Discovery Is the Real Advantage

When shipping is cheap, the only advantage left is knowing what to build - and why.

Delivery speed was always the wrong metric. The best product teams have known this for years, but arguing against it when "how fast can we ship?" was the only scoreboard didn't get you very far. Speed was the variable that could be optimized. Discovery looked like overhead.

AI has clarified the argument in a way that's hard to ignore. When delivery cost drops to near-zero, shipping speed stops being a competitive advantage. You can't win on a variable everyone has access to. What's left is the question underneath the speed: what are you building, and why?

Marty Cagan said it directly in April: the bottleneck has moved. It was delivery. Now it's discovery. "Build to earn" - shipping features fast - is table stakes. Any team with access to decent AI tooling can do it. "Build to learn" - running discovery, testing hypotheses, understanding what users actually need - is where product value now lives.

The teams that build better products aren't the ones who ship fastest. They're the ones who know what to ship.

This reframes a persistent tension in product organizations. Research and discovery have always been easy to cut. The counter-argument from research and UX teams was "you'll catch it in production" or "you're going to pay for this later." Both true, and neither especially motivating when speed was the only scoreboard.

The new counter-argument is harder to dismiss. When delivery is cheap, the cost of building the wrong thing is the entire cost. There's no inefficiency to hide behind. You can ship anything fast - but you can't ship the wrong thing fast and recover quickly, because the market doesn't slow down to wait. Every fast iteration on a bad insight is just accumulating speed in the wrong direction.

Discovery is now the only variable that separates teams.

Here's another thing that hasn't gotten enough attention: what AI exposes about PM work. AI can generate roadmaps, prioritize backlogs, synthesize customer feedback, and produce feature lists. It can do most of what occupied PM time in an activity-driven model.

But discovery isn't a PM-only job. It's a partnership between product, and UX - and it's strongest when these disciplines do the work together.

What it can't do is run real discovery. It can't sit in a user session and notice the thing the user didn't say but meant. It can't hold the tension between what metrics are telling you and what qualitative signals are suggesting. It can't make the judgment call about which user segment to optimize for when the data is ambiguous. Those decisions require human judgment, applied with context, over time.

PM value has concentrated at the point of discovery - and the same is true for UX, research, and design. The work AI has made redundant - pixel-level execution, component generation, responsive layout, basic prototyping - was never where design value really lived anyway. The work AI can't do - understanding what users need, designing for trust and agency, making judgment calls when the right answer isn't in the data - is exactly where design value has always been highest.

For teams that have been investing in discovery without always being able to justify it quantitatively, the math is now obvious. Delivery is cheap. Discovery is the leverage point. Every hour spent understanding users is worth more relative to every hour spent shipping features than it was two years ago.

For teams running pure feature factories, turbocharged by AI and increasingly uncomfortable about it, the exit is through discovery - not because the old model is broken, but because the value equation has shifted far enough that it stops generating returns.

The question "what should we build?" has always mattered more than "how fast can we build it?" AI has made that answer financially obvious.

What gets built when teams invest in discovery is what people actually want to use: features that solve real problems, and iterations that build on insight instead of accumulated assumption. Users notice, come back, and recommend it. Those outcomes are designed, and they start with genuine discovery, which AI can accelerate but not replace.