Most leaders at this level talk about AI. I build it. My AI Chief of Staff, Jeeves, runs my mornings, my portfolio, and half my admin, and I wrote every line myself, starting from zero code experience.
I've spent three decades building the partner infrastructure that lets large organisations actually adopt new technology: AWS, NVIDIA, Symantec, Citrix. Not the pitch. The plumbing. The relationships, the enablement, the go-to-market motion that turns a platform into something enterprises actually use.
Right now I lead Training and Certification Partnerships for AWS across EMEA: 189 partners, 50+ countries, a team of ten. Before that, NVIDIA's enterprise channel through the accelerated computing shift. Before that, Symantec, Citrix, and a board pitch to Cable and Wireless in 1997 that trained 2,000 staff through a technology transition most of my current peers weren't yet in the workforce for.
The throughline hasn't changed in thirty years: making the human case for technology change, at scale, at the executive level. Only the technology has changed.
I had never written a line of code before this. Jeeves is a fully agentic system I built and continue to build myself: he reads my portfolio, tracks career opportunities, manages my calendar and reminders, drafts and ships his own code updates with my approval, and talks to me over WhatsApp like a member of staff would.
He isn't a chatbot wrapped around a prompt. He has memory, a scoring engine for investment signals, a paper trading desk running a real strategy, and a build loop where he drafts his own code changes for me to review and ship. He runs on infrastructure I set up and maintain: GitHub, Render, Twilio, a proper email pipeline. I lead him the way I've led every team for thirty years: give him trust, clarity of purpose, and room to do the job.
Most people building an AI agent for the first time end up with something that answers questions. Jeeves does more than that, and the best way to show it is to walk through one real thing he did, start to finish.
The problem. Jeeves needed a way to spot promising investment signals without me having to check the market all day. Not a headline scanner. Something that could weigh several kinds of evidence together and get more useful over time.
What we built. A signals engine that pulls in fresh news, checks momentum, reads insider trading data, and scores each stock on a composite scale. It runs on its own schedule, re-checking names through the trading day, and it flags anything that crosses a meaningful threshold.
Where it got interesting. A scoring engine is only as good as its weights. So we added a second layer. Every week, Jeeves reviews his own closed trades, works out which factors actually predicted a good outcome, and proposes small adjustments to how much weight each signal gets. He proposes. He never applies. I get a message asking me to approve or skip each change. Nothing adjusts itself without me saying yes.
The governance bit that actually matters. Every change to Jeeves, including changes to how he scores investments, follows the same rule. Draft it, show me the diff, wait for explicit approval, then ship. Nothing merges into anything he actually runs without me looking at it first. That is not caution for the sake of it. It is the same discipline I would expect from any team I have run for thirty years. Trust people, give them room, and keep the checkpoints that matter.
I'm building toward a senior AI leadership chapter: Chief AI Officer, VP Partnerships, Chief Partnerships Officer, or a Non-Executive Director role where practitioner credibility and ecosystem experience both matter.
For FY26, an EMEA organisation set out to bring Generative AI adoption to the top 20% of its highest-spending customers. Fewer than 10% of that base had ever received a training offer, and fewer than 2% had been connected to a training solution. Across roughly 24,000 customers, the job was building a systematic route from customer intent to delivered training, through a network of 189 training partners across 50+ countries.
Earlier attempts had failed on data quality and internal buy-in. Account teams held the customer relationship while training partners held delivery capacity, with no agreed rules connecting the two and no incentive for either side to make it happen. Some markets had strong field sponsorship; others needed a different, more scaled approach. And there was no governance model for scaling a fully subscribed programme responsibly.
Leading the programme, I designed a three-layer engagement architecture: a relationship-led motion for account teams already close to the customer, a scale-led motion using AI tooling to reach customers relationship-led engagement alone couldn't touch, and a conversion layer through training partners across every EMEA sub-region. Courses were matched to customer maturity rather than pushed uniformly, and the whole programme was built around utilisation and paid-conversion gates rather than an open-ended commitment, so growth stayed accountable to actual customer demand, not ambition.
Within three months of launch, 14.5k customers had received a Generative AI training offer, covering 54% of the customers in scope. 65% of account managers actively engaged with the programme across a dispersed team. 811 people registered and 532 attended, across 34 events. In the UK and Ireland, half the registered companies were entirely new to the partner network. Sub-Saharan Africa alone contributed 87 attendees across 66 customers in 6 countries. The results supported approval for a 4,255-place next-phase capacity plan, built on the same gated, controlled model.
It's the same throughline as everything else here: relationship-led and scale-led work well together when the governance is disciplined and the reporting is honest.
Separate from any operating role, I take Non-Executive Director and Advisory positions with organisations that are serious about AI. This is not the operational job of a Chief AI Officer, running teams and shipping systems. It is the board-level one: oversight, assurance, and asking the question that stops a bad deployment before it becomes a bad quarter.
The distinction matters. A Chief AI Officer owns the build. A Non-Executive Director owns the governance of it. I can do the first, and that is exactly what makes me useful at the second.
Most enterprises have already deployed AI. Very few have deployed the governance to match. Models are in production, agents are taking actions, and the oversight was designed for software that did not make its own decisions.
That is not a technology problem. It is an organisational design problem: who approves what, where the human checkpoint sits, and what happens when the system is confident and wrong. A Non-Executive Director without hands-on experience can read the policy. One who has built an agentic system knows where the policy quietly fails, because they have watched it fail on their own build and then fixed it. That is the gap I close.
A worked example, from my own system. The clearest way to show what I bring to a board is not a client story. It is what building Jeeves, my own agentic AI system, taught me about governing one.
The problem. Agentic systems fail in a particular way: autonomy outruns accountability. The system is given room to act, the checkpoints do not keep pace, and by the time anyone looks, it has already done the thing. The failure is rarely the model itself. It is the design around the model.
What I built. Jeeves does not reach a trading decision in a single voice. He runs three separated roles: an Analyst who makes the case, a Sceptic who argues the other side without ever seeing the Analyst's reasoning, and a Risk Officer who rules on both. Separation on a diagram is easy. Proving it is not. So the separation is mutation-tested: the suite deliberately tries to break the wall between the roles and confirms the checks catch it. An independence that cannot be broken on purpose is not independence. It is a diagram.
The discipline at its highest stakes. The same propose, approve, execute rule governs the part that touches money. The trading book runs on paper, and real capital sits behind a go-live gate that was defined in advance: a track record held over time, performance ahead of its benchmark, and drawdown kept within a set limit. Capital does not go live because the system feels ready. It goes live only when the evidence clears a bar that was set before anyone was tempted to move it. Restraint, written down ahead of time, is the governance.
The failure modes I caught myself. The instructive part was not the design. It was what went wrong inside it. A test suite that reported success while proving almost nothing, until mutation testing exposed that the checks were green without actually biting. Gaps where the automated guardrails did not cover what I had assumed they did, found only by running the full check instead of the convenient one. Behaviour that no ordinary test could pin down, because the model is non-deterministic, and so had to be proven with cases built on purpose rather than asserted and hoped for. Each of those is a governance lesson before it is a technical one, and none of them is visible from a policy document.
Draft it. Show the diff. Wait for explicit approval. Ship. Nothing executes without sign-off, and that includes changes to money. It is the rule Jeeves runs on every day, and it is the instinct I bring to a board: not to slow AI down, but to make sure that when it acts, a person chose to let it.
The best way to reach me is LinkedIn or email. I read everything myself.