You buy an outcome, not my hours.
Most AI consulting is sold by the hour or the day. That arrangement pays more when work takes longer, which is a strange thing to agree to when you are hiring for speed and judgement.
Why hourly is the wrong instrument here
Hourly billing makes sense when the buyer can supervise the work closely enough to judge whether the hours were well spent. In AI engineering that is rarely true — the client is buying the expertise precisely because they cannot yet evaluate the decisions in real time.
It also misprices the thing that matters. Building AI-native means a well-scoped system can land far faster than a traditional estimate assumes. Under hourly billing that speed is a pay cut; under fixed price it is the whole proposition. The incentive should reward getting it right quickly, not sitting with it.
The honest objection
Fixed price moves overrun risk onto the supplier, and suppliers respond in one of two ways: pad the number, or defend the scope aggressively. Both are bad for the client, and pretending otherwise would be dishonest.
The mechanism that makes it workable is the written scope. Acceptance criteria fix what “done” means before either party is invested, which removes the argument that fixed-price engagements usually end in. And where something is genuinely unknown, it gets named as unknown and handled separately rather than priced as though it were understood. A fixed price on a vague scope is not a commitment; it is a dispute with a date on it.
Engagement sequence
- 1
Architecture review
A paid, short, self-contained engagement. I read the system, the constraints and the failure history, then write up what I would build and what it would take. It stands alone — if we never work together again, you keep a document you can hand to anyone.
- 2
Technical discovery
The specifics that decide effort: data shape, tenancy model, latency tolerance, what already exists, and which parts are genuinely uncertain. Uncertainty is identified here rather than absorbed silently into a number.
- 3
Written scope
A concrete list of deliverables. Each carries acceptance criteria — the conditions that decide whether it is done. This is the document that makes fixed price possible, and it is written before anyone commits.
- 4
Fixed-price proposal
One number for that scope. Not an estimate, not a range that drifts upward. If discovery revealed something genuinely unknowable, it is carved out and named rather than buried in contingency.
- 5
Build and accept
Work proceeds against the document. You accept each deliverable against its written criteria. Changes of mind are welcome and are handled as scope changes with their own number, not as an argument about what was implied.
Not a retainer. Not open-ended.
This is not staff augmentation and it is not a monthly seat you rent. Each engagement has a defined end, and the end is defined by the document rather than by a calendar. If what you actually need is a long-term embedded engineer, I will tell you that rather than restructure it into something it is not.
Start with the review.
Send the problem — the system, the constraint, what already broke. See the four disciplines for what the work usually involves.