cryptorank.io/news/feed/07821-sec-raises-concerns-over-tokenized-assets-while-robinhood-shares-surge

Yes, artificial intelligence is close to taking our jobs. We need to find different jobs now. You have probably read this many times. This time, we are starting to see its effects in different areas. These effects, which almost nobody thought about, are ready to change our lives.

The Memory Crisis

There are very few companies in the world that produce RAM and SSD disk modules for computers. Three companies make 90% of all RAM and SSD memory production. Until recently, the memory market was quite predictable. Supply and demand were easy to follow and matched each other. New investments in this area were made with long-term plans. Until artificial intelligence changed everything. Artificial intelligence and data centers changed the memory market in a way that threw all plans in the trash.

All three big producers now give priority to making memory modules for artificial intelligence and servers. These three producers (Samsung, SK Hynix and Micron) think that making memory for AI servers is more profitable. At the same time, they cannot predict exactly where this huge growth will end, so they are not in a hurry to invest in new factories.

The final blow to the memory market came from Micron. In its official announcement on December 3, 2025, the company decided to exit the consumer market with the Crucial brand, which makes RAM and SSD disks for computers that we all use. According to the company’s statement, “AI-driven data center growth has led to a huge increase in memory and storage demand” and Micron made this decision “to increase supply and support for larger, strategic customers.” In other words, it will focus all its production on servers.

As a result of all these developments, the prices of memory modules that consumers buy increased by 171% in the last year. This is a truly incredible price increase. In fact, DRAM prices have increased so much that they have even left behind the increase in gold prices. Especially people who buy these parts and put them together in what we call custom-built computers have already started to feel this increase. Even though it is not felt as much in ready-made systems, meaning laptops and desktop computers, because of bulk orders, we will see these increases very clearly in 2027 and 2028. Big players like HP and Dell have already started to warn about this.

While AI companies are making big investments to train their own models, the hardware needs of cloud server companies running these models have reached incredible levels. The lack of capacity to meet these needs quickly is causing us to buy more expensive products. In other words, artificial intelligence will cause us to pay higher prices for electronic products like computers and phones that everyone must have now. Just like with high-end graphics cards. Who could have known?

Should We Sell Our Bitcoins?

We have read in different articles before that artificial intelligence could bring the end of Bitcoin. It was already being written that AI and quantum computers could solve hash operations that take time to solve faster, and this could bring the end of Bitcoin. Before we even reached this stage, a completely different development happened.

You have probably heard of Bitcoin mining. It was quite popular once. Actually, it is still popular. If you have made this investment on a scale that can afford expensive electricity and chip costs, you still have significant returns. Of course, I am excluding the miners working with illegal electricity that appeared in our country. They have almost no expenses anyway. We say mining, but what these people actually do is check the accuracy of transactions in the Bitcoin network and give approval. The Bitcoin network, which is a decentralized system, needs these miners/validators. They are the backbone of the system. If they don’t exist, we cannot buy and sell Bitcoin either.

By design of the Bitcoin network, these miners receive Bitcoin rewards for the work they do. And these rewards are decreasing day by day. This is a requirement of the design. Because of the decrease in Bitcoin prices and the decreasing rewards, companies that do mining as a business have started to leave this business. Guess what they do now? Yes, you guessed right. They started renting their infrastructure to artificial intelligence companies. In the last 18 months, at least eight major mining companies in the US, such as Bitfarms, Core Scientific, Riot, IREN, TeraWulf, CleanSpark, Bit Digital, MARA Holdings and Cipher Mining, have announced plans to transition partially or completely to AI. Both sides are happy. Miners earn more money, and AI companies can quickly use the installed computing power.

Analysts say that this transition in industrial-scale miners could cause different damages to the strength of the Bitcoin network in the long term. Both the risk of a 51% attack increases in technical terms, and there is a possibility of concentration in countries that can do this mining business even at a loss. For example, China, El Salvador. In this case, the Bitcoins in our wallets could suddenly disappear.

A 51% attack occurs when more than 51% of the total computing power or validation authority on the Bitcoin network comes under the control of a group or organization. This allows the attacker to make important changes to the network. For example, they can change the order of transactions, prevent some transactions from being approved, or perform double spending. In other words, when the control of the majority is gathered in malicious people, the Bitcoin network is completely in danger.

