Building is easy now. Knowing what to build is the hard part.
Why we started Dodger, what we mean by product intelligence, and the kind of work the next decade of product teams will actually do.
A founder told me last month that his team had shipped 41 features in the last quarter. None of them moved the needle on retention. He didn’t know which ones to roll back, because rolling them back felt like admitting that shipping them was a mistake. The build velocity was the only thing he had left to point at.
This is the new shape of the problem. Building software has gotten genuinely, structurally easier — agents that write code, infrastructure that scales itself, design systems that come pre-baked. The cost of producing a feature has collapsed. What hasn’t collapsed is the cost of producing the right feature.
The cheap thing and the expensive thing
There is a useful frame: in any market, value flows toward whatever stays scarce. When code was the bottleneck, value flowed to the people who could write it. Now that code is cheap, value flows to whoever can answer the question that comes before it.
That question is, roughly:
Of every possible thing we could build for these people, which one — if we built it well — would actually change their lives enough that they would pay us for it?
It’s a hard question. It requires real curiosity about a real audience. It requires patience to sit with ambiguity. It requires the discipline to throw away your favorite ideas. None of those are the things AI is currently great at.
But there are parts of the work that AI is great at, and we noticed them when we started talking to product teams about how they actually spend their weeks.
What teams told us they hated
When we did the early customer-development calls for Dodger, three things came up over and over:
- Synthesis is the slowest part. Reading 40 transcripts and pulling out a coherent point of view takes days, and most of it is mechanical.
- Personas rot fast. A persona deck written six months ago describes a market that doesn’t exist anymore. Nobody has time to redo it.
- You can’t run the cheap experiment. It is too expensive — in calendar time, in recruiter cost, in goodwill — to run a 12-person interview round just to validate one branch of one idea.
That last one is the one we kept circling. The cost of thinking through an idea is high, so people don’t do it. They just build, and then justify what they built.
What we’re trying to do
Dodger is trying to make the cost of thinking through an idea low enough that it becomes a real step in the workflow again — somewhere between “have a thought in the shower” and “start a sprint.”
The way we’re doing it has three parts. None of them are individually new; the combination is what we think is new.
- Synthetic personas you can argue with. Not bullet-pointed slide-deck personas — characters with budgets, gripes, and competing priorities, that you can interview the way you’d interview a real person.
- A canvas, not a chat. Research lives somewhere. You should be able to point at a quote from one persona and have it inform the wireframe two columns over.
- Commitments instead of suggestions. A research output that says “users probably want X” is worth nothing. A research output that says “users with this profile, asked this exact question, said this exact thing” is worth something.
The part where we hedge
The honest version of all of this is that synthetic users are not real users, and synthesis is not a substitute for talking to people who pay you money. Anything Dodger gives you is a hypothesis. Hypotheses are valuable; certainty is not what we sell.
What we are trying to sell is a faster loop. Talking to ten real users a week is hard. Stress-testing a positioning statement against fifteen synthetic personas before you call the ten real users — so the calls are sharper and the questions are better — is something most teams will do, if you make it easy.
That is the bet.
If any of this resonates, start a free canvas and tell us what you find. The free tier exists exactly so you can argue with the output before you pay us anything.
We’d rather you push back than nod along. There’s enough nodding-along software in the world already.