Post #4042102

2026-07-23 17:31 UTC

Those who value speed over quality will inevitably see quality drop. LLMs bring speed that makes reviews, design and just *thinking* about what you're doing very slow in comparison to the code generation. Therefore, you'll take shortcuts to keep up. Therefore, quality will drop. Guardrails discussions are a red herring. You want quality, but the biggest bottleneck was always the thinking, and the many decisions that you made along the way. Now you delegate these decisions to chance.

Replies (3)

@ianh@phpc.social 2026-07-23 18:24

@afilina@phpc.social In $previousJob, the junior on the team was leaning heavily on AI. Sometimes, this made significant changes to certain Eloquent queries (Laravel codebase) that changed nothing in way of functionality, but went from a single chained method to more than half a dozen. The cognitive load trying to do code reviews on that is 100% wasteful. Like life, everything is a compromise. Quicker code output is slowing reviews at best, worse, contributes to needless burnout.

Open ##4043122
@Traumkaempfer@sueden.social 2026-07-23 18:33

@afilina@phpc.social someone - I still don’t know if it’s management or the consultants who are supposed to support in the QA team - are heavily promoting the use of „AI“ to write test cases and automate them. My internal colleagues and I are responsible that our payment services are working reliably and we all doubt „AI“ will help us because we know that doing that manually is how we learn the most about our systems.

Open ##4043308
@bobmagicii@phpc.social 2026-07-23 17:33

@afilina@phpc.social but but but "my custom guardrails" aka "i wrote a paragraph that tends to sway the rng to work more often than not with the current state of the non-deterministic code im talking to"

Open ##4145702