Jan van den Berg opens with a scene anyone in tech recognizes: four people failing to connect a laptop to a screen, each additional helper making things worse. It's a neat setup for revisiting Frederick Brooks's 1975 insight that adding people to a late project makes it later — and asking whether AI agents break the pattern. The answer van den Berg arrives at is a qualified no. The sequential-task constraint (his "Baby Problem" — nine agents can't deliver a baby in one month) remains unchanged. What shifted is the nature of overhead: instead of onboarding humans and navigating office politics, you're now defining done-states, reviewing AI output, catching hallucinations, and stitching agent work together. The coordination tax didn't vanish; it moved from the team to the individual operator. The essay's sharpest insight is that unclear requirements become MORE dangerous with AI agents, not less. A confused human team at least argues about scope. Nine autonomous agents with ambiguous instructions will "happily rewrite each other's work in a high-speed loop." Speed without direction is just expensive chaos. Van den Berg's proposed 2026 update — "Adding agents to a project with unclear requirements gets you the wrong answer faster" — is punchy and earns its place. It captures something real about the current moment: the bottleneck has migrated from labor supply to specification quality. The person who can articulate what "done" looks like is now the entire critical path. Where the essay falls short is depth. It doesn't engage with the real counter-arguments: multi-agent orchestration frameworks (CrewAI, AutoGen) that attempt to solve exactly the coordination problem he describes, or the empirical evidence from teams actually shipping with AI agents at scale. The Brooks analogy is load-bearing, but the piece doesn't stress-test it enough to know if it holds or just rhymes. The format — short blog post, personal anecdote, one big idea — is well-matched to the argument's weight. This isn't trying to be a treatise, and it shouldn't be. It's a lunchtime provocation that makes you think for five minutes about why your AI-augmented workflow still feels bottlenecked. That's enough. Still, the essay is more conversation-starter than finished argument. The reader who already manages AI agents daily will nod along but won't learn anything new. The reader who hasn't yet will get a useful mental model for what's coming.
