The short answer
Fixed-scope engagements profit by finishing; time-and-materials engagements profit by continuing. Both are legitimate for different work, but for buildable systems the incentive tell decides outcomes: a vendor paid to finish scopes hard, ships, and leaves, while a vendor paid by the hour has no structural reason for "done" to arrive. AI uncertainty does not break fixed-scope; phasing absorbs it.
Pricing models read like accounting detail and behave like destiny. Before any code exists, the pricing model has already decided what the vendor is optimizing for, which makes it the single fastest tell in an engagement letter.
The incentive tell
Strip the terminology: fixed-scope sells an outcome, time-and-materials sells hours. Under fixed scope, every inefficiency is the vendor's problem, so scoping gets sharp, seniors do the work (juniors cost the vendor rework), and finishing is the payday. Under T&M, every inefficiency is your problem, politely: discovery expands, meetings multiply, and the definition of done grows a committee. Neither behavior requires bad faith. Incentives do the work that intentions claim to.
The comparison
| Fixed scope | Time and materials | |
|---|---|---|
| What is sold | A defined outcome | Hours and attention |
| Who absorbs inefficiency | The vendor | The buyer |
| Vendor payday | Finishing | Continuing |
| Natural team | Small and senior | As many as the budget bears |
| Scope discipline | Forced, up front | Optional, forever |
| Honest use | Buildable systems, phased | Unknowable scope, embedded staffing |
When time-and-materials is honest
- –Genuinely unknowable scope: research-shaped work where nobody can define done without doing.
- –Ongoing embedded staffing: when the relationship is the product, as with fractional arrangements, a transparent retainer is the honest shape.
- –Post-handover option work: small enhancements to a system you already own, metered openly.
The dishonest use is the common one: T&M for a buildable system, justified by AI's uncertainty, billed until the budget rather than the work runs out. It is the pricing model behind the "long-term partnership" pitch in the red flags list.
Incentives do the work that intentions claim to. Read the pricing model as the vendor's true roadmap.
Fixed scope under AI uncertainty
The standard objection is that AI work is too uncertain to fix. The answer is not bigger contingency; it is smaller scopes. Phase the uncertainty: a fixed-price Embedded Discovery answers what should be built (including "nothing"), a fixed-price production build builds it, and hardening gets scoped when reality demands it. Uncertainty lives between phases, where you can stop cleanly, instead of inside an open meter you can only watch. Each phase's price arrives after the workflow has been seen, which is what keeps fixed from meaning guessed.
The clauses that matter
Fixed-scope only protects you if the scope is an outcome. Check that deliverables are systems, not activities: named users in production, evals passing thresholds, the six-part handover, never "four development sprints." Check the change mechanism: real projects change, and a sane clause prices changes at the same discipline as the original scope. And check what happens at each phase boundary, because the right to stop is the buyer's half of the incentive bargain. Vendors who price this way are betting their margin on their own competence. That is the bet you want to be on the other side of, and it is the model the mid-market tier runs when it is being honest.