There's a quiet shift in how the sharper operators are describing their organisations. They've stopped talking about AI as a tool they've bought and started talking about it as a kind of capital they're accumulating - something that sits on the balance sheet of the business alongside the people, and grows or wastes depending on how it's managed.
Satya Nadella put a useful pair of names to it. There's human capital - the knowledge, judgement, relationships, taste and pattern recognition of your people. And there's token capital - the AI capability you build and own: the agents, the prompts, the context and data you've made usable, the loops that turn all of it into work getting done. The organisation of the future, in this framing, isn't all of one or all of the other. It's a deliberate ratio of the two.
The short version: Human capital and token capital aren't competitors fighting over the same headcount. They compound. The moat isn't the model everyone can rent - it's the learning loop where your people and your AI make each other better. Manage the ratio; don't just cut one to fund the other.
Why this isn't the usual "AI replaces jobs" story
The instinct, when a new kind of capital shows up, is to treat it as a substitute. More token capital, less human capital - swap the expensive thing for the cheap thing. That's the spreadsheet version, and it gets the dynamic exactly backwards.
Nadella's argument is that human capital becomes more valuable as token capital grows, not less. The reason is that AI capability without human direction is, in his phrase, compute running in circles. Someone has to set the ambitious goal, connect the dots across domains, hold the relationship, and recognise which pattern actually matters. The more capable your token capital, the more leverage those human judgements carry - and the more it costs you when they're missing.
The last thing any organisation wants is to cede its distinctiveness to a model that every competitor can rent by the hour. The advantage was never the model. It's what only you can teach it.
- on Satya Nadella's human-capital / token-capital framing
The moat is the loop, not the model
If the frontier models are available to everyone - and they are - then buying access to one gives you no advantage your competitor can't buy on the same afternoon. The advantage lives one level up, in the learning loop: your people teaching your AI the specifics of your business, your AI giving your people more time and reach, and each turn of that loop making the next one better.
That loop is the thing that compounds, and it's the thing nobody can copy, because it's built from your context. Two firms can license the identical model and get wildly different returns, entirely depending on whether they've built the loop or just bought the licence. It's the same point we make about treating AI as infrastructure rather than a product - the value isn't in the thing you purchased, it's in everything you wire into it.
Rest in your uniqueness
For the people inside the organisation, the practical instruction that falls out of this is oddly reassuring: lean into what's uniquely yours. Token capital is very good at the commodity work - the first draft, the summary, the lookup, the routine routing. The way you add value on top of it is to spend your now-freed attention on the things only you can do: the relationship, the judgement call, the taste, the thing your particular experience sees that a general model never will.
"Rest in your uniqueness" is the mindset, and it's a healthier one than the anxious version where everyone tries to out-type the machine at the machine's own game. The machine wins the typing. You win the judgement.
How an organisation actually builds token capital
The unglamorous truth is that token capital is built, not bought, and it's built out of the things most organisations never bother to write down. The operational know-how that lives in people's heads - how this process really works, what good looks like, where the exceptions are - is exactly the context your agents need to be useful. The old operational manual nobody read becomes a living set of instructions your AI actually uses, and keeps getting better as your people refine it.
That's why the organisations pulling ahead are the ones investing in the plumbing: making their knowledge usable, building reusable agents instead of one-off prototypes, and treating each thing they build as an asset that compounds rather than a project that ends. It's the same structural work the GIVE framework describes - governance, ideas, value and experimentation - pointed at a single goal: growing token capital that your human capital can compound on.
Thinking about the ratio? Building token capital that compounds with your people is exactly what our Accelerator is for. Get in touch if you'd like to talk through where your organisation is starting from.