The Butterfly

Every step we take for development results in us living and consuming everything faster. We are experiencing incredible events like a butterfly flapping its wings in the Amazon rainforest causing a storm in the USA. And unfortunately, the impact of artificial intelligence is much bigger than a tiny butterfly flapping its wings.

See you in the next article...

Sources

1. Micron Technology. “Micron Announces Exit from Crucial Consumer Business”. https://investors.micron.com/news-releases/news-release-details/micron-announces-exit-crucial-consumer-business

2. Klotz, A. “DRAM prices surge 171% year-over-year”. Tom’s Hardware. https://www.tomshardware.com/pc-components/dram/dram-prices-surge-171-percent-year-over-year-ai-demand-drives-a-higher-yoy-price-increase-than-gold

3. Del Rey, J. “America’s Biggest Bitcoin Miners Are Pivoting to AI”. Wired. https://www.wired.com/story/bitcoin-miners-pivot-ai-data-centers/

Brides-to-be planning their weddings, birthday groups in prep mode, white-collar professionals renovating their homes, incurable manifesters building their 2026 vision boards, millennials wondering “what should I wear and post on Instagram?”, newly married women determined to prepare such an impressive table that even their sister-in-law can’t stop praising it. Pinterest is their common ground.

Anachronistic yet trendy. Friendly yet commanding. Close to us, but slightly smug. Pinterest is no longer just our platform; it’s firmly on brands’ radar. Because its user base is growing every day and its advertising models are becoming more sophisticated. For sectors with strong visual storytelling, Pinterest is no longer “nice to have”; it’s a genuine opportunity.

In this article, we take a comprehensive look at Pinterest’s rise in Turkey, 2026 trend expectations, which sectors benefit the most, how Pinterest drives sales, and which campaign types perform best. We examine what happens if you do this versus that.

(And yes, there’s a small surprise for agencies at the end.)

Why Is Pinterest Usage Increasing?

Pinterest’s global growth is already well known, but its rise in Turkey is particularly noteworthy. Estimates suggest between 7.5 and 10 million active users, with the majority aged 18–45 and a strong female skew.

Below, you can see the steady growth in numbers from February to November.

This audience isn’t on Pinterest just to browse. They’re there to decide.

They’re looking for decoration ideas, figuring out how to build a capsule wardrobe, saving hairstyle inspirations, collecting shopping ideas, planning holidays and crucially, organizing all of this into boards they can revisit and compare later. Because the economy no longer allows trial and error, Instagram posts feel too permanent for mistakes, and life is simply too short to live in an ugly room.

Pinterest’s biggest difference from Instagram and TikTok is this: users are at the very beginning of the decision-making journey. That means brands can reach potential customers earlier than almost any other social platform. This is where the cold audience lives.

And importantly, users come to Pinterest intentionally to research products. We’re not inserting products into entertainment, we’re showing up exactly where people are already looking for inspiration and solutions. The question is: how?

Pinterest now sits just behind the “big four” platforms in terms of brand attention. Yes, it still has a long road ahead but it’s an exciting one.

How Do You “Be” Pinterest?

Pinterest doesn’t work on the “share the moment” logic of other social networks. Content here functions more like a visual storefront or a curated moodboard — and the algorithm reinforces this behavior.

Highly stylized imagery, coherent interior aesthetics, harmonious color palettes, clean compositions, well-staged product photography, and an overall sense of visual order perform significantly better on Pinterest.

And unfortunately (or fortunately), even TV series are now part of this ecosystem.

Quick, casually shot videos that might perform on TikTok don’t translate well here. Pinterest rewards content that is intentional, designed, and close to editorial quality. Even in high-traffic categories, every piece of content demands serious post-production and visual planning. Brands that can deliver this are rewarded with strong, sustained traffic.

So which areas perform best?

According to Pinterest’s own data, traffic is expected to grow in the following categories in 2026:

• Minimal living spaces built with natural materials

• Sustainable fashion and personalized style

• Home production content: crafts, DIY, small renovations

• Short-form educational videos

• Shopping pins directly connected to purchase

And yes, even categories you wouldn’t expect can thrive. (No one would guess grocery retail — yet Whole Foods receives over 10 million monthly views.)

However, succeeding in these categories isn’t as simple as “posting what’s trending.” All of this requires high visual quality. Color consistency, composition clarity, narrative strength of the object or product, typography and layout (especially in video or Idea Pins), and even the shooting language directly affect performance.

