Bigger context windows do not fix multi-turn degradation. The post describes how Deep Planning front-loads codebase investigation so long tasks start with clean context, a pattern you can copy.
Key takeaways · AI-distilled
Cline illustrates compounding multi-turn loss: 95% accuracy on turn one might fall to 92% on turn two and 70% by turn ten, as the model's own explanations, error fixes, tool outputs and failed attempts pile up.
/deep-planning runs four steps: silent codebase investigation, targeted clarifying questions, an implementation_plan.md saved to the project root, then a fresh task seeded with that plan and @-referenced key files.
The Focus Chain is on by default in v3.25 and re-injects the todo list into context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → every six messages by default, reminding 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 → what is done and what comes next.
Auto Compact replaces bloated history near the context limit with a summary of decisions, code changes and progress; Cline says this lets a task needing 5 million 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 → of interaction finish in a 200k window.
Terms in this piece · Glossary
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.
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.
