Harness, Loops, Graphs: breaking down the 7 levels of agent autonomy, from babysitting to running a fleet

blaizedsouza · x · 2026-11-01

The post quotes an article titled "Harness, Loops, Graphs: What Each Layer Fixes and What It Breaks," arguing that almost every "autonomous" agent setup still has a human in the loop: it writes, you read; it says done, you verify; it breaks things, you explain. The piece maps 7 levels of autonomy, from babysitting agents to running a fleet.
The quoted tweet amplifies Lauren Tan (ex-Cursor, now senior engineer at SpaceXAI): she no longer writes code by hand but runs 50+ agents — a Chief of Staff planning her day, 3 PMs splitting work, 40+ worker agents building overnight, outputting the equivalent of 20 engineers. She claims 97% of SpaceXAI engineers already work this way.
Tone is hype-heavy ("99% of people use one chat tab"), but the underlying framework on where agent autonomy actually breaks down is worth reading for anyone designing multi-agent workflows.

More from coding & agent