How we work
Prove it works before you pay for the build
Five steps. Each one ends with something you can check, and a price you knew before it started.
How an engagement runs
-
Fit call
20 minutes, free.
- What you get
- A straight answer on whether AI fits the problem, and whether I'm the right person for it.
- What it costs
- Free.
-
Audit
Two weeks, fixed price, credited against the build.
- What you get
- A map of your systems and data, ranked candidate workflows with baseline numbers, pass/fail criteria for a pilot, cost per run, and a fixed build quote.
- What it costs
- $6,500. Credited in full against a Build Sprint signed within 60 days of delivery.
-
Pilot
Two weeks on your real data, pass/fail written down first.
- What you get
- A working first version on your own data, measured against criteria we agreed before starting.
- What it costs
- Phase 1 of the Build Sprint. If it fails, you pay for phase 1 only and keep everything built.
-
Build
Fixed price, change control.
- What you get
- The production build of one workflow or one MCP server, with its evals and cost per run.
- What it costs
- $18,000–$45,000, fixed, including the pilot.
-
Run
Month to month.
- What you get
- Eval regression checks, cost monitoring, model migrations, small changes and on-call.
- What it costs
- $1,500–$3,000 a month. Month to month. Cancel with 30 days' notice.
Principles
-
Start from your data layer
Not from a chatbot. The work begins where your data lives.
-
Prove it before you pay for the build
The pilot has pass/fail criteria written down before it starts.
-
You own everything
The API accounts, the prompts and the code are yours.
-
Nothing resold
Model usage is billed to your accounts. I make no margin on it.
-
One engineer, no hand-offs
The person on the fit call is the person who builds it.
What I will not promise
Anyone who promises these is guessing.
- That a model never makes a mistake.
- That AI replaces a role.
- A return before we have a baseline.
- That a demo is production.
- A timeline before the audit.
Contract terms in plain language
The contract says the same things, in longer words.
- The work is done to a professional standard. That is the warranty.
- No guarantee of model output or of a business result.
- You own the API accounts, the prompts and the code.
- A human reviews consequential decisions.
- Liability is capped at twelve months of fees.
- AI use is disclosed in deliverables.
Questions
What happens in the fit call?
Twenty minutes. What you run, what hurts, and whether I’m the right person. If I’m not, I’ll say who might be.
What do I get from the audit?
A map of your systems and data, three candidate workflows ranked by hours saved and error cost, baseline numbers, pass/fail criteria for a pilot, cost per run, and a fixed quote. Two weeks, $6,500, credited in full against a build signed within 60 days.
What if the pilot fails?
You pay for phase 1 only and keep what was built. That is the point of writing the criteria down first.
Who owns the code, the prompts and the API accounts?
You do. Accounts are opened in your name; the code and prompts are yours on payment.
Is my data used to train models?
No-training terms are configured on the accounts you own and confirmed in writing for each engagement. Which vendor and configuration we use is set out in the audit.
Which models and vendors do you use?
Whichever fit the job; models are swappable by design. I don’t resell any of them.
Do you work with health data?
Not yet. Work that touches protected health information is out of scope until the controls for it are in place.
How long does a build take?
Four to eight weeks for one workflow or one MCP server, including the two-week pilot.
What don't you do?
Chatbots for their own sake, subscription receptionists, anything I can’t measure, and work that needs a team I don’t have.
Where are you?
Boston and New York; most work is remote.
Start with a 20-minute fit call
Free. What you run, what hurts, and whether I'm the right person. If I'm not, I'll say who might be.