Twitter/X, my friends in tech, my colleagues at CBA. We all talk about AI. Claude Code here, agents there. It seems omnipresent. But these past weeks, I’ve been talking to devs, companies, CTOs outside my bubble. And I noticed something. Most haven’t switched.
Not because they’re incompetent. Not because they don’t know AI exists. But because there’s real friction. Lack of time first, teams are already overloaded, impossible to take 2 weeks to “test AI.” Uncertainty next, “we use ChatGPT for snippets, and then what?”, not clear how to transform entire workflows. And then the fear of investing time in a tool that’ll be obsolete in 6 months. Which tool to choose? Cursor? Claude Code? GitHub Copilot? All of them? None?
The numbers confirm what I observe. According to the Anthropic Economic Index from September 2025, only 9.7% of American companies use AI. In Europe, it’s even lower. The WEF reports in 2025 that over 60% of European companies are at the earliest stage of maturity. McKinsey shows in January 2025 that 88% of organizations say they use AI, but only 1% consider themselves “mature.” The gap between “we’re testing ChatGPT” and “we’ve actually transformed our workflows” is massive.
What these past 3 months taught me
I wrote One year after Hello World to tell how I got here. Here’s what I take away from these past 3 months.
Before, learning a new stack took 2-3 weeks. Reading docs, testing, breaking, starting over. Lots of dead time. Now it takes 2-3 days. Claude guides me, explains, corrects in real time. No dead time. And the paradox is that before, either I was learning, or I was producing. Now I learn while producing. Claude explains while I code. I understand by building.
And honestly, it’s become the most satisfying thing I’ve done in a long time. The feedback loops are ultra-fast. I test an idea, Claude corrects me immediately. Iteration in seconds, not hours. Every day I learn something new. Not just “how to code X,” but “why architect it this way,” “how to think about this problem,” “what’s the right abstraction.” It’s addictive.
What it teaches me goes beyond code. Architecture. Systems thinking. Product thinking. Claude doesn’t just code for me, it explains why. I’ve been grinding these past months. BMAD. Claude Code terminal-first. Spec-driven, test-driven. Olympus and Pantheon failed, but I learned. MVP BVN. scanr. Personal site. CBA hackathon this week, 1 week to recode a big chunk of the app.
But this head start won’t last. LLMs are getting simpler to pick up. Anthropic, OpenAI, Google, all working to simplify UX. What took me 2 weeks to master will probably take 2 days in a year. The learning curve is shrinking. Adoption is accelerating. By end of 2027 in my estimation, tools will be trivial. Everyone will have access to the same power. The advantage will go back to raw talent, not tool mastery.
What I’m doing with it
I have, by my estimate, 18-24 months. Maybe less. To make use of this head start before it dilutes. Not just “coding faster,” anyone will do that in 2 years.
What seems interesting to me is sharing what I’ve learned. Helping other teams through this transition. Auditing current workflows, training on BMAD methods and Claude Code, supporting pilot projects with knowledge transfer. The idea isn’t to sell AI or pitch a “magic SaaS,” but to transform how people work by structuring human-AI collaboration. And it’s not just about devs, marketing, ops, support, any role with structurable workflows.
SMEs and tech startups, development teams, companies that feel the gap but don’t know where to start. Why now? Because in 18-24 months in my estimation, this expertise will be commoditized and tools will be trivial. Today I have a head start I can share. Tomorrow, probably, everyone will be at the same level.
I’m on this train, learning, building, iterating every day. I’m going to use these months to grind XP, share what I learn, and help those who want to get on board.