AI Adoption for Dev Teams

If you have a dev team building your product and you're mulling over the idea of them starting to use AI for coding, you'll likely go through three stages. Here's what I've seen in practice.

Usually, the founder has had a few conversations with people from the dev team and has heard some stories. He or she assumes that software developers are smart people, so they must have figured out this entire AI thing on their own.

In my experience, the reality is different. There are two or three people on the team who have a $20 Cursor or Claude Code subscription. And what's interesting, their opinions about AI for coding are usually pretty mixed.

Nothing consistent is happening at this stage. There is no shared tool, no shared training, no standard workflow. That is why results don't match the marketing hype and all those stories on the internet about someone building an amazing app in a single evening.

This is where you start taking serious action. You pick a tool, Claude Code, Cursor, Codex, or whatever, and you run a short training. In my experience, it usually takes about two to three weeks for software developers to really learn how to prompt AI, how to review output, and other things.

You start to see the first results pretty quickly. More code being shipped faster. Shorter product iteration cycles. More issues being closed compared to what you had before.

But it's still mostly personal productivity. Everyone does it differently. Best practices stay inside your developers' heads. And in my understanding, many companies stay at this level for some time, because founders or product owners can't generate ideas and feature requests fast enough to keep up with the newly accelerated dev team.

You have to be really observant and attentive at this stage. You have to watch what your software developers are constantly asking AI to do. You should see certain repetitive patterns.

For example, someone keeps asking AI to generate documentation in a certain way. Or notify people about something in a certain format. When you notice these patterns, ask your developers one question: "How can we trigger this automatically?"

This is where AI automation starts. The most classic examples people talk about online are automated release notes generation and automated code review. But there are many, many other things.

The Bottom Line

These are the three stages I've seen in my experience. I believe there are more, fourth and fifth stages that teams will discover as tooling matures. But for most companies today, the jump from Stage 1 to Stage 2 is where the real unlock happens. And the jump from Stage 2 to Stage 3 is where it gets genuinely interesting.