Should you build an internal tool because an AI agent can build it in 20 minutes?

THE SHORT ANSWER

Not for that reason. When an agent can build almost anything in minutes, build time no longer separates good builds from bad ones. Jason Lemkin's SaaStr replaced Calendly with an agent-built booker in about 20 minutes (2026-10-03), and it was a good build because it uses data only SaaStr has, not because it was fast. I approved 39 agent builds in 80 days because each was cheap, and 13 ended with zero references anywhere else in the system. Keep SaaStr's four questions (what the vendor cannot do, whether an integration covers it, what data only you have, what happens if it breaks) and add three: the named person who notices when it breaks, the existing workflow it joins, and the date you switch it off for a week to see who complains.

The longer argument is in Twenty Minutes Is Not a Reason to Build. Jason Lemkin's Calendly post gives four questions SaaStr asks before an agent builds anything, and then one line that sets the bar by build time: at 20 minutes it was low, and at two weeks they would have kept Calendly. The four questions are worth keeping. The bar is the part that breaks once every build is minutes.

The three lines I add come from counting my own fleet in 39 PM AI Agents Deployed: What Stuck, What Died, and Why: 39 agents in 80 days, 13 with nothing depending on them, and four traits shared by the ones that lasted (a named owner, narrow scope, an existing workflow, and stable inputs). The switch-off test is from The Seven-Agent Reset: A Lean PM AI Fleet, One Agent Per Stage. The whole thing belongs to Building a Company in Public, because the evidence is a set of my own calls that turned out wrong.

SOURCES

THE LONG VERSION

RELATED ANSWERS

Last reviewed 2026-10-05 · 1 min read