Who owns the AI platform after launch?

The owner of an AI platform after launch is one named person inside the company who decides what the agents may do, approves changes to the rules they follow, and answers for the results. Hosting and running the software can sit with a partner. The decisions cannot, so name that person before go-live.

Why this question comes up late

Most AI projects plan the build in detail and the day after launch in one line. Then the first exception arrives: a customer with a special price, a supplier who changed terms, an order the agent should not have sent. Someone has to decide what the agent does next time. If nobody was named, the agent keeps doing the old thing, or people quietly stop using it. The steps up to go-live are in how to deploy AI agents in an enterprise and how long AI implementation takes. This note is about the day after.

Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027 "due to escalating costs, unclear business value or inadequate risk controls" (Gartner, June 2025). In our reading, two of those three reasons are ownership problems. Value stays unclear when nobody is measuring it, and risk controls stay weak when nobody approves them. Gartner makes the same point about accountability: agents "are not legal entities and cannot be held responsible for the consequences of those actions", so responsibility stays with people (Gartner, AI agent reality check).

What the owner actually decides

Ownership after launch is not about servers. It is about four kinds of decision:

What the agents may do on their own. Which steps run without a person, and which wait for an approval. See approvals and control for AI agents.
Which rules are current. When a price list, an approval threshold or a contract term changes, the owner confirms the new rule before agents use it.
Whether it is working. The owner reads the results against the benchmark agreed before launch, on quality, cost, time and the human effort still needed. See how to measure AI agent ROI.
What comes next. Which piece of work to add, and which to stop.

For high-risk systems, the EU AI Act puts part of this in law: deployers must assign human oversight to people "who have the necessary competence, training and authority" (Article 26(2)). Most business agents are not high-risk, but it is a sensible test for any of them.

Three ways to split the job with a partner

	Your team runs it	Partner runs it for you	Shared
Who decides what agents may do	Your owner	Your owner	Your owner
Who updates rules and workflows day to day	Your team	Partner, with your owner's approval	Your team for rules, partner for workflows
Who watches the runs and evaluations	Your team	Partner, reports to your owner	Both, with one shared view
Who hosts and operates the platform	Partner (or you, if you build in house)	Partner	Partner
Fits when	You have someone with time and interest to run it	The work matters but nobody internal has the hours	You want to build skills while the first pieces settle

The first row never changes. A partner can do the work of running agents, but deciding what your business allows stays with you.

What we agree with clients before launch

At Ortelian, after launch the client's team runs the work or we keep running it for them. Either way we host and operate the platform, which runs in the EU, and the evaluations keep running on the work in use. Before go-live we agree on:

The named owner. An internal owner with the authority to make decisions and drive implementation, ideally a founder or someone on the leadership team. It is the first thing on our list of what we need before we start.
Who approves which change. Every change to an agent or workflow is shown in full before it goes live, and each run records the version it used. The owner, or someone they name, approves.
The benchmark. The baseline from the audit and the standard the work must meet, built from real examples of the client's work.
What the client owns. The client owns its data and its world model, and can export both, along with the full history of how its agents were set up. Nothing about the setup is locked to us.

Workshops during the build prepare the client's team for this. The goal is that the owner can read what an agent did and why, without asking us.

Signs nobody owns it

The agent's rules were last changed by the people who built it.
Exceptions go to a shared inbox and nobody is sure who picks them up.
Nobody can say what the agent saved last month.
People work around the agent instead of telling someone it is wrong.

The fix for each is the same: name the owner and give them the four decisions above.

Questions people ask

Does the owner need to be technical?

No. The owner decides what the business allows and reads the results. They need authority over the work, not coding skills. Technical changes are made by whoever runs the platform, and the owner approves them.

Can the partner be the owner?

A partner can run and operate everything, but should not decide what your company allows. If the partner left tomorrow, someone inside would still need to know what the agents do and why.

What happens to our data and setup if we stop working with the partner?

That depends on the contract, so ask before you sign. With Ortelian, you own your data and world model and can export both with the full history of how your agents were set up.

How much time does the owner spend on it?

It depends on how much changes in the work. Plan regular time to review results and approve rule changes, and more in the first weeks after launch while exceptions are still being worked out.

Sources

Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027", press release, 25 June 2025.
Gartner, "AI Agent Success Requires Reliability Before Autonomy", section "Accountability remains a human responsibility".
Regulation (EU) 2024/1689 (EU AI Act), Article 26(2), obligations of deployers of high-risk AI systems.