DeepLearning.AI free course teaches agents to turn traces into reusable skills
DeepLearningAI · x · 2026-11-18
DeepLearning.AI and Oracle launched a free short course, Building Adaptive AI Agents (1h, 7 video lessons), addressing why agents repeat the same mistakes: give them mechanisms to carry experience forward. It covers three adaptation layers:
- Behavior: a skill induction pipeline that turns agent traces (conversations, tool calls, errors, fixes) into reusable skills, each human-approved before reuse;
- Knowledge: a code knowledge graph connecting files to surface relevant functions and context, shown to beat keyword search;
- Model: where the model itself can adapt.
Core thesis: today's coding agents start every session from zero, re-burning time and tokens on fixes discovered yesterday.
More from coding & agent
- Punchcutter skill turns a photo of your handwriting into a usable font — vista8 · 2026-11-18
- AI researcher John Frankle: AI-written code made software more brittle; correlated outages aren't a conspiracy — jefrankle · 2026-11-18
- Copilot reportedly getting multiplayer Workspaces for teams and AI agents to collaborate — testingcatalog · 2026-11-18
- Microsoft Copilot adds multiplayer Workspaces for teams and AI agents to work together — testingcatalog · 2026-11-18
- Nadella frames Copilot in three modes: Home assistant, Code factory, and Autopilot teammate — donasarkar · 2026-11-18
- supermemory launches Antigravity plugin sharing one memory across Claude Code, Cursor, Codex — Scobleizer · 2026-11-18