The code deletion pattern is real and underappreciated. Each new model release makes you realize how much of your scaffolding was compensating for model limitations rather than solving actual problems.
The bottleneck shift - from writing to review - is where I keep getting surprised too. Shipping faster exposes review as the actual constraint. The METR finding is more honest than most productivity benchmarks I've seen: the developers who felt faster were slower.
For my setup the wins show up in the boring parts - context loading, cross-file consistency - not the impressive demos. The TypeScript fluency argument for the tech stack is clever though. Self-maintaining codebases change the calculus on language choice.
Ha, the "code you thought was clever but was actually a model workaround" graveyard must be growing fast. What's the biggest thing you deleted and went "wait, that's all it needed?"
The code deletion pattern is real and underappreciated. Each new model release makes you realize how much of your scaffolding was compensating for model limitations rather than solving actual problems.
The bottleneck shift - from writing to review - is where I keep getting surprised too. Shipping faster exposes review as the actual constraint. The METR finding is more honest than most productivity benchmarks I've seen: the developers who felt faster were slower.
For my setup the wins show up in the boring parts - context loading, cross-file consistency - not the impressive demos. The TypeScript fluency argument for the tech stack is clever though. Self-maintaining codebases change the calculus on language choice.
Ha, the "code you thought was clever but was actually a model workaround" graveyard must be growing fast. What's the biggest thing you deleted and went "wait, that's all it needed?"