The Board Wants AI. the Data Is Still on the Ship.

A close friend of mine recently finished onboarding with a new employer, a large international company that operates vessels offshore. When we caught up, he told me about a company-wide event where the CFO made a strong case for AI. The CFO wanted the business to start building its own applications with AI.

My friend works in IT, and he had just become responsible for approving his colleagues’ requests to use AI assistants. I asked what he checked those requests against. Apart from his own judgment, nothing: no checklist, no guidance. Each request started with a budget of about $10 for AI usage. And then, almost as an aside, he mentioned a pilot. The plan was to capture raw events from several of the company’s ships and collect them on a data platform the company owns. A systems integrator was standing by, the investment was modest, and IT had blocked it.

I have thought about that conversation a lot since then. Nobody in it was being unreasonable, and that is what stayed with me.

What I Recognized  

At board level the logic is sound. AI will change how the business works, and waiting looks like the bigger risk, so the message goes out: build with it.

Further down, every request lands on someone like my friend, who has to decide whether a colleague may send company information to an AI service, and who has no criteria to guide that decision. I can see why someone in his position would hesitate when nobody can tell him what a dataset contains, who owns it, or where it will end up.

I don’t know exactly why IT blocked the pilot. But I’ve seen this pattern before: without an approved way to collect, describe and govern operational data, waiting is often the safest answer. In twenty years as a data and analytics consultant, I watched leaders set goals at the top while data teams were still getting the basics in place: knowing which systems held what, landing the data somewhere reliable, and keeping track of how it changed.

Gartner’s May 2026 warning  about AI agents and semantics makes a related point: schemas alone don’t give agents the business context they need. But the company in my friend’s story faced an earlier obstacle: its pilot to collect operational data was blocked.

Why We’re Building Amfelt  

Amfelt does the unglamorous part. Its agent runs inside your network, inspects your source systems, documents what it finds in an open data contract, and can recommend how each table should be loaded. Your team can review any recommendation, and those the model is unsure of wait for a person. Amfelt then loads the data, builds the data platform layers, tracks changes over time, and keeps the documentation up to date. The agent reports back a description of the data and the outcome of each run.

It would not have written my friend’s AI policy. It is also a batch platform, so it is not the tool for streaming live telemetry off a ship. But it is built to establish what the data contains and where it will live. That gives IT something it can say yes to.

We’re opening Amfelt to a small group of teams first. If you’re trying to get a data pilot approved, or you’re the person expected to approve it, we’d like to hear from you. Join the Waitlist .