Two engineers discovered the same thing from different angles: AI that works for hours, not minutes.

Ryan set 17 failing tests as the goal at 6 PM. By morning, Claude had iterated through them all - diagnosing, fixing, learning from each failure. No babysitting. The computer worked while he slept.

Brooker, a former hedge fund analyst, faced a different problem: 5-hour earnings previews that were outdated by next day. Now he describes what he wants - "build an earnings preview for Meta" - and gets an interactive dashboard in 20 minutes. Beat/miss history, guidance analysis, trend charts. All from natural language.

The pattern: clear success criteria + extended compute = AI that does the tedious iteration for you.

Whether that's autonomous loops that fix tests overnight or custom research infrastructure that runs as long as the analysis requires - the ceiling on AI work has lifted.

P.S. What's the most tedious iterative task in your workflow? Reply with one - I'm collecting patterns.