The Worst Instructions You Can Give An Ai Coding Agent
Source
cline.bot
Author
Cline evaluation and engineering blog
Date
Why it matters
Vague instructions like 'make it better' make agents guess and burn tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → on reverted output. Concrete scoping of the change and its reason cuts wasted runs.
Key takeaways · AI-distilled
Prompts that pair conflicting constraints, such as "add OAuth but keep it simple," lead the AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → to pick one side rather than weigh the tradeoff. The author recommends stating which priority wins and any hard limits, such as no new dependencies.
Citing Laban et al., the post says LLMs lose 39% of performance on average when instructions arrive across turns rather than up front, so it recommends restoring a checkpoint and editing the prompt where things went wrong instead of piling on corrections.
A useful bug request names the symptom, the fixes already tried and why they failed, what was ruled out, and the suspected location. Without that history, the post says, the agent tends to propose fixes the developer already tried.
Terse, commit-style prompts get literal execution: asking only to convert a method to async/await yields a behavior-identical rewrite. The post says adding the reason, such as uncaught database errors, lets the agent handle unstated edge cases.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
article(Im)Paired Programming: Coding Agents Improve Productivity but Harm UnderstandingNishant Balepur, Connor Baumler, Valerie Chen, Eunsol Choi, Rachel Rudinger, Jordan Lee Boyd-Graber
videoStop babysitting your agents... — Brandon Waselnuk, UnblockedAI Engineer
postBuilding CLIs for AI Agentseric zakariasson