Late to Claude Code. Or just in time.

I came to Claude Code later than I probably should have.

That was my first instinct. Then I thought about it differently.

Twelve months with Cursor meant I'd already done the reacquaintance work -- GitHub, Vercel, deployments, pull requests. The practical foundation you need when you've spent several years leading product and engineering teams rather than writing code yourself. I hadn't lost the instinct. I'd just been operating at a different layer.

That foundation transferred directly to Claude Code. The tooling changed. The underlying fluency didn't.

But the bigger realisation wasn't about tooling at all.

The thing that actually matters hasn't changed

Agentic coding tools are genuinely impressive. The speed at which you can move from idea to working software has shifted in ways that felt implausible two years ago.

And yet.

The constraint was never the coding.

The constraint was always clarity of direction. What are you building? For whom? What does success look like? What are you explicitly not doing?

Those questions didn't become less important when the tools got faster. If anything, they became more important. You can now move a long way in the wrong direction very quickly. The cost of starting without a clear outcome in mind has gone up, not down.

A well-structured PRD -- not a lengthy document for its own sake, but a clear articulation of the problem, the user, the constraints, and the outcome -- is still the thing that determines whether an agentic session produces something useful or something that looks finished but solves the wrong problem.

What product thinking actually contributes here

There's a version of this conversation that positions product thinking as a soft skill sitting alongside the technical work. That's not how I experience it.

Product thinking is the north star. It sets the direction before any code runs. It's the thing that decides which of the many possible outputs is the right one.

Without it, agentic tools are fast but directionless. You get output. You don't necessarily get progress.

This is not new. Jumping straight into code without proper discovery and definition rarely worked before AI made it faster. The developers I worked with who consistently shipped well weren't the ones who wrote code the quickest. They were the ones who understood what they were building and why before they started.

Agentic coding hasn't changed that dynamic. It's amplified it.

The timing question

Most people haven't seriously started with these tools yet. Adoption curves for genuinely new capabilities tend to look obvious in retrospect and feel uncertain in real time.

Feeling late is a natural response to watching others move fast. It's also usually wrong.

What matters more than when you arrive is what you bring with you. Prior experience with agentic workflows -- even imperfect, iterative, learn-as-you-go experience -- compounds. The reps transfer.

If you have product instincts and you're prepared to reacquaint yourself with the technical layer, you're not late. You're probably better positioned than most.

The tools keep changing. The need for direction doesn't.