In short, what needs to be built is this:

A consistent, visually coherent source of inspiration.

That’s why success on Pinterest is both a creative production process and a strategic design discipline.

Infiltrating Special Occasions and Hobbies

Pinterest’s real power lies in inspiring people around their special moments and personal interests. But how do we do that? How do brands generate organic traffic or launch campaigns on Pinterest?

In simple terms: how should brands use Pinterest?

Because Pinterest functions like a search engine, well-structured content remains visible for a long time. Brands should reimagine their product catalogs through a board-based aesthetic. Instead of posting single products in isolation, creating inspirational combinations, theme-based moodboards, color stories, and usage scenarios significantly improves organic performance.

Optimizing titles, descriptions, and alt-text for SEO; maintaining consistent color palettes and clean composition; and designing content that feels worth saving all give brands algorithmic advantages. Tying content to recognizable cultural codes or established aesthetic languages amplifies this effect even further. Pinterest Predicts reports already outline dozens of upcoming trends across style, fashion, home textiles, and travel for 2026 they’re well worth reviewing.

User-generated content, before/after visuals, mini how-to videos, and Idea Pins also play a strong role in increasing organic reach. Since Pinterest favors long-lived content, each visual asset gradually contributes to a brand’s own discovery ecosystem.

But let’s say you’ve done all of this and now you’re launching your hero product. If you’re operating in Turkey, this often means setting up campaigns through a UAE-based entity due to advertising limitations. What then?

First, an important note: brands with strong organic Pinterest performance tend to see significant returns from this investment especially in catalog ads and retargeting campaigns. Because Pinterest performance emerges from the intersection of content, context, and correct targeting.

1. Idea Pins and Video Pins

Short, educational, fluid content captures attention quickly.

Example: Three simple ways to make a small living room feel larger.

2. Product Catalog Integration

Shopping pins that show products with pricing and descriptions are extremely effective for e-commerce.

3. Theme-Based Boards

Boards like “Capsule wardrobe ideas”, “2026 home trends”, or “Earth-tone interiors” place products within a lifestyle context, accelerating conversion.

4. SEO-Optimized Pins

Pinterest works like a discovery engine. Keyword usage in titles, descriptions, and alt-text dramatically improves visibility. Globally, this works across categories ranging from textiles to organic honey.

5. Retargeting

Reaching users who added items to cart but didn’t purchase significantly increases conversion rates.

6. Seasonal Campaigns

Search demand peaks around periods like wedding season, Mother’s Day, and New Year. Campaigns aligned with these moments perform especially well. Because ultimately, on Pinterest, we’re selling a mood.

How Does Pinterest Contribute to Sales?

Sales on Pinterest usually begin at the planning stage.

When a user saves a product, they’re essentially saying, “I’m considering this.”

What happens next depends on the brand:

• Inspiration boards show where and how products fit into real life.

• Product pins drive direct traffic to the store.

• Retargeting accelerates the final purchase decision.

Pinterest traffic is consistently reported to have longer session durations and higher page views. Over time, this turns the platform into a low-cost, long-term traffic source.

Additionally, Pinterest provides valuable learning data about what audiences genuinely respond to — insights that can be leveraged across other channels.

Final Thoughts

Pinterest remains a significant opportunity for brands not just global giants, but creative startups as well. Because its search-driven structure allows well-designed content to stay visible for a long time. Brands that reframe their catalogs through inspiration, rather than isolated products — using combinations, moodboards, color stories, and usage scenarios win here.

In short: products need to be save-worthy.

Major brands like Anthropologie, H&M, and Mango invest in both paid and organic Pinterest strategies. Locally, it’s encouraging to see brands like Manuka joining the space as well. This really is a new, enjoyable playing field.

And one final note: agencies can now open Pinterest business accounts too.

Which means the game just got a bit more interesting

Breaking Points

Before we get to the reason, let’s briefly look at how Nvidia got here. You’ve probably read the historical information in many places. Nvidia started as a company that made graphics cards for computers and became very successful at it. But we can talk about two big turning points in the company’s history. The first was when people discovered that Nvidia graphics cards were very efficient for crypto mining. The second important development was artificial intelligence and training large language models, which also need a lot of computing power.

