Free DeepLearning.AI course teaches agents to turn their own traces into reusable skills
minchoi · x · 2026-09-10
minchoi highlights a free 1-hour DeepLearning.AI course built with Oracle, Building Adaptive AI Agents, on making agents carry experience across sessions.
Core thesis: most agents have retrieval plus prompt inflation, not memory. RAG looks things up; a memory system decides what's worth keeping (policies, preferences, facts, episodes, traces) and rebuilds the prompt from it each turn. The missing piece isn't a bigger context window — it's a write path.
The course covers three adaptation layers:
- Behavior: a skill induction pipeline that converts agent traces (conversations, tool calls, errors, fixes) into reusable skills, each human-approved;
- Knowledge: a code knowledge graph connecting files via imports and calls, shown to beat keyword search as context;
- Model: when fine-tuning is worth it.
Includes 7 video lessons, 3 code examples, and a graded assignment, taught by Nacho Martínez and Casius Lee of Oracle.
More from coding & agent
- Hermes Agent Adds QR Quick Setup, Telegram Config Now Nearly Effortless — Teknium · 2026-09-11
- Cognition's Astra Sees Unprecedented Demand, May Pause New Pro Subscriptions — taherdhanera · 2026-09-11
- Cognition Ships SWE-2 at 70% Lower Cost Alongside $2B Raise — DevinAI · 2026-09-11
- Agent-built small classifier labels 191K documents for $0.70, vs $13–26 with batch LLMs — vanstriendaniel · 2026-09-11
- Microsoft MVP Demos Personal AGI: Five Tools, Markdown and One LLM — dfinke · 2026-09-11
- Maestro diagnoses AI coding workflow issues with 25 targeted commands — tom_doerr · 2026-09-11