AI Subcontracting for Agencies: How Delivery Works and What the Contract Should Say

An agency that sells AI work but does not build it has a delivery gap. Subcontracting closes that gap. The agency keeps the client, the scope and the relationship, while a specialist team builds the agent or automation under a written agreement. The arrangement works well only when both sides agree on how the work is handed over.

Key takeaways

• In a subcontract, the agency is the client of the build partner, and the agency answers to its own client.
• The scope, the handover date and the acceptance test should be written down before work starts.
• Payment should be tied to milestones, and work should not start until the first payment arrives.
• Liability should be capped and tied to what has been paid for the project.

How delivery normally runs

A typical project has four stages. The agency and the build partner agree the scope. The partner designs the workflow and builds a first version. Both sides test it with real examples from the client's work. The agency then hands the finished system to the client, with documentation and a walkthrough.

Each stage needs a named person on each side. Without that, questions sit in a chat thread for days and the client starts to doubt the result.

Clauses that matter in an AI subcontract

• Scope: list the workflows, the systems they read from and write to, and what is excluded.
• Acceptance: define how the client signs off, such as a set of test cases that must pass.
• Payment: give the amounts, the due dates, and what happens if a payment is late.
• Intellectual property: state when ownership of the deliverables passes to the agency and then to the client.
• Confidentiality: cover client data, including any data sent to AI models.
• Liability: cap liability and exclude lost profits, which is the usual position for software and automation work.
• Termination: set a notice period and say how work done so far is paid.

Why client data needs its own rule

AI work often touches customer emails, invoices or internal records. The subcontract should say which data the build partner may access, where it is stored, which AI model providers receive it, and how it is deleted at the end. This matters more than most agencies expect, especially when clients operate in regulated markets.

Handover is part of the job

The handover should include the workflow documentation, the list of connected systems, the approval steps, the known limits, and the contact for support. A client who understands how to switch off or review the agent is far less likely to blame the agency when something unusual happens.

If you want a partner to handle the build under this kind of arrangement, the details are on the become a partner page. The guide to choosing an AI agency lists the questions to ask any build partner before you sign.

Frequently asked questions

Who should sign the client contract in a subcontract?

The agency usually signs it, because the agency owns the client relationship. The build partner signs a subcontract with the agency.

Can the subcontractor talk to the client directly?

Only if the agency agrees in writing. Most agencies prefer to control client communication, at least until the handover is complete.

What if the client changes the scope mid-project?

Treat every change as a written change request with a price and a new date. Do not let changes slip in through informal messages.

Do I need a lawyer for this?

For anything beyond a small project, yes. A lawyer who knows software and data rules in your country and your client's country should review the contract.