FullStack VC

Archived · LP Update Bot WG (archived)

LP Update Bot

An early experiment in agent-drafted LP updates — archived after we concluded the bottleneck wasn't drafting, it was approval.

All projects

Why we archived

Six months in we had a working drafting loop — agents could ingest the portfolio data, produce a credible LP update draft, and route it for review. The reviews never got faster.

The real cost of an LP update isn't the drafting — it's the partner-level decisions about what to disclose, what to soften, and what to celebrate. Agents don't reduce that cost. They just make the unedited draft show up faster.

What we kept

  • The data-ingestion side (portfolio metrics → structured artifacts) graduated into MemoPop AI.

  • The "which sentences need partner judgment" classifier graduated into Context Vigilance as a generic review-boundary detector.

What to read if you're tempted to try this again

  • The pre-archive retro doc (in repo).

  • The MemoPop AI spec — different angle on a similar surface.

Lesson

If the bottleneck is approval, automating drafting doesn't help. Find the bottleneck first.