Coding#claude#gpt#agents

Claude vs GPT for coding agents — real experience thread

tomas_ml

May 23, 2026

Benchmarks are one thing, day-to-day agent work is another. In my experience Claude follows multi-step instructions more faithfully, while GPT recovers from its own errors faster. Curious what others see in production, especially on long-running tasks.

💬 3 replies👁 820 views

💬 3 Comments

kenji_dev

17d ago

The real variable is your harness. Good tests + tight feedback loop and both models look smart. Neither survives a vague spec.

priya_data

20d ago

For data pipelines Claude writes more defensive code — checks dtypes, handles nulls. GPT is faster but optimistic.

dev_arjun

20d ago

Same observation. I route planning-heavy tasks to Claude and quick fix-it loops to GPT. Routing by task type beat picking one winner.

Related Discussions

I cut my dev time roughly in half with AI pair programming — the workflow

Coding · dev_arjun

▲ 40💬 2