I believed the hype: this thing would give us economic freedom.
Learn the tools and the rest would follow. I did learn them. I can build today, alone, what used to need a small team. The honest accounting is less comfortable: building became cheap, and I became isolated, and I pretended those two facts were unrelated.
For me it was Claude Code in December 2025, then OpenClaw at the end of January 2026, right after I made my predictions, then Codex, and then Hermes. And now Jev.
My weeks became backlog tickets of 24/7 tasks and babysitting harnesses. I learned to meta-analyze myself while interacting with these intelligent systems. But somewhere I got stuck, and I can see it in one number: 400 skills across Codex and Claude Code. You don’t have 400 friends. You have 20 or 30 real ones; the other 370 are mostly acquaintances, Russian dolls of familiarity. I was collecting acquaintances and calling it capability. Worse, I collected deep-research reports because clarity felt like reducing risk before confronting the real world when I was looking for answers. It was the illusion of safety, and pretending to be busy was a comfort and a lie.
The obvious explanation is that the tools are the problem. I have seen it up close: friends having their first baby, realising they don’t care about any of this, because to them it is big tech serving its revenue and shareholders with their IPO agenda, however the PR hypes the mission. They aren’t wrong about the incentives: a Stanford review found all 6 frontier AI companies train on user chats by default, some indefinitely. Also, more people hate AI, and I was misunderstood for defending it. But blaming the tool doesn’t survive one question: why did the same tools amplify judgment and experience for some, and isolate others, like me? A suspect is not a culprit.
The explanation cost me 4 years. The tool is a mirror. It amplifies everything, including the foundation you bring. If the foundation is corrupted, corruption amplifies faster, especially on principles and integrity. The sycophancy grows more complex; Anthropic’s own researchers showed that preference models reward convincingly written agreement over correct answers. These systems cut the cost of production and raised the cost of verification, and verification is judgment, and judgment is the human part. In one randomized trial, experienced developers worked 19% slower with AI tools and still believed they were faster. The mirror doesn’t create the foundation. It scales it.
It played out beyond my office. Before AI it was the digital nomad era: everybody travelling, living the dream, selling the lifestyle. I wasn’t selling it; I was enjoying it more than performing it. That era ended anyway this year. The Minneapolis Fed counts remote work slipping: hybrid schedules grew while fully remote shrank, and Amazon and Meta called people back full time. The tools compressed exactly the digital work the nomads lived on. When everyone can produce, everyone produces the mean: last year everything was purple, this year everything is codified. Measured, too: generative AI raises individual creativity and lowers the collective diversity of what gets made.
My own loop has 6 steps:
1-Consume: the TikTok loop.
2-Think: process what you consumed on paper.
3-Build: productize.
4-Distribute: the hardest step, marketing drowned in its own sludge; the goal is not likes, it is a person with a problem reaching a solution.
5-Monetize: the moment of truth, where the real world hits you in the face, and most people fail.
6-Then reinvest, so the feedback loop compounds, slow and invisible until it is exponential. That is how the rich get richer while the smartest guy in the room stays smart and broke.
I was stuck at build. I still am. Building felt like progress because it is an aura moment: look at the dashboard. But building for building is an ego thing. When you build, you must always have a goal: solve a pain point, open source it, or monetize it. An outcome, not an aura.
The reframe that freed me came from friction. Not all friction is useful. The reports and the 400 skills removed the wrong friction: contact with reality. The useful friction is the other kind: real people, real clients, angry or happy. A friend pulled me away from the computer into nature and toward people; that did more for my judgment than any model release. We are in a big identity shift: we are defined by what we do, not by what we pretend or wish to be. My mentor, may he rest in peace, said it in French:
“Il y a une différence entre ce que l’on croit être, ce que l’on veut être, et ce que l’on est.” In English: “There is a difference between who we think we are, who we want to be, and who we truly are.”
Believing I was a builder who distributes is not being one.
What changes now: my stealth season is over. The next friction is distribution: real people, real clients, real money or real rejection, with this time a slightest precision, a compounding feedback loop. Even my second paper draft, about preserving our human intent when AI drifts, tied to the nonprofit I’ve been trying to build over the last few months, only counts once it touches real people. Otherwise, it’s just sophisticated AI slop, with AI talking to AI about humans. If you recognize yourself in that mess, the diagnostic is simple: name the step you are stuck on, and trade one collection habit for one real contact.
4 years, and the mirror has not changed; what I bring to it has started to. The compression is this: production is cheap, judgment is the cost, and the loop only closes when you reinvest in reality. They want you to believe you are a regular guy repeating his day and escaping on weekends. But we all have an alpha inside us. All of us. We undersell it, or have not found it yet. The sun will rise with or without you. I intend to be moving when it does.
Sources
1. Federal Reserve Bank of Minneapolis, “Is remote work declining? What the latest data show” (June 2026). Supports: remote-work shift - about 22% of US workers worked from home at least some of the time in 2025, hybrid share grew while fully remote shrank; Amazon and Meta return-to-office.
2. Stanford HAI, “Be Careful What You Tell Your AI Chatbot” (October 2025). Supports: all six frontier developers (Amazon, Anthropic, Google, Meta, Microsoft, OpenAI) use chat data for training by default; some retain it indefinitely.
3. METR, “Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity” (July 2025). Supports: randomized trial, experienced developers 19% slower with AI tools; they expected 24% faster and still believed 20% faster after.
4. Anthropic, “Towards Understanding Sycophancy in Language Models” (ICLR 2024). Supports: sycophancy is a general behavior of RLHF models; humans and preference models prefer convincingly written sycophantic responses over correct ones.
5. Doshi and Hauser, Science Advances, “Generative AI enhances individual creativity but reduces the collective diversity of novel content” (2024). Supports: the sameness-of-output claim.
6. OpenClaw, open-source AI assistant. Tool named in the personal timeline. openclaw.ai
7. Hermes Agent, Nous Research, open-source AI agent. Tool named in the personal timeline. hermes-agent.nousresearch.com
8. Jev, an AI model that classifies text and checks it against questions you define. typesafe.ai/blog/introducing-system-one-models-and-jev
9. The golden age of digital nomadism isn’t over, it has evolved. linkedin.com/posts/ashkanborouj_digitalnomad-hybridwork-futureofwork-share-7481294317051916288-nWzc/