Months In, a Bioinformatics Dev Says Claude Code Still Can't Write Production-Grade Code
adacomb · reddit · 2026-10-06
A bioinformatics developer shared a detailed Reddit postmortem of his team's months with Claude Code: the more critical the code, the less useful it gets. He concedes Claude is superb for non-critical, low-readability tasks, but on production code where correctness, readability and consistency matter, the baseline is too poor to bother with.
His daily fixes include: designs and abstractions that don't look human-written down to function signatures; invented terminology and bad naming; comments and docs that all get rewritten; error-swallowing and silent failures that are unpredictable; algorithmic inefficiency and recomputed values; band-aid fixes instead of optimal solutions; and pervasive verbosity.
He addresses the standard advice—repo init, planning, precise prompts, skills—and argues it doesn't fix the core problem: if planning must go down to variable naming, writing it yourself is faster; CLAUDE.md instructions get forgotten within a few messages; and the cognitive cost of repeatedly teaching fundamentals isn't worth it. His conclusion: Claude's utility is highly polarized, and its value shrinks as code importance rises—the opposite of what a good tool should do.
More from coding & agent
- Meta paper: dual coding agents cross-review lift correct patches from 45.8% to 62.5% — rohanpaul_ai · 2026-10-06
- Rewriting from scratch beats depending on a single maintainer with feelings — l4rz · 2026-10-06
- Use auto-review instead of full-access for coding agents — soumitrashukla9 · 2026-10-06
- Zeroization can make things worse: how wiping secrets creates more copies — jedisct1 · 2026-10-06
- Indie dev launches agent-first creator marketing platform Clipatra, pays out $3,000+ — tibo_maker · 2026-10-06
- TensorFold bonds dual Thunderbolt 5 links for 83% throughput boost on Apple Silicon — AIFlow_ML · 2026-10-06