Maybe my job is just too easy, but all the hoops that folks jump through to get the magic oracle to do the thing takes longer than if I just Did The Thing
Comments like this are not much worth without context. Each model has wildly different perf per each language and framework, project architecture (if it can be followed up successfully). No two devs on different projects have the same experience. Even insights like "Anthropic has a lead" is a broad generalization.
Can’t give an extensive reply as I’m on mobile, but FWIW I’ve tried Claude and Codex and Copilot. We use Python and standard microservice architecture, nothing special at all.
I’ve had the most success with using one agent to write a plan document and another to follow that plan doc.
Or maybe it's too difficult? Or maybe you are just holding it wrong
The unpredictability for using things like cursor or Claude code is just a showstopper and indeed I'm not sure it has ever saved me time in the grand scheme of things