I’ve (unsurprisingly) seen and been part of a lot of discussions over the last few months about how AI will impact the world.

When disruptive technologies are first introduced there’s an awkward early phase where everyone agrees something important is happening, but nobody is entirely sure what the long-term consequences will be.

We’re very much in that phase with AI right now.

The conversations I’ve been part of have ranged from AI being an existential risk through to it being an enabler of technological utopia. Most sit somewhere in the middle.

What strikes me is how quickly the impact is already being felt in the design and development of digital products and how unevenly distributed that impact is currently.

I’ve spoken with people at companies who have gone all-in on AI. Others are still experimenting cautiously. Some teams are deeply enthusiastic. Others remain unconvinced.

The experience of working with AI in product today seems to vary wildly depending on where you happen to work.

The one thing that does seem broadly agreed upon is that building and shipping software is becoming easier - even if at some cost to code quality.

Whether that’s through AI-assisted coding, design tools, research tools or something else entirely, the cost - especially in terms of time taken to ship - of turning an idea into a working product is falling.

Most of the discussion I’ve seen focuses on that point.

Teams can build more.

Founders can build without needing as much specialist support.

Experiments that previously wouldn’t have justified the investment can now be explored.

All of which is interesting.

What I’ve found myself wondering, though, is whether the hype and uncertainty that AI presents is leading people to pay too much attention to how quickly we can now build and not enough to what we’re actually learning.

For years, product teams have talked about the importance of learning quickly. Anyone who’s worked with me will be bored of hearing me talk about how we might “optimise for speed of learning”.

Shipping fast and often is a core component of learning quickly, but the goal was never to ship features for the sake of it. It was to better understand customers, markets and opportunities, so that future decisions could be made with greater confidence.

If AI lowers the cost of building, then our ability to generate ideas and ship experiments increases.

But, that doesn’t automatically mean our ability to learn increases too.

In fact, I can see a world where the opposite becomes true.

A company that ships fifty experiments and learns very little from them is unlikely to outperform a company that ships ten and meaningfully changes its understanding of customers after each one.

I’m not convinced that the winners of the AI powered future will be the companies that build the fastest.

I suspect they’ll be the companies that use AI to learn, adapt and iterate faster.

Building has always been a means to an end.

Perhaps that is the disruption worth paying attention to: not that AI makes it easier to build, but that it makes it easier to mistake building for learning.