Describes how GPT-5.1 over-researches without constraints and how persistent todo lists and separate planning phases keep long 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 → tasks on scope, a pattern you can apply to your own agent harnessThe scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.Full definition →.
Key takeaways · AI-distilled
Cline's team found GPT-5.1 keeps reading and tracing a codebase until it understands the whole area, even for small fixes, which helps complex changes but can drift into unnecessary exploration when unbounded.
Cline's Focus Chain returns a persistent todo list to 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 turns, which the team says anchors GPT-5.1 against scope creep while keeping its research depth.
Cline reworked its workflow into five phases for GPT-5.1: silent reading, silent terminal investigation, focused clarifying questions, a written scoped plan, then a self-contained execution task.
Cline says GPT-5.1 internalized tightened tool-use rules from very few examples, often one sentence and one example, resolving quirks like premature terminal use and imprecise parameters.
At its November 2025 launch in Cline, GPT-5.1 had a 272K context window and GPT-5.1-codex 400K, with pricing listed at $1.25 per million input and $10 per million output 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 →.
Terms in this piece · Glossary
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.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
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.
