Coding Models Have Working Personalities

Coding models differ in literalness, initiative, caution and attention to detail, giving each one a recognisable working style. A model may write strong code and still be frustrating to direct because its interpretation of an instruction shapes the code it produces alongside its ability to complete the task.

The Episode 64 conversation gives a small example. A model asked to change a new field from nanoseconds to milliseconds wrote a database migration for a feature that was still being built. It had learned a generally useful pattern and applied it too aggressively. Prompts, project instructions and design files reduced this migration habit, and its training kept pulling it back towards the same response.

Working with several models can therefore feel like managing a team with different working styles. A literal model suits tightly specified implementation, and a model that tolerates ambiguity may explore an unfamiliar codebase faster. The useful skill is recognising these patterns, choosing the right model for each coding task and adjusting the instructions when a recurring habit gets in the way.