It starts innocently. Someone on your team builds a small tool with Claude, vibe-coded on a personal account or inside a chat artifact: quotes pulled from a spreadsheet, a summariser for approval requests, a checker for freight documents. It works. A colleague asks for the link. Then their team asks. Within a quarter, a workflow that ran on copy and paste now runs on this thing, and nobody signed anything.
Ownership is decided the day the app is born, not the day IT asks. Who owns the code, where the data lives, what happens when the prompt changes, who is liable if it hallucinates, what IT will demand before it touches company data: those questions cost almost nothing to answer on day one, and a great deal once the tool has users.
The questions nobody asked at the start
Who owns the code? Built in a personal account, the app sits in a grey zone between the author, the employer and the platform’s consumer terms. Consumer subscriptions are built for individuals, not companies, so the IP, data and support terms a business needs are usually not in the plan the tool was born on.
Where does the data live? If colleagues paste customer records, pricing sheets or contract clauses into a personal chat to make the tool useful, company data has left the company’s control, and possibly its region. Data residency is a much cheaper conversation before the data exists in the wrong place.
What happens when the prompt changes? A prompt is the app’s logic. Change one paragraph and behaviour shifts for every user on Monday morning, with no version history, no release note and no way back. Colleagues who rely on the tool are now relying on something that can change silently overnight.
Who is liable if it hallucinates? A plausible-sounding wrong quote or a mis-summarised clause is harmless in a demo and consequential in a decision. Without evaluation evidence or a log of what the tool produced, nobody can even reconstruct what went wrong, let alone decide who is responsible for it.
What will IT demand? Authentication, an access matrix, an audit trail, backup and restore, a data processing agreement, a named support owner: the review arrives the moment the tool matters. Apps that can pass the review survive it. Apps that must be rebuilt under pressure often do not.
Five things to check today
If any of this describes a tool in your organisation, five checks take an afternoon:
- Account and export paths. Whose account was the app built in, and can the code and data come out in a usable format, today, without anyone’s goodwill?
- Data residency. What company data has passed through personal accounts, and what did the terms in force at the time say about it?
- Access control. Who can reach the tool now, who can see what it stores, and is that the list you would choose?
- Evaluation evidence. How do you know the tool is right on your cases? Golden examples, spot checks, a record of what it produced for real users: any evidence beats a shrug.
- An exit story. If the author is unavailable tomorrow, who can rebuild, run and defend the tool? If the answer is nobody, you have a single point of failure wearing a friendly face.
Two routes to production
The fix is not to kill the tool. The instinct that built it was right: the workflow was worth automating, and someone proved it cheaply. The fix is to give the app what software needs, which is an owner, an environment, controls and evidence for the people whose job is risk.
There are two clean routes, and both start with the same short fixed-fee readiness review of what you actually have, with a firm written price before anything is committed. Route one is managed EU hosting: the app keeps running, now inside a managed platform with separate test and live environments, authenticated access, encrypted storage, backups with rehearsed restores, and the access matrices and release records your IT team will ask for. A first pilot runs in about three weeks. Route two is a documented IT handover: the app is adapted into a corporate environment you choose, with full documentation and a rehearsed handover, and your people operate it. Either way, the builder keeps building. The point of production is to protect what they made, not to confiscate it.
We run both routes as Claude App to Production, and the readiness review produces a firm written price for your specific application before you commit.
The best day to answer the ownership questions was the day the app was born. The second-best day is today, while the answers are still cheap.