I build AI agents that do real work for operating businesses.
An agent that reads your operation every night, investigates what looks off, and proposes what to do about it — then stops. It never acts alone. A human always holds the pen: nothing happens without your sign-off.
How it works
Pulls in what your business already writes down — counts, orders, logs, notes.
Flags what's worth a closer look. Most nights, most things are fine.
Digs into the flagged threads with real tools — history, rates, cross-checks.
Writes up what it found and what it would do, with a confidence score.
A human reads the proposals over coffee. Approve or dismiss. Always.
Only approved changes happen — and every one of them is on the record.
// runs nightly · halts at a human · nothing ships without sign-off
Proof
This isn’t hypothetical. Every night, an agent I built clocks in at a bar & grill in Bigfork, Montana. It runs a three-phase pipeline — triage the day’s operational data, dig into the flagged threads with investigation tools, then synthesize insights with confidence scores and proposed actions. In the morning, a manager approves or dismisses each proposal, and every decision lands in an audit trail. It’s been in production since 2025, part of a platform the kitchen and bar staff use every day.
Engagements
I spend two weeks inside your operation: what your team tracks, where the hours leak, what an agent could genuinely take over. You get a written roadmap with the honest version — what's worth automating, what isn't, and what I'd build first. The roadmap is yours to keep, whoever ends up building it.
I build the first agent from the roadmap — reading your real data, proposing real actions, human-gated from day one. Fixed scope agreed up front, working software in production at the end, and your team trained to supervise it.
Operations drift, and agents should drift with them. A monthly engagement to tune thresholds, extend the agent's tools, and review what it's been proposing. Entirely optional, cancel anytime — the code is yours either way.
Questions you should ask
What if it makes a mistake?
On its own, it can't make one that costs you anything. The agent's job ends at a proposal — every action waits for a human to approve it, and every approval and dismissal lands in an audit trail. A bad night for the agent means a suggestion you dismiss with one click, not an order you have to unwind.
Do we have to change our systems?
No. I build around what you have. If your operation runs on spreadsheets, a point-of-sale export, and a group chat, that's what the agent reads. Asking your team to change how they work is a cost — and the whole point here is to remove work, not add it.
Who owns the code?
You do. Full stop. Everything I build for you lives in your accounts and stays yours when the engagement ends — no licenses, no lock-in, no pieces I keep.
How long does it take?
The audit is two weeks. A first agent build is scoped before we start, so we both know exactly what's being built and when it lands — weeks, not quarters. And it runs human-gated from its first night.
What does it cost?
A fixed fee per phase, agreed before any work starts. No hourly billing, no open-ended retainers, no surprises on the invoice. The audit is the inexpensive way to find out whether a bigger build is worth it — and you keep the roadmap either way.
Why one engagement at a time?
Because small operations deserve full attention. When I'm working on your business I'm not splitting the week across four clients — I learn your operation properly, and you always know what I did this week. The trade-off is a wait list; the availability date below is real.
Contact
If you run a real operation and wonder what an agent could quietly take off your plate, the cheapest way to find out is a plain email.
Email me — the worst I’ll say is “not a fit”