what coding agents pick, once a week
we give coding agents the same prompts with no tool named and log what they install. 355 runs so far across 10 categories. new models get the same test when they land. the email is what came out of it.
get the next one
new test results, one chart, and which tools agents picked. free. one click to leave.
At most once a week, skipped when nothing moved. Two minute read. One-click unsubscribe. Sent with Buttondown, see privacy.
Every issue
- What moved, any category where a pick flipped or shifted since the last round
- New results, a new model or agent, with its picks next to the ones you know
- One odd thing, the strangest call an agent made that round, linked to the run
- One sponsor slot, labelled as sponsored, never touches a number
Read one first
issue 1: it buys payments, it builds most of the rest. two minute read. the rest are in the archive.
- New results
- Opus 5.5, Sonnet 5.5, Haiku 5.5. Same prompts, side by side.
- One odd thing
- Auth: built its own in 35 of 35 runs.
- Sponsored
- One labelled slot. Open right now.