During the pandemic, global supply chain problems made it very hard to get Nvidia graphics cards. People who really wanted these cards for gaming couldn’t find them because crypto miners were buying them all. This point can be seen as the peak for Nvidia in crypto mining. I think this was a situation that developed without much planning. People who rushed to buy these cards to mine cryptocurrency found more efficient ways to do it at this peak point. When people face difficulties, they look for new alternatives. Because of this, ASIC circuits were created for crypto verification, and Nvidia’s dominance in this area moved to ASIC circuits.

An Application-Specific Integrated Circuit (ASIC) is an integrated circuit (IC) specially designed for a specific task or application. Unlike FPGA cards that can be programmed after production to meet different conditions, ASIC designs are adapted to special needs in the early stages of the design process.

While Nvidia was hiding from its investors that most of its GPU income came from crypto mining (they went to court over this and paid a large fine to the SEC), it was also producing chips for growing AI companies. When OpenAI announced ChatGPT in November 2022, all the computing power behind it was Nvidia chips.

With ChatGPT’s huge success, all AI companies started knocking on Nvidia’s door to train large language models. Orders that no one could predict made this graphics card company the world’s most valuable company in a very short time. Nvidia still prefers to hide its Achilles heel. Looking at its last quarter results announced last week, 61% of its total revenue comes from four big customers whose names are not revealed. These are probably Amazon, Microsoft, Meta, and Google (Alphabet). Maybe Oracle could be in place of one of these. This concentration naturally creates a systemic risk.

Warning Signal

Google, probably one of Nvidia’s biggest customers, announced its new large language model Gemini 3 and related tools on November 20. The success of Gemini 3, which the whole world could access on the first day, was really amazing. Unlike OpenAI’s relatively less successful new versions recently, Gemini 3 really showed serious progress compared to its previous version.

The real bomb is hidden in the details. Google used its own designed TPUs to train this model. TPU, which stands for Tensor Processing Unit, is a processor type that Google and Broadcom designed together. Actually, this can also be called an ASIC. With these circuits that are designed only for model training by Google, brought to production by Broadcom, and of course manufactured by TSMC, Google is becoming able to offer a “full-stack” solution from beginning to end (from chip design to model development and cloud integration).

Also, looking at the shared data, they completed this training process almost 5 times cheaper than OpenAI, which uses Nvidia GPUs. This is a very important step in terms of efficiency.

Google has been investing in this area for almost 10 years. With Tensor processing units, it seems to be getting the return on this 10-year effort. Actually, like Google, other big AI companies are also working to produce their own chips. This includes OpenAI too. Nobody wants to depend on Nvidia forever. And as we understand from this first warning signal, they will be successful on this path.

After all, there are very powerful chip designer companies besides Nvidia. Like Broadcom and Qualcomm. At the end of the day, they all go and have these chips manufactured by companies like TSMC anyway. So Nvidia is also a fabless manufacturer like the others, meaning it has no factory. We know AMD also has serious work to have a say in this field. The quiet giant Intel, even though it might seem like an unimportant competitor, is still a manufacturer that should not be ignored with its production capacity.

Nvidia’s Dark Future

For Nvidia, it’s time to make a new plan. When we look at the distribution of Nvidia’s revenue items in the graph above, we see a company completely leaning on data center revenues. Also, the fact that four big customers whose names are not revealed make up 61% of its revenues is a very big risk. Even one or several of its customers switching to their own chips will cause a huge drop in revenues.

What’s Waiting for Us

Unfortunately, it’s not possible to see the future from today. But as Ilya Sutskever, one of OpenAI’s founders, also mentioned, we have reached the end of language models trained with more processing power and bigger datasets. The part we will focus on from now on is efficiency. That is, models that can be trained cheaper and process cheaper. Google seriously shook Nvidia’s throne with its new Tensor processing units. I expect other companies to announce their models trained with their own chips one after another. All of this could start an AI era where we can access it cheaper with competitive advantage.

(Since Nvidia, the subject of this article, is a publicly traded company, I want to state that what I write here is not investment advice.)

Which company do you think will be next to produce its own chip?

See you in the next article.

Over the past two years, vibe marketing has been everywhere. Building communities, creating micro-connections, tailoring strategies for every single user… It all sounds amazing. But there’s a hidden cost no one mentions: brand consistency.

