TL;DR

Gartner used the opening leg of its IT Symposium to tell CIOs something vendors will not: the leading AI labs do not yet operate like enterprise suppliers, and the more AI an organisation runs, the harder both control and return on investment become. Its research puts 86% of CIOs in the camp that AI risk is growing faster than AI value. The proposed remedies are structural, not technical.

The supplier problem

Distinguished VP analyst Daryl Plummer was direct about working with frontier labs: “Trust in these vendors is not warranted yet. They are not enterprise grade. They don’t understand enterprise terms and conditions. They don’t understand enterprise liability.”

The concrete grievance is versioning. Models get altered frequently with little apparent regard for the applications depending on their output, and with a working life of about six months there is no legacy support to fall back on. Stay on an older version, Plummer said, and the vendor stops paying attention. His summary — “It’s an out-of-control pace of innovation, and the sad thing is you can’t afford not to follow it” — captures the bind precisely.

Careless consumption

Kristin Moyer’s half of the argument concerns demand rather than supply. Early wins generate more appetite for AI than IT can safely service, and some of that appetite is what she terms careless consumption: reaching for a model when nothing required one. She cited findings that 40% of workers have run into AI slop, and that untangling it costs about two hours a time — which she scales to roughly $9m of wasted work annually in a 1,000-person organisation.

Visibility is the underlying issue. Counting your agents is hard when AI arrives inside software you already licensed. Plummer’s counsel is unfashionable: a database query does not need an agent when a function call has always worked.

Looking forward

Three recommendations stand out — an “AI central bank” owning systemic oversight, guardian agents empowered to kill misbehaving ones, and disaster recovery teams for AI-created messes. The accountability point is the sharpest, since ERP and CRM leave an audit trail and AI frequently does not. UK boards should note Plummer’s closing line: “CIOs will be held accountable for AI failures.”