AI as a means, not an end

Most of the conversation about AI and software treats the model as the destination. Ship the model, mention the model, be an AI company. I think that’s backwards. AI is a means, a very good one, to the same ends we’ve always cared about: useful products, durable systems, and teams that can move fast without breaking the things that matter.

The transformation is real, just misplaced

The industry is being reshaped, but not where the headlines point. The interesting shift isn’t “AI writes the code now.” It’s that the cost of the first draft (of code, of designs, of prototypes) is collapsing toward zero. When drafts are cheap, the scarce skills move elsewhere: taste, judgment, system design, and knowing which problem is actually worth solving.

In other words, AI doesn’t remove the hard parts of engineering. It relocates them.

What gets more valuable, not less

• Judgment. When anything can be generated, deciding what should exist is the bottleneck.
• System thinking. Models are great at local correctness and indifferent to global coherence. Someone still has to own the whole.
• Verification. A cheap draft you can’t trust is a liability. The teams that win invest in tests, observability, and tight feedback loops: the boring fundamentals, now with leverage.
• Communication. Translating fuzzy intent into something a model, or a team, can execute is the new core skill.

The means-not-end test

Before adding AI to a product or a workflow, I ask one question: does this make the end better, or does it just make the deck better? If the honest answer is the latter, it’s theater. The best uses of AI I’ve seen are almost invisible: they remove toil, shorten loops, and let small teams behave like large ones. Nobody on the outside calls them “AI features.”

That’s the transformation worth betting on. Not AI as the product, but AI as the thing that quietly lets you build a better product, faster, with fewer people and more confidence.

The end hasn’t changed. The means just got a lot more powerful.