Every brand has been swept into this personalization vibe. But if everything moves this fast, where does the brand’s structure stand?

Does Personalization Have a Limit?

According to McKinsey, 71 percent of consumers expect personalized experiences, and 76 percent feel disappointed when they don’t get them.

Users want more personalization—always more. But no one talks about the other side: as personalization increases, the brand’s direction starts to blur. Until now, brands maintained consistency through large teams—creative directors, brand managers, strategists. Their job was to keep the line straight.

Now AI says, “I can create content for each individual.” But isn’t that level of variation a risk for the brand?

A thousand versions of the same brand? Or a brand morphing into something unrecognizable?

Can you show users messages tailored to them… but can you still find the brand among all those variations?

Does Personalization Increase Indecision?

When every touchpoint surfaces a different message, the user falls into an inconsistent flow. Extreme personalization can even turn into a trust problem. More options also create a “Which one should I trust?” paradox. Content that should help the user becomes a tiring cloud of choices. And when that happens, the brand’s conversion curve starts dropping.

And here’s the part no one discusses enough: producing at a 1:1 level is easy; managing the outcome is the real challenge.

AI can generate unique content for every user. But deciding which version stays on-brand, which one is risky, and which one aligns with what the company actually wants to say is still a human job.

AI speeding things up is great; no argument there. But organizing that speed is still a human responsibility. Yes, hyper-personalized campaigns can help the customer, but without control, they also increase indecision. And even though AI makes production easier for teams, the burden of controlling those variations grows. Letting AI police itself is an option, but that loop becomes a trap quickly.

Personalization is fascinating—but it doesn’t always create a better experience.

The Magic Word of Vibe Marketing: Connection

“Build a community, create a vibe, grow organically.” Beautiful words. Reality doesn’t work that way. True organic growth needs something beyond emotion; it needs a system.

Vibe marketing and personalization often get stuck in the idea of content variety. The goal becomes feeding micro-communities by producing different content for everyone. But variety can disrupt how communities relate to the brand.

Think about it: if AI shows everyone a different tone and message, the brand’s core message can drift and fade. Does that create a shared feeling? I don’t think so. Loyalty doesn’t come from content variety—it comes from a shared emotional anchor.

Brands should be asking:

1. Who is responsible for conflicting campaign messages?

2. How will AI-generated content be governed?

3. If every message is personal, what shared experience will people remember the brand by?

4. What is the organizational cost of this system?

Vibe marketing is a hype right now. But tomorrow, this flood of hyper-personalized messages might grow personalization while weakening the brand itself. Don’t get swept up in the vibe.

Uncontrolled Power Isn’t Power

Is 1:1 marketing possible? Yes. But you can’t take AI output and declare, “We’re doing vibe marketing now.”

A sustainable model isn’t offering 1,000 different campaigns to 1,000 people—it’s offering a limited range that feels personal without breaking the brand apart.

Using AI output without guardrails will tear a brand to pieces. So what needs to be done?

1. Redesign the brand’s core structure around clear personalization levels.

2. Pass AI’s speed through a brand filter.

3. Focus personalization on meaningful segments, not the extremes.

Marketing has passed the era of “produce more.”
This new stage rewards brands that know what not to produce.

The most common advice in management strategy is “don’t put all your eggs in one basket.” I’m sure you’ve heard this at least dozens of times. It’s simple but very valid advice. Of course, being simple doesn’t mean it’s easy to apply.

In my article where I asked “Can a Billion-Dollar Company Be Built Alone?”, I mentioned that a portfolio approach with this meaning would be the most important method on this path. Actually, there are investor companies that work this way. For example, Bending Spoons, whose AOL acquisition surprised us recently. Actually, the part that surprised me was: was AOL still alive? AOL (American Online), one of the first and most important internet service providers in the industry, was still alive and worth $1.5 billion. Most of us know AOL, but who is this Bending Spoons?

Bending Spoons was founded in Milan, Italy, in 2013 by Luca Ferrari and his partners. Describing itself as “private equity hipsters,” Bending Spoons specializes in acquiring and revitalizing software and content companies. The company focused on this area with the experience it gained from an initially failed startup (Evertale) and added businesses with potential but not shining enough to its portfolio.

With these acquisitions, Bending Spoons became one of Europe’s most valuable technology startups in just a few years. Its annual revenue is over $1.2 billion and it has dozens of companies in its portfolio.

