I help organisations work out where agents fit and what has to change around them, working with FTSE 100 and public sector clients at Elsewhen. I also run the Machinepower Index, which measures which countries can actually build and control AI, and write about AI and geopolitics for The Spectator, UnHerd and VentureBeat.
Most enterprise AI stalls somewhere between the pilot and production, and the model is rarely the reason. It gets dropped into structures designed around people, with approval chains, handoffs and job descriptions that leave no obvious place for an agent to do the work.
of enterprise AI projects fail, either never reaching production or never earning their keep, which is roughly twice the failure rate of IT projects without AI.
RAND Corporation ↗a year is the shortfall Bain expects by 2030 between what AI earns and what it needs to pay for its compute, even counting the savings it brings.
Bain & Company ↗of generative AI pilots make no measurable difference to the P&L. One in twenty shows up in the accounts.
MIT Project NANDA ↗AI doesn't take your job. It takes your tasks, and that splits the work in two: what agents run and what people own. Most organisations haven't drawn that line yet, because they are still built on the assumption that every task needs a person.
My argument is that AI is the defining strategic capability of the century, and the index measures who actually has it, scoring countries on watts, weights and will across 12 datasets to separate the AI makers from the AI takers.
View the indexAt Elsewhen I work with FTSE 100 and public sector clients on agentic AI, and coined the Agentic Enterprise, now central to the company's strategy. Before that I founded ACQUAINTED, Europe's first generative AI consultancy.
Visit ElsewhenTalks and lectures on agentic AI and AI sovereignty at the University of Oxford, the Swiss National Museum and FTSE 100 companies, as well as at conferences and executive offsites. I'm represented by Champions and the London Speaker Bureau.
Book a talkReporting and argument on AI, geopolitics and who ends up controlling the technology, from China's AI push to the stand-off between the frontier labs and the state, with longer essays on the Machinepower Substack.
The Spectator VentureBeat UnHerd Quillette SubstackFor three years everyone ran the same play of bigger models, more pilots and a chatbot bolted onto the old systems. My 2026 report argues that consensus has broken, and follows the five forks and twelve trends where the paths now split, from agentic warfare and world models to the intimacy economy.
Email is the quickest way to reach me about consulting, the Machinepower Index or a talk, and speaking bookings can also go through Champions.
