I only sell fixed-price builds and a monthly retainer, never hourly project billing, so I have a position here. But the deciding question is not which model a vendor prefers. It is how well you already understand the problem, and who carries the risk of the parts nobody knows yet.
Why AI builds make this choice sharper
Most of the uncertainty in an AI project sits in the data, not the code. Data preparation is usually 30–50% of an assistant build: scanned files, three versions of a policy, nothing marked current. A prototype tested on five clean documents looks finished. Then it meets the real documents.
Under hourly billing, you pay for that discovery as it happens, with no ceiling. Under a fixed price, the vendor has to find it during scoping, which is why a serious fixed quote asks for a sample of your real data before giving a number. A vendor who gives a fixed price without seeing any data is either padding it heavily or planning to argue about scope later.
How hourly goes wrong
- The bill grows with every surprise in the data, and AI projects have many.
- Nobody is paid to say "this part isn't worth building", because cutting scope cuts revenue.
- Handover becomes a favour asked for at the end rather than a deliverable with a line item.
How fixed price goes wrong
- The scope is vague, so every disagreement becomes a change-order argument.
- The vendor protects the margin by cutting the work you can't see: evaluations, error handling, documentation.
- "Done" is never defined, so acceptance becomes a matter of opinion.
Every one of those is a contract problem, and every one has a fix you can write down before work starts.
What a fair fixed-price AI contract includes
- The one workflow included, and what is explicitly excluded. Exclusions prevent more disputes than inclusions do.
- Acceptance tests. For an assistant, twenty real questions with verified answers that the system must pass. See why AI projects fail for how to collect them in an afternoon.
- A payment schedule tied to delivery. Mine is half on signature and half on handover.
- An acceptance window and a warranty. I use a 7-day acceptance window and a 30-day warranty written into the statement of work.
- A written update every week, with working software to look at, not a status report.
- Ownership transfer. 100% of the code and IP, documentation and a runbook. The AI software handover checklist lists what to collect.
- Running costs named separately. Model usage, hosting and third-party services are not part of the build fee unless the quote says so.
When paying for time is the right call
Some work genuinely has no fixed end: a product that ships continuously, a team that needs senior review and architecture help every month. Forcing that into fixed-price projects creates a new negotiation every few weeks. For that, I offer a Fractional AI & Backend Engineer retainer: dedicated engineering days each month, direct access, and you can pause or cancel whenever. What I don't do is open-ended hourly billing on a project that should have a defined end.
And if the scope genuinely can't be defined yet, neither model saves you. Whatever you buy while the problem is fuzzy, you will pay for twice. Freelance AI developer vs agency vs in-house walks through the same question from the hiring side.
How Kortex Labs prices a build
A free 30-minute scoping call, then a fixed scope and a fixed price within 24 hours, or an honest answer that it isn't worth building. Starting prices are public: a data and automation pipeline from $6,000, a RAG chatbot from $8,000 and an AI product MVP from $12,000. The AI development cost calculator shows how the number moves with your data, users, integrations and risk, and how much it costs to build an AI app explains each line.
Frequently asked questions
Is fixed price or hourly better for AI development?
Fixed price is better when you can define what working means specifically enough to test it, because the vendor carries the risk of surprises in the data. Paying for time suits ongoing work with no fixed end. If the scope cannot be defined yet, neither model helps; get clear first.
Why are fixed-price quotes sometimes higher than hourly estimates?
A fixed price includes the risk of unknowns, especially in the data, that an hourly estimate leaves to you. Compare the fixed price against what the hourly estimate would become if the data turns out messier than expected.
What should a fixed-price AI contract include?
The included workflow and explicit exclusions, acceptance tests, a payment schedule tied to delivery, an acceptance window and warranty, weekly written updates, transfer of all code and IP with documentation, and running costs listed separately.
Does Kortex Labs bill hourly?
No. Builds are fixed scope and fixed price, quoted within 24 hours of a free scoping call. Ongoing work is available as a monthly Fractional AI and Backend Engineer retainer that you can pause or cancel.