Seventeen AI agents and one rule: a human approves anything risky.

An AI company with six departments that writes code, drafts marketing and does research, while the founder talks to just one agent.

The situation

Running two businesses meant the back office never ended. We wanted AI that could take whole tasks off the list without being trusted blindly.

One orchestrator that takes requests and routes them to six department heads (development, marketing, finance, design, research, operations), each with specialist workers.
Three approval tiers. Reversible work runs on its own; a fixed list of risky actions always waits for a human; everything else escalates up the chain.
An outcome judge that checks every result against the original request before a task counts as done.
A native macOS cockpit with a live agent tree, approvals queue, chat and per-agent usage. Rex built the first version of its own cockpit.

The same approval-first design goes into every AI agent we build for clients.

A sample mission. The real cockpit runs on Shah’s Mac.

Rex

MissionsAgentsApprovals1ChatVaultMemoryUsage

Mission (sample)

Draft the October newsletter and schedule the posts

RexOrchestrator

Dev leadMarketing leadCFODesign leadResearch leadOps lead

Content writerMarketingCampaign plannerMarketing

17 agents · 6 departmentsSample data

Approvals

Needs your yesMarketing lead wants to send the newsletter to the subscriber list.ApproveDecline

Always needs a human:sending to customers, moving money, deleting data

Swipe to see the whole screen.

Free 30-minute growth call. I’ll tell you what I’d build first, what it would take and whether we’re the right team for it.

Toronto

You’ll talk to Shah, the founder. After hours, leave a message and Shah calls you back the next business day.