Your Team Built 50 AI Apps. Who Owns Them?

Why this paper exists

A working app can now start with a prompt.

It can be ready to demonstrate that afternoon. Organisations should welcome that. The harder questions begin when someone starts relying on it.

Who owns it? What can it access? Who supports it when the person who built it leaves?

Building is solved. Owning is not.

AI has made software easier to create. Organisations now need to make it easier to take responsibility for that software.

The second problem

A useful experiment quietly becomes part of the working day.

A caseworker builds a case board. A controller creates a dashboard. A project lead puts together an agent that drafts replies.

The people closest to the work can turn their ideas into useful tools without waiting for a development project. They can test an approach, learn from colleagues and improve it while the problem is still fresh.

That is worth protecting.

But a working prototype leaves operational questions unanswered. An app may run under a personal account. Its connections may depend on the builder’s credentials. Colleagues may start using it before anyone has agreed who maintains it.

The transition can happen quietly. A useful experiment becomes part of the working day. Then its creator changes roles, an integration fails or someone asks for a record of what happened.

By then, the organisation depends on something it has never formally taken ownership of.

An app and an agent need different controls

Displaying information is not the same as taking action.

An app that displays information needs appropriate access controls, reliable operation and a named owner.

An agent that takes action needs more. It may update a record, send a message or trigger a process across several systems. The organisation needs to define what it may do, whose authority it acts under and when a person must approve the next step.

Consider an agent that helps process invoices. Reading an invoice, suggesting an account code and authorising a payment carry different responsibilities. Access to one step should not automatically grant access to the others.

Those boundaries need to be enforced when the action happens. They also need to remain visible afterwards.

A log should help someone establish what the agent accessed, what action it took, which permissions applied and whether a person approved it. Where a decision requires an explanation, the workflow needs to capture the relevant evidence and rationale.

Build anywhere. Govern execution in one place.

Keep the builders’ tools. Share the way into production.

People should be able to keep using the tools that help them build. The organisation should provide a shared way to connect those tools to business systems, control their actions and operate them over time.

That requires:

• Named ownership. Someone is accountable for the app or agent, with an agreed support and handover process.
• Separate agent identities. Agents act through assigned identities and permissions instead of inheriting unrestricted access from a builder.
• Controlled integrations. Access to business systems is managed through shared connections with defined boundaries.
• Approval at the right steps. Actions with significant consequences require human review before execution.
• Traceable activity. Records make it possible to investigate actions, approvals and failures.
• Explicit data handling. Storage, processing locations and external services are understood and approved.

These controls give builders a clearer route into production. They also reduce the work required to assess each new solution from scratch.

Where Copyl fits

A shared operating layer for AI agents.

Copyl provides a shared operating layer for AI agents, with identity, permissions, integrations, approval controls and audit trails. Agents can be built in Copyl or with other tools, then connected to business systems through governed execution.

That gives organisations a common foundation for running agents across departments. Teams can develop solutions around their own work while IT manages access and operational controls centrally.

Deployment and data handling still need to be assessed for each solution, including any external model providers or services it uses. A shared platform provides controls; the organisation remains responsible for how they are configured and used.

Five questions before an AI app goes live

Before an app or agent touches real business data, ask:

The ability to build fifty apps is an opportunity.

Making ownership, access and support clear is how an organisation gets to keep using them.

Already running a growing collection of AI apps or agents? Talk to Copyl about bringing them into governed operation.