Ben Grinnell, chief AI and capabilities officer at Newton, says letting one consultant both hear the problem and build the first version removes the handoffs where context gets lost.
Grinnell warns that a fast AI demo can look finished while security, nonfunctional requirements and architecture are still missing, so experienced people must ask the right questions.
Grinnell says organisations only move at the speed of trust, and predicts chief of staff agents will become common within six months, while AI makes long-hidden corporate data far easier to find and expose.
Why AI Value Comes From Narrow Wins and Organisational Trust
AI value comes from shrinking the distance between the business problem and the first working solution, while keeping engineering discipline and trust in place. Fast builds look finished before security and architecture exist, and large programmes lose momentum. Narrow use cases that show real value, judged by revenue and service rather than headcount cuts, are what organisations will actually adopt.
Takeaways
One person hearing and building removes lost context. Putting one consultant in front of the client, who listens to the problem and builds the first version, removes handoffs where context leaks like Chinese whispers. Ben compares it to the earlier shift from outsourcing to DevOps.
Fast demos hide missing engineering fundamentals. A quickly built AI solution can look finished while security, nonfunctional requirements and architecture are still absent. Experienced people who know which questions to ask are still needed, and IT must know about AI-built tools.
Start narrow and create pull, not push. Eighteen-month projects are out of date on arrival, and large transformation programmes lose momentum. Ben favours one narrow use case that shows business value, creating demand for services in organisations already overloaded with big projects.
Adoption moves at the speed of trust. Use grows from individuals automating back office work to expert agents answering colleagues and agents shared by teams in Slack. Ben expects chief of staff agents to be common within six months. AI also makes long-hidden corporate data easy to find, raising access risks.
Measure value by revenue and service, not headcount. Ben would be pleased if one in twenty experiments delivered a P&L difference, against the cited 95% failure claim. Baselines shift because tools change before studies finish, and headcount targets provoke rejection. Revenue, service quality, profitability and removing toil work better.
Legacy systems can be rewritten with bugs intact. Ben suggests AI could rewrite a legacy system with its bugs preserved, so user teams keep working while change management proceeds separately. Tools like Blitzy examine legacy code bases and bring their context into AI-assisted work.
Newton is building sovereign compute for inference. Tokens are expensive and some client data should not leave the UK, so Newton is building UK inference capability. Ben predicts an inference market crunch. Microsoft pulling out of a contract weakened Newton's infrastructure hopes, but staff still get the best models.
Tap a name to go there · tap a line to see why they are linked