Actually, in our country, Mediazone, the parent company of Onedio.com, does what Bending Spoons does by acquiring content-producing companies. Similar logic, similar work. Mediazone is also quite successful, especially in our country. This should also be noted.

Acquisition Portfolio

Bending Spoons is known for its aggressive acquisition strategy. Since its foundation, the company has made acquisitions in a wide range, from small applications (the first acquisition was $10,000) to billion-dollar big deals. In recent years, we can see that it targets established but stagnant brands in the face of new competition.

Some of Bending Spoons’ notable acquisitions include:

• AOL: Announced to be acquired from Yahoo for $1.5 billion in October 2025.

• Vimeo: Acquired in September 2025 with a deal worth $1.38 billion.

• Evernote: (2022/2023) Well-known note-taking platform.

• WeTransfer: (2024) Dutch file-sharing service.

• Meetup: (2024) Community platform for in-person and virtual events.

• Brightcove: (early 2025) Enterprise video platform, taken private from a public company.

• Remini: (2021) AI-powered photo editing application.

• StreamYard: (2024) Live streaming application.

• Mosaic Group Assets: (2024) Assets of a New York-based mobile app developer (only intellectual property was targeted by laying off all employees).

Acquisition Reasons and Strategic Goals

Bending Spoons’ basic acquisition thesis is to focus on companies with consumer-facing products that have high value potential but have historically failed to achieve this or are under debt. Actually, looking at the companies they acquired, I outlined their strategies as follows:

Loyal User Base and Revenue Stream: Acquiring companies with a loyal user base that trusts the service and is unlikely to leave despite price increases. For example, AOL is an iconic business with approximately 8 million daily and 30 million monthly active users. Additionally, a common point is that these companies usually have a steady cash flow (mostly subscription-based software).

Operational Inefficiency Opportunity: The acquired companies are generally those where old managers made big mistakes in management. Bending Spoons can significantly increase profitability by correcting these mistakes and reducing costs.

Data Assets: The acquisition of old platforms like AOL must have been made strategically not only for user numbers but also for access to the data produced (training sets). Because these old platforms host large, proprietary historical datasets that will be used to feed Bending Spoons’ future AI development.

Long-Term Ownership: Unlike traditional investors, Bending Spoons approaches companies with the philosophy of “owning and operating forever.” In other words, the opposite of the buy-grow-sell logic. This long-term approach allows it to make long-term investments in the companies it acquires.

Operational Strategy and Value Creation Mechanism

Bending Spoons uses its operational and technological infrastructure as leverage in the companies it acquires. CEO Luca Ferrari also summarizes this approach as “we are 25 percent investment, 75 percent technology company.” This part about using operational and technology infrastructure as leverage is really important. For example, they use the workforce in their companies in different geographies as a kind of cost arbitrage. For instance, American employees in high-cost centers (like Silicon Valley) are laid off and operations are moved to the central team in Milan. The layoff of 75% of the personnel immediately after WeTransfer’s acquisition is another painful example of this. Ferrari states that this “painful transition” is necessary to make the company successful in the long term. I don’t know if employees think the same way.

It should be noted that using a central infrastructure platform technologically also has important contributions. This internal platform contains more than 50 pieces, from A/B testing infrastructure to payment management systems. When a new company is added to the portfolio, these central services are immediately applied, providing great advantages in terms of cost.

Culture and Talent Strategy

Bending Spoons’ operational capability is largely based on its extraordinary and high-performance corporate culture. To achieve this high performance, they focus on raw talent rather than experience, usually hiring fresh university graduates or young engineers.

They have a company culture coded with strict principles like “Uncompromising Excellence” and “Relentless Simplification.” Despite this, Bending Spoons claims that the employee turnover rate is at a low level of about 1% per year, which is much lower than the industry average. They are also very good at quickly shifting employees to opportunities that will create the highest impact to minimize organizational friction. This creates a cost advantage for the company while also accelerating employees’ continuous learning and development.

The Importance of Economies of Scale

Actually, if you’ve noticed, the most important concept that Bending Spoons relies on in these acquisitions is economies of scale. If you grow and diversify your business enough, you can provide cost and operational advantages in many items. In today’s world where artificial intelligence and agentic coding concepts are on the rise, I believe we need to rethink this concept of economies of scale.