Models

AI agency lock-in: what "you own everything" should mean

Daniel Kerambrum Bicourt · Founder, CO-ORDI · 27 October 2026 · 5 min read

Nearly every AI agency pitch deck has a line that says you own what they build. Almost none of them say what that means when you try to leave. A 2026 Expert Market survey of UK and US small business leaders found that among firms that deeply regretted switching an enterprise platform, high upfront switching costs were the single most common reason they felt stuck, cited well ahead of simple contract lock-ins. AI builds are more exposed to this than most software, because the thing you are buying is not a licence. It is a system trained on your process, and that makes it easy to make leaving expensive without ever writing a lock-in clause.

"You own everything" has to survive four tests

The phrase is marketing until it is tested against what happens on exit. Four things have to be true, in writing, before it means anything.

• IP assignment, present tense. Not "will assign on final payment", not "licensed for your use". The contract should say the work product is yours the moment it is created, full stop.
• Accounts in your name. Every tool, API key and login the system depends on should sit in your organisation's accounts, not the agency's. If their departure deletes your access, you never had ownership, you had tenancy.
• Documentation that survives the builder leaving. A system only you can read is a system only they can maintain. Ask for the logic, the data flow and the failure points written down in plain language, not just code comments.
• No proprietary runtime underneath it. If the automation only runs inside the agency's own platform, you do not own a system, you own a seat. The moment the subscription lapses, so does everything built on it.

Where lock-in actually hides

Lock-in rarely shows up as a clause that says "you may not leave". It shows up as a design decision made early, before anyone is thinking about an exit.

Platform dependency. The automation is built inside the agency's own no-code tool or managed app, rather than in standard infrastructure you can host anywhere. Ask whether the system would still run tomorrow if the agency vanished overnight. If the honest answer is no, you are renting.

Prompt and configuration lock-in. With AI systems specifically, the business logic often lives inside prompts, agent configurations and fine-tuning held in a vendor's console rather than in a file you can export. A model trained or tuned on your data rarely reproduces the same results when moved to a different provider, which makes switching feel riskier than it should.

Data lock-in. Export formats that are technically possible but practically painful: a CSV dump with none of the relationships that made the data useful, or a database schema nobody outside the agency understands.

Knowledge lock-in. The quietest one. The build works, the documentation exists on paper, but the one person who actually understands how the pieces connect left the agency eighteen months ago. This is the same failure mode we cover in why AI pilots die before production: a system nobody owns past launch decays whether or not the contract says you own it.

What it costs when it bites

Nobody budgets for an exit at the start of a build, which is exactly why lock-in is cheap to create and expensive to discover. The same Expert Market research found that most small businesses who switch a core platform and regret it point to the same handful of causes: the upfront cost of moving, a system that was judged "good enough" to avoid dealing with, and the fear of downtime during the switch itself. None of those are about the new vendor being worse. They are about the first build never being made portable. That cost does not show up on an invoice. It shows up as a renewal you accept because untangling the system looks harder than paying again.

Four questions before you sign

Ask these before any AI build starts, not when you are trying to leave.

• Which accounts will this run in, mine or yours, and can I see the credentials today?
• If I asked for everything tomorrow, source, configuration and documentation, what would I actually receive?
• Does this system run on infrastructure I could host myself, or only inside your platform?
• Who, specifically, understands this system well enough to hand it to someone else?

A provider with nothing to hide answers all four without a pause. We cover the wider version of this screening in how to vet an AI expert, and the comparable test for choosing a provider at all in best AI automation agencies UK.

How we handle it

Every system your embedded AI Architect builds runs in your accounts, documented as it ships, with IP assigned to you from day one, not on final payment. If you walk away after month one, under the CO-ORDI Guarantee, you keep everything already built. There is no platform of ours standing between you and your own systems, because there is no platform of ours to begin with. The fee buys outcomes, not a seat on something you would need us to keep running.