Skip the fine-tune: for an apartment-hunting agent, an MCP search tool is all you need
dbasch · x · 2026-09-18
Responding to someone eager to fine-tune their first model to find a dream apartment in Buenos Aires, dbasch argues fine-tuning is the wrong tool: what you want is an MCP letting Claude or ChatGPT search listings. With data in hand, sorting through it is trivial for the model — the hard part is having the data, not training.
Related event: Don't Fine-Tune: An MCP Server Is Often Enough for Personalized AI(2 posts)→
More from coding & agent
- AI-run company makes its first $75: what running a business with Devin taught us — sandylikesfrogs · 2026-09-18
- Codex ships Code Mode: JS programs filter MCP data before it hits context — alex_frantic · 2026-09-18
- W&B floats chart annotation feature: pin notes like '58k: added attention gate decay' to training runs — yoavartzi · 2026-09-18
- Claude Code relaunches Projects to manage multiple cloud agents — The Verge AI · 2026-09-18
- Scale AI CEO Alexandr Wang hypes Muse agent: 'surprised by how much it can take on' — alexandr_wang · 2026-09-18
- Included Health built federated healthcare agents on LangGraph, lifting chat engagement 75% — LangChain · 2026-09-18