I build AI that home services owners trust to run their day.

What runs every day

I built Titan, an AI operating system for home services companies, and ran it across 120+ locations in two franchise systems. In my own two-location company, an AI operations agent now works next to the GMs on top of it, reading our field-service software, phone lines, shared inbox and crew WhatsApp groups. A day looks like this:

• 3am, data check. Field-service data is reconciled overnight, so the morning numbers are right.
• 5am, the numbers. Every manager and salesperson gets yesterday's results against target. Field labor went from 50% to 30% of revenue once managers saw dollars per van every morning.
• 8:15, closeout. Were yesterday's jobs closed out and invoiced? The agent reads job notes and messages first, so it only nudges when something is actually missing.
• 8:30, unpaid invoices. Follow-up texts go out, except to anyone with a complaint, a dispute, a cash payment or a recent text. Nearly 100% of invoices get collected, and about $20K of bad debt was recovered.
• 9am and 4pm, GM follow-through. Before it flags a slow reply, the agent checks whether the work already got done.
• During the day, dispatch. The GM books new jobs next to existing work and pulls future jobs into slots that open up.
• 6pm, what's left. Every visit from today that still needs invoicing or payment.
• Any time someone applies. They get a text within minutes, a short screening conversation and an interview booked with the GM as soon as possible. Cost per hire fell from about $1,000 to $100 to $150.
• Mondays, the week ahead. The weekly report, plus appointment confirmations. One Monday the agent found 95 visits and sent 27 confirmations after a single approval.

Messages to customers go out after a person approves them, on the same software, phone lines and group chats the team already uses. See a full day in the case study or watch the 6-minute walkthrough.

How I build

I built Titan by running AI coding agents in parallel, 80+ at a time, and shipped 975 changes, about the output of a seven-person engineering team. Every change goes around the same loop, and the loop runs as devflow, a tool I built that I now use on every product.

1. 1 · InFeedback and requests
   • A feedback bubble on every page
   • Thumbs up or down on every agent reply
   • A roadmap people vote on
2. 2 · PlanA written plan
   • One for every change
   • AI reviewers on the big ones
   • I approve it
3. 3 · BuildAgents in parallel
   • 80+ at a time
   • Each in its own copy of the code
   • They email me when stuck
4. 4 · CheckGates
   • Four AI reviewers
   • 21 automated checks
   • A live preview of every change
   • My approval, then a merge queue
5. 5 · ShipDeploy
   • Tested after every deploy
   • Rolls itself back if it breaks
   • The people who asked get an email
6. 6 · HealOvernight
   • Data checked and repaired
   • Errors scanned
   • A fix agent has a tested fix ready by morning

Client work gets the same setup. Agents only write where that's been turned on, approvals depend on risk and role, spending is capped, and there's an off switch.

How I work

Every engagement starts with the 30-minute call below.

Track record

Bring me one problem

I'm an open book. Bring one problem to a 30-minute call, like a report you don't trust, a tool you're tired of maintaining, or the thing that eats up your day.

We'll work it out together on the call. Afterwards you get a write-up of the whole thing, so you can run with it yourself or have me build it.