Stephens AI · our own tooling · Stephens AI · v2 live 2026-03-01

The AI operating system my own company runs on: 9 agents, 51 skills, 15 connected services.

I test everything on myself first. The same architecture I install for clients runs my own practice.

9agents each on a least-privilege tool list, plus 51 reusable skills and 15 connected services

07:30 every morning: new meeting recordings become pages, priorities, decisions, and a text with the action items

What I built

A searchable knowledge vault with retrieval search, and a local dashboard over the whole company: a 3D graph of the vault, calendar, Gmail, iMessage, Slack, and health data in one place, plus a chat over the vault that can read but never edit or send.
9 specialist sub-agents, each on a least-privilege tool list, and 51 reusable skills for proposals, research, QA, deploys, and reporting.
Scheduled jobs: a daily 07:30 meeting sweep that writes the meeting page, updates the week's priorities, logs decisions, and texts the action items; a weekly vault health check. Every run leaves an inspectable log.
Doctrine, written down: verify the output, augment rather than replace, understand the why, loop humans in, stay transparent. What stops earning its place gets retired.

What changed

By hand

After a client callNotes typed up later, or not

Finding a factGrep and memory

A repeated jobRe-explained each time

Sending anythingWhatever the model decided

The system

After a client callPage, priorities, decisions, text by 07:30

Finding a factRetrieval search over the vault

A repeated jobA skill with the gotchas encoded

Sending anythingA human approval step, always

Where the numbers come from. Counts are from the system's own registry on 2026-09-10. They change weekly.

For engineers

Architecture: A markdown vault with frontmatter as the database, a generator that rebuilds the index, sub-agents defined as files with tool allow-lists, skills as files with the gotchas at the top, a local dashboard server, and scheduled jobs on the Mac that deliver to iMessage.
Where it runs: On my Mac, with connectors into Google Workspace, Slack, Fathom, and the clients' tools.
Tests and evals: A weekly lint over the vault (contradictions, stale facts, orphans), a layout lint on every branded page, and a two-reviewer gate (adversarial fact-check plus a domain reviewer) before anything client-facing ships.
Cost and latency: One subscription plus about $10 a month of API credits.
Guardrails: Every agent runs on a least-privilege allow-list. Nothing is sent, deleted, or deployed without a fresh yes. Every scheduled job leaves a trace.
The tradeoff we chose: Why this way and not the other
Files over a database. A vault of markdown is slower to query than Postgres, but every agent, every human, and every future rebuild can read it, and the whole thing is in git.