Liquid labour: my take on agentic AI's impact on the future of work

A few weeks ago I was in a conversation with a founder who told me, very confidently, that AI was going to "eliminate most white-collar jobs within five years." He wasn't being provocative -- he genuinely believed it. And I've heard some version of that claim a lot recently.

My read is that it's wrong. Not slightly wrong -- structurally wrong. And I think the people who get this right will have a significant competitive advantage over the next few years.

Here's the thing most people miss: we're not in a period of labour surplus. Most developed economies are facing genuine workforce shortages. Australia's skills gap is well-documented. The US has had more job openings than unemployed people for years. The EU is dealing with an ageing workforce and declining birth rates across most member states.

In that context, AI doesn't function as a replacement. It functions as a multiplier.

The data backs this up. Only about 4% of organisations actually reduced headcount following AI adoption. The dominant pattern was reallocation -- people freed up from routine tasks moved toward higher-value work. This is what economists call the reinstatement effect, and it's been the dominant pattern in every prior technological shift.

So what is actually changing?

Middle management, mostly. And this is real. AI is very good at the things middle managers spend most of their time doing: synthesising information, routing decisions, tracking progress, scheduling coordination. If your role is fundamentally about information flow and coordination overhead, that's the layer getting compressed.

But this isn't the same as "managers are being replaced." It's more that the coordination layers are being automated, which flattens structures and pushes accountability up and down simultaneously. Senior people take on more direct oversight. Individual contributors get more autonomy. The middle compresses.

The organisations that handle this well aren't the ones eliminating the most roles. They're the ones figuring out how to run a hybrid model -- a rigid core and a fluid periphery -- at the same time.

The liquid labour matrix

I've been thinking about this through a simple framework I'm calling the 'liquid labour matrix.'

The horizontal axis is how humans are being used -- as task-doers executing specific work, or as system-overseers managing and directing AI and specialist agents. The vertical axis is the organisational structure -- rigid roles and permanent headcount, or fluid networks of temporary and project-based teams.

The competitive advantage lives in the right-hand column. Not pure AI automation (bottom-left) and not pure human hierarchy (top-left). The winning model is running a rigid core of experienced people alongside a fluid periphery of AI and specialist agents -- simultaneously.

The rigid core is the institutional memory of the business. Long-term employees who carry the knowledge that can't be documented: client relationships built over years, the unwritten rules of how decisions actually get made, the context that makes data meaningful. These people become system-overseers -- directing the AI, interpreting outputs, making the calls that require accountability.

The fluid periphery is where you get scale without fixed cost. Project-based specialists, AI agents handling execution work, contractors brought in for specific capability. You spin them up, you spin them down. The work gets done without the overhead of permanent headcount.

Why I think this model holds

Three reasons I keep coming back to.

Shareholders still require human accountability for major decisions. When something goes wrong -- and it will -- someone has to stand in front of the board or the regulator and own it. AI can't do that. The institutional knowledge and the accountability function are inseparable, which means experienced human judgment stays valuable at the top of the stack.

Institutional memory is hard to replicate. The stuff that makes a business defensible -- the client relationships, the product intuition built from years of iteration, the knowledge of what's been tried before -- lives in people. You can try to document it, but most of it doesn't survive the documentation process intact. The rigid core protects this.

Sensitive operations require human gatekeeping. Legally, ethically, and practically, there are categories of work -- anything touching personal data, high-stakes negotiations, regulated decisions -- where you can't hand off to an AI agent and call it done. Someone has to be in the loop. The question is just how much of the preparation and execution work gets automated before that person makes the call.

The strategic implication

The organisations that will struggle are the ones trying to make a binary choice. They're either protecting every role because "AI can't replace human judgement" (true, but not the right frame) or automating everything they can as fast as they can without thinking about what they're preserving (faster, but fragile).

The ones that will win are treating this as an organisational design question, not a technology question. They're asking: what's in the core, and what's in the periphery? What do we protect because it's genuinely irreplaceable, and what do we automate because it's overhead?

The window to ask those questions thoughtfully is now -- before the fluid periphery becomes the default, and the rigid core gets hollowed out by accident.

So that leaves the question I keep coming back to -- if your organisation ran this framework, which quadrant are you in right now? And which quadrant are you building toward?

Disclaimer: Thoughts are my own and do not represent any other parties.