Don't hand over requirements definition when outsourcing AI, adviser warns

The column opens from the author's own advisory experience: IT departments at large companies frequently tell him they want to hand AI development to a vendor, but they have heard of hand-offs that failed. His framing is that even big enterprises rarely have enough people who can build AI agents tailored to a specific business task, so outside help is a given.

The fictional case makes the tension concrete. A manufacturer with about 10,000 employees is under pressure from its corporate planning unit to show AI results aligned with company strategy, specifically by automating responses to internal inquiries to its quality-assurance department. Its IT staff, however, is consumed by a core-system replacement and counts only a few AI-savvy employees. That is the gap the systems integrator steps into, offering requirements definition through operations plus resident engineers.

The author's stated concern is not whether to use outside help but what to retain control of, and he names requirements definition as the thing a company should not simply hand over. The rest of the piece promises to work through which elements must stay in-house, using the fictional case as the vehicle.