How much should a data agent learn upfront? A team's discovery-layer design
prous5tmaker · reddit · 2026-10-19
A team building a proactive data agent (surfacing growth/revenue issues and posting to Slack) shares its design trade-off between upfront 'discovery' and on-the-fly lookups. V1 used an LLM agent to inspect schemas and write a catalog of tables, columns, and joins. Their new path moves schema collection into code — reading names, columns, types, and key constraints from metadata, grouping shard tables/partitions, and matching layouts — while the model only decides which tables matter, whether matching layouts are copies, and what tables mean; code validates outputs against the source catalog. Open questions: whether to sample categorical values during discovery or defer to each investigation, and how to keep descriptions and join suggestions checked against live findings.
More from coding & agent
- Ethan Mollick's new book charts 30 months of AI progress, from cyborg to conjurer work — rseroter · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Jensen Huang details Vera Rubin agentic stack: NVL72, MGX, CMX and RTX — rohanpaul_ai · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23
- Capy 0.5.0 lets Claude Pro/Max subscribers reuse their plan quota, adds OIDC cloud access — garrytan · 2026-10-23