AI products that have to work

For an investor deciding on an AI-native company, or an executive deciding whether an AI product ships. Two weeks, fixed fee, one review at a time.

The review answers four questions. Which of the product's claims are supported by evidence, and which are demos. What breaks under realistic usage: data, load, edge cases, the model behaving badly. What human work, vendor dependencies, and costs are hidden under the current numbers. What must change, at roughly what cost, on what timeline. I judge the system against the gates I run on my own product: schema and data validation, evaluation before release, tiered review, monitoring, and a person who owns each consequential call.

You get a written memo tied to your decision, an evidence appendix a technical reader can check, and a call with whoever decides. From $25,000, scoped in writing before we start. Half on start, half on delivery. If the first three days show the work does not need me, I say so and bill only those days. Under your NDA. I never carry one company's material into another company's room.

One system, one question, one week. The question is usually of the form: can this ship, or should we fund this, or why does this keep breaking. I read the code, the data flow, and whatever evaluations exist, run the checks that apply, and give you a written answer and a 60-minute call.

$5,000, paid on booking. If the question cannot be answered from what you can share, I say so by day two and refund in full. Commission a review within 30 days and the diagnostic fee is credited against it.

For a product or engineering team that wants an agent running behind verification gates by the end of the day, and still running on Monday. One day in your repo with up to twelve people: a prep call the week before, the day itself, and a written plan for the two weeks after. You leave with a tested workflow, evaluation results, accountable owners, and a release plan.

From $20,000. Half to hold the date, half on the day. The deposit is refundable up to 14 days out. On site in the San Francisco Bay Area or remote.

For a founder or product lead with something to ship this quarter. Ten hours at $250 an hour, used over two to six weeks: a two-hour teardown of the product, the data, the code, and the customers, then weekly working sessions on the real artifact, with review in your repo and docs between them, inside one business day. We work on whichever of these your product needs: the bet on one page; the system and the gates that decide when a release is ready; shipping with agents that are checked rather than trusted; data foundations; the team; and strategy under a clock, from pricing to what to tell the board. Live and async time bill the same, and you see the ledger.

$2,500, paid up front. Stop any time; unused hours are refunded. After ten hours, another block at the same rate or single sessions. I take two ongoing clients at a time.

One decision that needs one clear answer fast. Sixty minutes plus my written notes the same day. $600, paid on booking. Reschedule up to 24 hours before.

Both exist, by conversation, when the timing works on my side. Ask on the form.

An investor with a live decision on an AI-native company. An executive who has to decide whether an AI product ships, and what it will take. A founder, CEO, or head of product or engineering with real users inside the next quarter, seed to Series B or inside a larger company with a mandate and a budget.

It is not for a product that exists only as a deck, and not for anyone who wants the building done out of sight. I work in the room and in your repo. Your team keeps ownership.

Zoom or Google Meet, any time zone. In person in the San Francisco Bay Area when a whiteboard beats a screen.

Card, bank transfer, invoice, or purchase order. Pick it on the form. Prices in US dollars.

I sign your NDA. What you share stays with me.