❯ git show --details cases/studio-agents.md
Internal Tools

Using the tools on my own work before recommending them to anyone else

A venture studio with a small team has the same back office as a bigger company: research, reporting, pipeline, outreach. I built the studio's internal workflows in Slack, Mattermost and Jira, and then a set of Claude Code skills and scheduled agents that do the repetitive parts in the house voice.

role
Built and operate them
timeline
2025 – present
users
The studio team
stack
Claude Code skills, hooks, scheduled agents, Slack, Mattermost, Jira, Python
outcome
Lead research, market reporting and outreach drafting run on a schedule

What runs

01
Lead research and enrichmentA skill that takes a target segment, builds and enriches a list, and scores it against the pipeline criteria for JournalRelay's CPA-firm channel. Output lands as a reviewed sheet, not a sent email.
02
Biweekly market reportA scheduled agent that snapshots a defined set of sources, writes the report in a fixed structure, and keeps the history so trends are visible.
03
Outreach drafts in the house voiceA skill with the studio's writing rules baked in: short, plain, one ask. Drafts only. A person sends.
04
Hooks and guardrailsHooks that stop an agent from doing the things we decided an agent should never do, such as sending, deleting or touching client files.

The rules

  • Drafts, not sends. Everything that leaves the studio is sent by a person. The agents remove the blank page, not the judgment.
  • Simplify before automating. Two of the first five workflows I set out to automate were removed instead. One was a report nobody read.
  • Measure hours and cost per run. I track both, so the next decision about a tool rests on what people actually used rather than on a demo.

What it taught me

Where these tools save hours is obvious within a week. Where they quietly cost more than they save takes longer to see: review time that nobody counted, outputs that are plausible and wrong, and workflows that only exist because the tool made them cheap. Having lived both sides is the most useful thing I bring to a conversation about AI adoption inside an operations team.

06.

connect()

Building something where a wrong number costs money?
Fintech, accounting, operations, applied AI. I am always open to talking with people who ship.

$ echo "Let's build something that survives an audit"