Notes on web, AI, cloud, and enterprise engineering — written by the people building it.
By 2023, we had delivered 150+ projects. I thought we had it figured out. Then a friend asked me one simple question — “Have you heard of ChatGPT?” — and nothing was ever the same again.
Every junior dev today has access to the most powerful coding assistant ever built. So why are they still stuck on the same problems, asking the same questions, and shipping the same bugs? The answer isn’t the tools. It’s what the tools can’t replace.
AI didn't kill jobs — it killed excuses. In 2026, the question isn't whether machines can do your work. It's whether you've built the skills that machines can't replace, and the judgment to use them. Here's the honest breakdown.
Every prompt you write, every task you delegate, every "AI-assisted" output you ship — you are not using a tool. You are training your own successor. The question is no longer whether AI will replace white-collar knowledge work. It's whether you'll notice before it's too late.
I didn't expect a language model to confidently fabricate a memory, defend it under pressure, and then apologize for the fabrication using the same fabricated memory as proof. But there it was, playing out on my screen at 1 a.m., and I couldn't look away
Retry storms, cost-per-query, prompt versioning, and guardrails — the unglamorous engineering that decides whether an AI feature survives its first year.
A demo is one input. A product is an unbounded set of them. Here's what actually closes that gap once real users show up.