Your Riskiest Assumption Might Live in the Workflow
Maurya's lean canvas forces you to name what has to be true for your business to work. When agents do the work, a new class of assumption enters the canvas: that the agent will perform reliably enough to deliver the promise. Before building elaborate autonomy, ask whether the path is actually fixed. Most tasks that feel like agent problems are workflow problems in disguise — a deterministic sequence that needs no branching at all. Prove the workflow version first, then graduate to agency only where genuine branching is required.
Problem Interviews Apply to Eval Design
Maurya runs problem interviews to test whether the pain is real before writing a line of code. The analog in agentic systems is building your evaluation harness before you build the agent. Talk to the people who will consume or supervise the agent's output. The failure modes they describe become your red cases. An eval built from real frustrations is worth more than a suite of cases you invented at the desk, because it attacks the assumption you most need to break.
Solution Interviews Expose Tool and Context Gaps
When Maurya moves to solution interviews, he is checking whether his proposed fix actually addresses the discovered pain. For agent builders, this maps to testing whether your tool surface and context window genuinely support the task. A tool named vaguely or scoped too broadly produces outputs nobody trusts. Running structured interviews with early users — watching where they hesitate, where they override the agent — tells you exactly which tools need sharper edges and which context is missing or cluttered.
Humans Hold the Judgment the Canvas Cannot Automate
Maurya's whole method assumes a human founder is making the calls: deciding which assumption is riskiest, choosing when the evidence is sufficient, pivoting or persevering. That role does not disappear when agents scale the execution. Setting goals, owning accountability, and drawing the boundary between what the system decides and what a person must decide — these are the founder's job regardless of how much the agents can do. The lean canvas is a tool for human judgment, and so is every eval harness you build around it.
