Custom agent · Built in-house
Replace a CRM with plain files an agent keeps up to date
The partner of a two-person venture fund replaced the fund's CRM with a git repository of plain text files that an AI agent fills in from call transcripts, emails and decks, with Claude-written daily summaries for every active deal.
What they did
Every person, company and deal is a folder with a structured record file plus notes, emails and document text. After a 45-minute intro call, the transcript, email thread and three PDFs went to the agent, which checked for existing matches, created six linked records, logged notes, extracted the PDFs and committed everything in one go. A rules file of about 600 lines sets naming, allowed statuses and matching rules, and anything destructive or agent-initiated must go through a pull request for a human to approve.
How it went
The author says the whole intake took about 90 seconds versus 20 to 30 minutes of manual CRM entry, that the system runs in production for every deal and meeting at the fund, and that they no longer use the CRM at all.
Worth knowing
The author is clear it only works for a small team comfortable in a terminal, with no dashboards, and that the rules file took many iterations after the agent made duplicates and wrong statuses. The post does not name the agent product used for intake.
Try it with Custom agent
Set up a folder in <your notes or repo location> where each contact, company and deal is a subfolder with a short structured record and a notes file. When I give you <a call transcript, email thread or deck>, first search existing records by email, domain and name; if only one signal matches, stop and ask me. Then create or update the records, link them, and add a dated note with next steps. Never delete or merge records without asking.
We wrote this prompt from the case. Replace the parts in angle brackets with your own details.
What this label means: User reports
Someone using it at work posted about it. Any numbers are theirs, so we write "the poster says".
Real cases of AI agents at work, from every major agent, each linked to where it was first reported. Rewritten in our own words, most with a prompt you can try. These are not Wize customers.