How to ship an MVP with AI agents in two weeks

Our updated MVP playbook ships every build with a working AI agent layer from day one — tool registry, eval harness, typed actions and a Claude-powered planner, ready on day fourteen.

Architecture diagram of the krylov® MVP playbook with an AI agent layer

Our two-week MVP program has always been about one thing: the smallest shippable proof that the idea is real. From this month on, every MVP includes a working AI agent layer by default — and we think it should have been that way from the start.

What ships in the new MVP playbook

Previously, AI agents were an optional Phase 2. Now they are Phase 1 plumbing. Every MVP ships with an auth-aware tool registry, a lightweight LLM eval harness, a typed action schema and a default planner wired to Claude. Teams still pick their own product logic — but the agent foundation is tested and instrumented from day one.

AI agent architecture — tool registry, planner, evaluator — The three components every MVP now ships with from day one.

Those three components are not a diagram drawn for this post. They are the same registry, harness and planner that land in the repository on day one of every build.

What the two weeks look like from inside

Fourteen days leaves no room for ceremony. Discovery on day one, a walking skeleton by day four, the agent layer wired and evaluated by day nine, and the remaining time spent making it survive real users.

Studio reel.

Why AI agents are the new MVP baseline

Two observations pushed us here. First, founders cannot credibly raise without an AI story any more — investors ask on the first call. Second, agents went from "brittle magic" to "boring glue" over the last twelve months, which drops the cost of adding them to an MVP by an order of magnitude.

What it means for new builds

Same timeline, same price, more shippable surface on day 14. Teams that were going to add agents in Phase 2 get them four weeks earlier. Teams that were not get the option with no extra engineering budget.