Friday evening is when most free-tier experiments begin. An engineer assigns a coding agent a large refactor, armed with a generous token allowance and a free server. Monday morning the allowance is gone, the diff runs to fourteen hundred lines, and nobody has reviewed a single one of them. The pattern dominates this account's postmortem archive, and it explains why a recent DEV discussion concluded that AI promoted every developer to reviewer while nobody tested the reviewer.
That thread asked what developers actually do while the agent codes. The honest answer, drawn from the runaway-diff and cold-start failures documented here, is that they build a review queue they never scheduled time to drain. Tokens were never the bottleneck in any of those stories; attention was always the bottleneck. This article is a myth-busting FAQ built from those failures, where each claim is a sentence developers repeat and each correction is a mental model that survives a Monday-morning diff.
Myth 1: A free token allowance makes bad plans cheap.
The arithmetic looks generous on paper, because ten million tokens and a free server carry no invoice at all. The hidden line item is the review queue, since every wasted plan still produces a diff that a human must read, judge, and repair. A bad plan at free-token prices simply converts the cheapest resource into the most expensive one, which is a reviewer's attention. The corrected model is that a quota is a rate limiter, never a value meter.
