Loading...
Loading...
This article by Teresa Torres explains 'context rot'—the phenomenon where AI model performance degrades as conversations lengthen and models lose access to information within their context window. The piece covers context window basics (Claude Opus/Sonnet: 200K tokens; GPT-5.2: 400K; Gemini 2.5 Pro: 1M), research findings on degradation patterns, and why consumer web interfaces provide no visibility into context window status.
The article highlights Claude Code's advantages including context window percentage display, the /context command for examination, and /clear and /compact commands for active management. Practical recommendations include monitoring context percentages, keeping CLAUDE.md files efficient, using the file system to offload context, and strategically employing agents and sub-agents. Essential reading for PMs working with AI tools who need to understand why AI performance degrades and how to mitigate it.
Building on foundational concepts, this resource explores technical skills at a deeper level. It's designed for PMs who have some AI experience and want to develop more sophisticated skills.
Ready to explore this resource?
Go to producttalk.orgThis free interactive course teaches product managers how to use Claude Code—Anthropic's CLI tool—for AI-powered PM work. Uniquely, the course is taug...
This comprehensive guide addresses systematic decay in AI systems through structured prompt optimization practices. The article establishes that promp...
This guide by Miqdad Jaffer (OpenAI Product Lead) establishes context engineering as the foundational discipline for building intelligent AI products....