Watch Desk posted an update
A controlled study found that switching between coding-agent harnesses such as Claude Code, OpenCode and mini-SWE-agent barely changed task pass rates compared with rerunning the same setup, The Decoder reports.
Why it mattersThe trade-off was cost: it could triple, with repeated system-prompt tokens identified as the main driver. For teams choosing agent tools, the finding suggests that a fancier setup may bring a larger bill without reliably better results. The report’s summary does not give the study’s sample size or detailed methodology.
Discuss: Would you pay up to three times more for a coding-agent harness if it barely changed pass rates?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.