2 min read
Three AI product bets for 2026
Write the eval set before the spec, ship agents like teammates, and route each task to the right model instead of the biggest one. AI product work is starting to look like systems design.
Three calls I'm making as a product leader for the back half of 2026. None of them are about the models themselves. They're about how product work changes when generating things is cheap and judgment is the bottleneck.
1. Evals are the new PRD
The team that writes its test set first ships faster than the team that writes the spec first. An eval set is a list of example inputs, each with a clear pass or fail. With one, you can try ten model variants in an afternoon and pick the winner on evidence. Without one, you argue in Slack for a week.
It doesn't need to be fancy. Twenty hand-written cases with clear criteria gets you most of the value. I'd attach one to every AI feature before it reaches production.
2. Agents are products, not features
The 2024 reflex was to bolt a chatbot onto the side of the app. The better move now is to ship an agent the way you'd bring on a teammate: a clear job, a defined scope, the tools it's allowed to use, and a point where it hands back to a person.
The question stops being “how do we add AI here?” It becomes “what is this agent allowed to do, when does it escalate, and where does a person stay involved?” That's a product spec, not a model spec.
3. Route each task to the right model
I'm watching teams default to the largest model they can afford for every feature. Cost goes up, speed goes down, and quality often doesn't move. The right call is task by task, sometimes step by step inside a task.
Haiku for fast, structured calls. Sonnet for judgment and writing. Opus when the problem justifies the time and money. A small routing layer in front of each feature, with budgets and fallbacks, beats “use Opus everywhere” on everything the user and the person paying the bill care about.
What this means for the next hire
The AI product leader I'd hire in 2026 isn't the one who writes the best prompt. It's the one who writes the eval set, designs the routing, scopes the agents and puts people in the loop in the right places. The work looks more like systems design than copywriting.
If you're building toward this and want to compare notes, get in touch.
Questions people ask about this note
- What are the most important AI product bets for 2026?
- Three. Write the eval set before the spec. Ship agents like teammates, with a clear job, scope and handoff to a person. Route each task to the right model instead of using the biggest one everywhere.
- How is the role of an AI product manager changing in 2026?
- The job moves toward writing eval sets, designing how tasks are routed between models, scoping agents, and deciding where people stay involved. It looks more like systems design than prompt writing.
- Why are AI evals more important than PRDs in 2026?
- With an eval set you can run a feature against ten model variants in an afternoon and pick the winner on evidence. Without one, the decision is opinion. Evals turn AI product work into measurement.
Photo: Sunset, Georgian Bay. EriktheZ, CC BY-SA 4.0, via Wikimedia Commons.