A roadmap for the agent era: three bets, not thirty

Why long roadmaps struggle now

Features are cheaper to build than they used to be, and cheaper for competitors to copy. A roadmap that is a list of things to ship doesn't give you an advantage. It gives you a backlog.

What's hard to copy is a position: the data you hold, the workflow you own, the trust you've earned with a particular kind of customer.

Three bets

Frame each item as a bet, with a sentence of the form: "We believe X, so we will do Y, and we'll know we're right if Z."

For example: "We believe our customers' biggest cost is manual review, so we will automate the first pass, and we'll know we're right if review time per case falls by a third."

Aim for three. One is about deepening what already works. One extends into an adjacent problem for the same customer. One is a longer shot you could be wrong about.

Say what would prove you wrong

This is the part teams skip. Before building, write the result that would make you stop. It keeps you from defending a weak idea for six months because you've already invested in it.

Where AI fits

Treat AI as a means, not a bet in itself. "Add an AI assistant" isn't a bet. "Cut the time to prepare a quote from a day to an hour" is, and AI might be how.

What to do this month

1. Write your three bets in the "We believe, so we will, and we'll know if" format.
2. Add a stop condition to each. A number and a date.
3. Cut the rest from the active roadmap. Put it in a parking list.
4. Pick one measure per bet. One metric that moves if the bet is right.
5. Schedule a quarterly review. Keep, change or drop each bet.