Jev Daily

Theo re-ran it: 5% fewer false reports, still bad

Theo says fixing broken ACP bindings cut his false reports by about 5% and the model is still, in his words, utter garbage for real world work. Greg Isenberg, meanwhile, summarised his entire company stack in five words.

Built with it

Isenberg boils his 50-person company down to voice notes and inference

Greg Isenberg restated his agent stack as a single line: running a company is now voice notes and inference. The underlying post describes a 50-plus person team across 12 countries where specs start as a voice note to grokbot, agents run overnight in parallel git worktrees, and humans get version five.

GREG ISENBERG
@gregisenberg
X
my team is 50+ people in 12 countries, but i talk to my agents more than my human teammates
Nov 3, 2026 · View on X

Read the full story

The case against

Theo re-tested a coding model on CLI, says it is still unusable

Theo says the ACP bindings were broken during his earlier test of a coding model, so he re-ran it through the CLI and false reports dropped by about 5 percent. He still calls the result unusable for real codebases, having earlier said it hallucinates in egregious ways and is not 5 percent worse but 10x worse.

Theo - t3.gg
@theo
X
Better, but still utter garbage for real world work
Nov 4, 2026 · View on X

Read the full story

Get the next one by email

Jev, read daily so you do not have to. The builds, the benchmarks, the criteria that worked and the cases where it lost, from the people shipping on TypeSafe AI's System One model.