Experienced dev: autonomous agents are overkill for 90% of coding — one-pass edits win
cgouguen · reddit · 2026-09-12
A senior developer who maintains large codebases and built his own lightweight AI CLI (Frugaast) argues against the default of fully autonomous coding agents:
- Autonomous loops are unpredictable: they hallucinate unnecessary refactors, burn tokens on multi-step runs, and bloat the context window with every iteration.
- By the time you've written the perfect spec and reviewed a 15-step plan, you could have shipped the feature yourself.
His 3-step workflow
- Strict context curation: manually pick exactly the files needed — seconds of work if you know your architecture.
- One-pass execution: simple tasks via free web UIs (Claude/ChatGPT), complex ones via direct-edit CLIs (Aider, Claude Code, Frugaast).
- Diff & ship: with tightly human-scoped prompts, the AI gets it right on the first try 99% of the time.
His warning: leaning on agents erodes your mental map of the codebase. Curate context, keep architectural control, and let AI handle only the typing.
Related event: Veteran Developer Argues 90% of Coding Needs No Autonomous Agents(2 posts)→
More from coding & agent
- GitHub Copilot app is building richer dashboard widgets for personalized data views — mariorod1 · 2026-10-31
- Todoist's self-healing agent auto-fixes 1 in 5 Sentry issues and opens PRs — zeeg · 2026-10-31
- Dev's Multica 1.0 build log: 432M tokens, 89 agent runs, just $183 — jiayuan_jy · 2026-10-31
- Stardock's QA now runs agents that play the game and file reports — draginol · 2026-10-31
- Limma MCP structures unstructured data end-to-end so agents never miss a detail — FlimsyIngenuity7248 · 2026-10-31
- Agent audit metric: can you name which tests ran and files changed without reading its summary? — ronbodkin · 2026-10-31