Two ways to leverage AI
There are two ways to leverage AI, and they're two different games, not two steps of the same journey. Mix them up without a mental model and it gets confusing fast.

On software architecture, AI in engineering, and building things that last.
There are two ways to leverage AI, and they're two different games, not two steps of the same journey. Mix them up without a mental model and it gets confusing fast.
Most advice about AI is about speed and leverage. Almost nobody talks about the prerequisite: knowing what you're building and why. Without that, AI is a drift accelerator.
I used AI to harden a platform for public-sector compliance anyway — and the whole trick was making sure it never made a single security decision.
With AI, move too fast in the wrong direction and you don't just ship a buggy feature — you lose user trust, or land in legal trouble. The teams that skip this sequence are the ones you read about in the headlines.
The best tips for building a personal AI assistant don't come from the docs. They come from the room.
Unparalleled programming speed leaves you with volumes of code you can barely peek into. Luckily, there's a very powerful tool to counteract all these shortcomings.
A single confused backoffice user broke everything. Years later, building with AI in the same stack, I caught the same pattern. The world does not change. It just moves faster.
AI removes the language barrier when you're exploring. You can spin up a working proof-of-concept in a stack you've never touched. That's genuinely new. Production is a different conversation.
I just shipped a learning product. No backend. No frontend. No install. Zero lines of code. The whole thing is a set of markdown files.
The internet is obsessed with learning AI. Almost nobody is talking about leveraging AI to learn. Not the same conversation.
The AI had no memory of what I'd practiced the session before. Every time: cold start. So I started building what I was missing.
Docker networking, a quick-fire language drill, and the unexpected value of side-learning — session 2 of an AI-coached reactivation.
Don't ask AI to explain things. Ask it to challenge you. A 20-minute Docker reactivation that actually stuck — and why.
AI tools have lowered the bar for non-technical founders to validate ideas. But the moment real users get involved, the nature of the project shifts — from experiment to operation.