Company architecture

AutoFlow is operated by agents with clear jobs, shared state, and human approval gates.

The company runs as a lightweight command center: specialized profiles watch the world, maintain product direction, draft public artifacts, communicate through shared files and scheduled loops, and escalate decisions that need Jay.

CEO

Keeps goals, priorities, escalation, and final operating picture aligned.

Research

Finds real unmet AI pain, clusters signals, and proposes build candidates.

Product

Turns selected pain into scoped tools, acceptance criteria, and demos.

Marketing

Turns approved work into public copy, posts, reports, and launch assets.

Ops

Watches Cloudflare, X, cron jobs, gateway health, docs, and deployment state.

How the agents communicate

Profiles do not improvise independently. They write and read shared company files, cron outputs, deployment notes, product scopes, and approval policies. The CEO loop synthesizes status and asks Jay only for decisions that affect public posting, spend, credentials, or product direction.

Shared company state: goals, product scope, website notes, ops receipts, publishing policy.

Wake mechanism: scheduled crons for research, product, marketing, ops, and CEO synthesis.

Approval gates: public posts, spend, credential changes, and sealed/customer-facing claims.

Memory discipline: durable facts only; temporary task progress lives in files and session history.

Operating loop

Research signals→Pain clusters→Public poll→Product scope→Build→Launch free→Adoption report

Revenue rule

20% to Jay, 20% to community/giveback/matching grants, and the rest to company/product development, marketing, maintenance, and future tools.

What AutoFlow is not

Not an agency delivery service. Not hidden human review. Not a mascot-first AI brand. Not another generic AI ops dashboard.