AI Collapses the Roadmap

The one-year plan becomes a two-day build

Ryan Petersen, founder of Flexport, describes a change in how product gets planned rather than just how fast it ships. "There's way less planning than there used to be at Flexport in the road map. It used to be a one-year strategic plan...Now you're kind of like: alright, let's go see if you can make this work, come back to me two days later, and it works."¹

The mechanics: Flexport has pivoted roughly 30 to 40 percent of its engineers to building agents, with about 50 agents in development at once. The canonical example is a bill-audit agent built in a single week that did not exist when the week started. It checks every incoming carrier and trucking bill against what was quoted, the procurement database, and what actually happened on the shipment, reading message logs and event logs to decide whether the amount is correct and whether the customer was billed properly. Petersen: "I didn't know last week that we would have that. I thought it would be cool, I told them to work on it."

Why the roadmap itself is the casualty

When a feature takes a week instead of a quarter, the expensive resource is no longer building, it is deciding what to try. Planning a year ahead becomes negative expected value because the cost of attempting something has fallen below the cost of debating it. The roadmap turns into a queue of cheap experiments, and the leadership task shifts from sequencing scarce engineering capacity to generating and triaging ideas.

Petersen's own account of the unlock is blunt: AI is "better at writing code than the engineers," and an engineer paired with frontier coding models is "10x, maybe 100x" more productive.

The tradeoff

The two-day cadence assumes uninterrupted access to frontier models, so the whole approach is exposed if that access is cut off. There is also a build-debt risk in spinning up roughly 50 agents at speed: fast to build is not the same as done, and Petersen's own bar is getting a large majority of in-flight workflows to real impact rather than just existing. The aspiration he describes, operations dialed in without anyone needing to log in and check, is a destination; the near-term reality is a large pipeline still mostly under construction.