Could AI wipe out middle management?

To actually make a difference, AI must tackle the weakest link of any given problem. For corporate structures, this could well mean the erosion of middle management, writes Lewis Liu

If you look at the last 150 years of American economic growth, by and large it has stayed on an average trend of around two per cent a year, despite enormous technological disruption: the steam engine, internal combustion, electricity, airplanes, computing and the internet. How is it, as Stanford economist Chad Jones recently pointed out, that we have access to computers with 100m times more transistors than in the 1970s, yet he, as a researcher, is perhaps only two or three times more productive?

This is the central question for many policymakers and economists trying to understand how AI will impact the economy, and, for me, what kind of bet I should be making in building a product that anticipates where value will accrue in this new AI-enabled economy.

Having just completed our recent seed round for my start-up, Twin1 AI, my co-founders and I were sitting with a few investors at an Indian restaurant in Palo Alto celebrating. One of our investors, someone I have known for many years and who is a deeply thoughtful man, asked us how our product would be positioned in this wave of AI revolution, and how that related to some of the emerging theories around AI’s impact on the economy.

Which is how he brought up Professor Chad Jones’s recent work.

What is the ‘weak links’ theory?

Jones maps out America’s remarkably stable historical growth rate of around two per cent and asks: why, given all these technological revolutions, has it remained so consistent?

His explanation is the “weak links” theory.

The analogy Chad gives is the iPhone. Think about the entire chain of processes required to make one: design, sourcing, manufacturing, logistics, shipping, retail and marketing. If any one of those parts falls down, the whole thing can fall apart. Most activities in our economy are the same. Making one part of the chain dramatically more productive helps, but ultimately the entire system remains constrained by its slowest or weakest parts, just as a chain is only as strong as its weakest link.

We see this in real life in the knowledge economy with AI.

I recently had a client tell me they had bought an HR workflow AI agent that was meant to reduce HR headcount by, say, 10 per cent. The AI agent automates one of perhaps 10 tasks an HR officer performs, so the theory goes that the HR team should become 10 per cent more efficient – or perhaps 10 per cent smaller.

Reality doesn’t work that way.

You still need a human to connect the output of that AI agent to the other nine tasks in order for the overall job to get done. In the end, the HR team felt threatened, the company spent a significant amount of money on the AI agent (so costs actually went up), but the team stayed exactly the same size. No productivity was gained, but people got angrier.

How long does technology take to make structural differences?

Now, I’m using this example not to say that all AI agents have this problem. I hear plenty of examples of significant productivity gains in specific areas of knowledge work. But it illustrates the broader point.

One implication of Jones’s argument is that technology takes time to diffuse across an economy. Making 17 links in a 20-link chain fantastically strong doesn’t transform the chain if the remaining three are still weak. By the time the gains from one technological revolution have diffused throughout the economy, another technological revolution emerges and the cycle begins again – yet over the long term, we keep compounding at roughly two per cent a year.

Jones argues that perhaps we can be more optimistic with AI. Perhaps this time we can accelerate beyond that two per cent trend. But he also suspects we won’t get the hockey-stick effect for quite some time, because AI is nowhere close to permeating the entire economy, and nor is much of the technology ready yet.

Consider one of the most important technological transitions in economic history: from steam to electricity in manufacturing.

In the steam era, factories typically had a central engine driving a network of overhead shafts, pulleys, gears and long leather belts to distribute mechanical power throughout the building. This created significant layout constraints and inefficiencies because everything ultimately had to connect back to this central source of power.

Then came electricity and the electric motor.

Initially, factories often simply replaced the steam engine with an electric motor while leaving much of the existing paradigm intact: one central source of power feeding the same shafts and belts. Unsurprisingly, that captured only a fraction of what electricity could actually do.

The real transformation came as electricity diffused throughout factories and smaller electric motors began powering individual machines. Suddenly, machines could be placed where they made sense for production rather than where a belt could reach them. Factory floors could be completely reorganised around workflow, becoming more flexible, efficient and safer, and ultimately helping to pave the way for modern mass production.

What’s the lesson here?

It wasn’t the electric motor itself that revolutionised productivity. It was the entire paradigm shift around it. Electricity had to diffuse across the ecosystem, and organisations had to redesign themselves around what the technology made possible, before the economy could reap its full rewards.

Could middle management be wiped out by AI?

So where does that leave us in this world of AI?

I say two things.

First, the weak-link theory suggests that value will accrue to improving the weakest links – which will often also be the scarcest.

I think some of those weak links are going to be aspects of “humanness”.

If we can find ways of improving or amplifying human judgement, decision-making and knowledge, rather than merely automating around them, we can strengthen some of the weakest links in the knowledge economy. Moreover, as AI-generated content becomes ubiquitous, I think simply being able to identify what is human and what is not will become increasingly important.

Second, knowledge work needs to fundamentally change its current paradigm if it is to fully benefit from AI.

A really interesting perspective on this was recently written by Jack Dorsey of Block and Roelof Botha of Sequoia. They argue that one reason organisations have layers of middle management is because information can only travel at certain velocities. Human communication creates bottlenecks, so companies construct hierarchies to route information up and down the organisation.

Yet much of the actual work and value creation happens at the edges of the organisation. Think about a law firm: ultimately it is the lawyers doing the legal work and billing the client, not the management hierarchy sitting above them.

Dorsey and Botha argue that, in an AI-enabled organisation, some kind of AI coordination or “intelligence” system could perform much of the information-routing function traditionally handled by management. Instead of humans spending their time passing information up and down a hierarchy, everyone can become closer to being a “doer”.

In other words, AI might not simply make today’s organisation more productive. It could fundamentally change what an organisation looks like.

They don’t really offer a complete solution, as they just posit the idea, but I think the implications are profound. Perhaps middle management, at least in its traditional form, is itself one of the weak links.

I don’t know what the future will hold, or what kind of growth curve we will ultimately be on. But I do know that AI is the next great economic horizon.

The question is whether humanity can remain part of that chain, not merely as the weak link, but as a necessary link.

I certainly hope so.

Lewis Liu is co-founder and CEO of Twin-1 AI