Who Owns Your AI Agent? Ownership, Accounts, and Handover Terms to Settle Before You Hire

You should own the prompts, workflows, custom code, data and every account the agent runs on, or have a written licence that lets you keep using them if the agency disappears. Settle this before signing: put accounts in your name, define the IP clause, and require documentation and an exit plan.

Key takeaways: Create the accounts yourself and give the agency access, not the other way round. Get ownership, documentation and exit terms in writing before work starts, because they are hard to negotiate once you depend on the agent.

Who actually owns what an agency builds for you?

It depends entirely on what the contract says, and many agreements say very little. In many places the default for commissioned work is that the creator keeps the copyright unless there is a written assignment or licence, which means paying an invoice does not automatically make you the owner. An AI agent is also not one thing. It is a bundle of prompts, workflow configurations, custom code, connections to your tools, a knowledge base and accounts with third-party providers, and each piece can have a different owner. So ask for the answer piece by piece rather than accepting a general sentence like "the client owns the deliverables."

A second complication is that some parts were never the agency's to give. The underlying model, the chatbot platform and any open-source library keep their own licences. Model providers generally address who owns outputs in their own terms, so read them directly, for example the Anthropic commercial terms and the OpenAI terms of use. What you can realistically ask for is ownership of what was made specifically for you, plus a clear right to keep using everything else.

Which accounts should be in your name from day one?

Every account that holds your customers, your data or your billing should be created by you, under a company email you control, with the agency invited as an admin or collaborator. The reason is practical: if the agency creates an account under its own name, you cannot move it, cancel it or prove it is yours without their cooperation. This is the most common lock-in in small projects, and it rarely looks like lock-in at the start. It looks like a helpful agency setting everything up for you to save time.

• Model provider API accounts (OpenAI, Anthropic, Google): your organisation, your billing, API keys you can rotate.
• Messaging channels: your WhatsApp Business account and Meta business portfolio, and your Instagram and website accounts. See the WhatsApp Business Platform documentation for how accounts and numbers are tied together.
• Your own systems: CRM (such as HubSpot), Shopify, Slack, Notion and calendars, with the agency added as a user you can remove.
• Automation platforms such as Zapier: a workspace owned by your company, not a personal one.
• Hosting, domain and code repository: registered to your business, with the agency as a contributor.
• A password manager vault or equivalent where credentials live, shared with the agency and revocable.

For the messaging side in particular, our guide on what an agency sets up for WhatsApp Business API and what you need to provide goes through who creates what.

What should the IP clause say about prompts, workflows and code?

The clause should name the asset types explicitly and say who owns each one on final payment. "Deliverables" is too vague, because nobody will agree later on whether a prompt library counts. A fair agreement usually gives you ownership or a perpetual licence over the work made for you, while the agency keeps its general tools, templates and know-how that it reuses across clients. That split is normal and reasonable; the thing to avoid is a contract where reusable agency material is mixed up with your business content.

Also check what happens to data you supply. If you upload policies, price lists or customer conversations, the contract should say they are confidential and are not used for other clients. Our piece on training a chatbot on your own business data explains what typically goes into the knowledge base, which helps you judge what needs protecting.

What documentation should you require at handover?

You should require enough documentation that a different competent person could run and change the agent without calling the original builder. Think of it as the test: if the agency vanished next week, could a freelancer take over in days rather than start from zero? Writing this into the agreement as a deliverable, tied to final payment, is far easier than asking for it politely at the end. Agencies that work well already produce most of this as they go.

1. A system diagram showing the channels, the model, the tools it can call and where data is stored.
2. The full text of prompts and instructions, with a short note on why key rules exist.
3. A list of every account, integration and API key in use, who owns each and where billing goes.
4. The workflow or code exported or committed to a repository you control, with setup notes.
5. The test cases used to check quality and the results at launch, so you can spot regressions. For ideas on what to measure, see evaluating AI agents beyond simple accuracy.
6. A runbook: what to do when the agent gives a wrong answer, when a key expires, when a channel is blocked.
7. A walkthrough session with you or your team, ideally recorded.

Which lock-in terms and red flags should make you pause?

The red flags are usually not dramatic; they are small clauses that add up to dependence. An agency that is confident in its work does not need to trap you, so a refusal to discuss these points is information in itself. Raise them early, in the first conversations, and notice whether the answers are specific or evasive. If you are still comparing providers, the questions to ask before you hire an agency sit alongside this checklist.

• The agent runs only on the agency's own platform, with no export or self-hosting option.
• Accounts, API keys or numbers are registered to the agency.
• No written statement of who owns prompts, workflows and code.
• Exit fees, long minimum terms or automatic renewals with short cancellation windows.
• Monthly fees that mix hosting, support and licence with no breakdown.
• Documentation described as "available on request" rather than a named deliverable.
• Your data may be reused for other clients or for training, with no opt-out.

What if the agent runs on the agency's platform or a SaaS tool?

Running on a platform is not wrong, but you need to know exactly what you are renting. A no-code tool or a vendor platform can be a sensible choice for a small business, as we compare in custom agent versus no-code tools versus SaaS. The question is portability: can you export the conversation flows, prompts and knowledge base, and can you move the account into your own name? If the answer is "the platform keeps it," you have a subscription, not an asset, and the contract and pricing should reflect that.

With an agency's own platform, ask what happens if they close, are acquired or stop supporting a feature. A reasonable arrangement might be a licence that survives termination for a set period, escrow of the code, or a documented export in standard formats. Not every small agency will offer all of these, and that is fine, but they should tell you plainly which ones they do offer. Recheck the same question for each paid tool in the stack, since your agent depends on all of them.

What exit terms belong in the agreement?

An exit clause should say how you leave, what you get on the way out and how long the agency keeps helping. Cover notice periods, a handover of all credentials within a stated number of days, deletion or return of your data with written confirmation, and a short transition-support period at an agreed rate. Also decide who pays ongoing costs such as model usage and messaging fees, since those are billed by the providers to whichever account owns them. Costs and scope are covered in what drives the cost of a custom AI agent project, and the phases in what a custom AI agent project involves help you place handover milestones where they belong.

Finally, remember that this article is a practical checklist, not legal advice. Have a lawyer in your jurisdiction review the final contract, especially the IP assignment and data-processing terms, because the rules differ by country. At Kuan AI we integrate with the tools a business already uses, and the specifics of ownership and handover are agreed per project, so you can raise these points with us through the contact form and get a straight answer before you commit to anything.

If you only do three things before signing, create the accounts in your own name, name each asset in the IP clause, and make documentation a paid-for deliverable. Those three steps remove most of the risk of being stuck with an agent you cannot move, change or fully understand.