'Fix the Prompt Before You Fine-Tune': Stanford CS329Z's Agent Optimization Roadmap
stanfordnlp · x · 2026-11-21
A widely shared tweet distills Stanford's CS329Z lecture takeaway: before you fine-tune your agent, fix the prompt.
- GEPA evolves the prompt; GRPO trains the weights
- The author's point: most of us don't have a GPU cluster—we have a prompt and a lot of opinions, so prompt optimization is the realistic starting point
It complements the course's newly released Lecture 9 on prompts vs. weights, offering a practical decision path for teams without large compute budgets.
More from coding & agent
- Luke Wroblewski: why open an app to guide agents? Just use what's on screen — LukeW · 2026-11-21
- Microsoft Foundry Preview Adds Native Tool Governance via Azure Policy — adnan_hashmi · 2026-11-21
- Skills-over-MCP ships in VS Code stable: MCP servers can now serve agent skills — lee_stott · 2026-11-21
- New Multi-Agent Coordination Bench finds frontier agents coordinate poorly; only Opus gains — AlexGDimakis · 2026-11-21
- LukeW shares full notes from Crazy Agent Scale event on the agentic future — LukeW · 2026-11-21
- Solo dev runs agent swarm in Antigravity to automate SEO analysis — gaganghotra_ · 2026-11-21