The CBI's “execution divide” in AI adoption

A report from the CBI and Oliver Wyman, published in late August, made a point that's easy to miss.

63% of firms surveyed have moved AI into scaled or agentic use in design and R&D. That sounds like progress.

But the report calls out something more useful than the stat: an “execution divide”. The gap isn't between firms using AI and firms that aren't. It's between firms running isolated pilots and firms that have wired AI into the whole workflow, concept through procurement through delivery.

A tool bolted onto one stage of a project doesn't change how the work gets made. The same report names what still won't get automated: taste, critical judgement, emotional intelligence, and the ability to audit what the AI produces. That's the part of the job that was always the actual job.

There's a harder number buried further down. Graduate design vacancies are down 50% year over year. If entry-level roles disappear before firms redesign them around AI-assisted work, there's no pipeline left to produce the judgement the report says matters most. There's also a quieter risk worth flagging to anyone briefing a design partner. Training models on client assets carries real legal exposure, and most firms don't have clear data licensing or provenance protocols yet.

If you're briefing a design partner, ask which side of that divide they're on, and who they're training up to replace them.