Roads, bridges and the traffic that never existed

I recently wrote about how AI agents are like the car in 1908: on a race track a private driver could hit top speed, but on muddy roads congested with horses, walking pace was common.¹ There I argued that individuals can already go fast with AI agents, but organisations need to build and re-organise their roads to capture the productivity benefits.

The analogy lands well both online and in my client conversations, so I want to extend it. What type of roads does an organisation need to build, and what happens to the traffic?

I disagree with the zero-sum framing of AI: the idea that because it can do some of what we do today, there will be less work overall. How many of today's journeys are ones people would once have taken by horse? I think people will find an endless supply of new intellectual journeys to take at the speed of AI agents, and our role is to enable them.

Motorways and interstates carry vastly more journeys than horses ever did, because once a journey is fast and cheap people make far more of them (Jevons' paradox).² Paving the road induces its own demand. Go further, and a bridge or a car ferry opens up journeys inland that no horse could ever reach.

In general terms, reducing the cost of a journey, intellectual or physical, does three things:

1. Substitution. The journeys previously taken by horse are made by car or bike today. This is zero-sum, and it is a small portion of the eventual market the new technology enables.
2. Induction. People take more journeys than before because each one is cheap. This is positive-sum, and often the largest effect. It is the motorway commute, and it is Jevons' paradox.
3. Extension. Across the water, entirely new destinations open that no route reached before. Also positive-sum, and the growth that displaces no incumbent.

Only substitution is zero-sum. Induction and extension are incremental journeys. The incumbents own today's mud roads: they need to pave them to stay relevant, or risk a startup laying a new, faster, parallel road. The motorway and the interstate gave rise to suburbs, logistics hubs and modern retail, extensions that were never possible with horses.³

Why the growth is hard to see

Put the three effects together and the puzzle, that AI "has yet to clearly show up as an economic growth driver", resolves without having to pretend the growth isn't real.⁴

Substitution shows up first and loudest, because it has a victim: a visible incumbent losing a visible margin. It is also the only one of the three that is zero-sum, so when commentators tally it up they see redistribution and conclude little is being created.

Induction and extension are where the growth actually is, and both are slow to register. Induction waits on the roadworks: the corridor only carries its multiplied traffic once the workflow around the agent is rebuilt, the J-curve where productivity dips before it climbs.⁴ Extension is revenue in markets too new to be measured. The tutoring the family never bought and the legal help the tenant never sought show up as a loss nowhere, and barely yet as a gain.⁵

The car is the precedent, and it was emphatically not zero-sum. Its contribution was never the horse's journey done faster. It was the explosion of total travel along the paved corridors (induction) and the suburbs, logistics and retail the road made reachable (extension). The horse economy, the farriers and coachmen and stable hands it displaced, was the substitution story, and on its own it reads as pure loss. The road economy that ended up employing far more people than the horse ever had was induction and extension, and it took decades both to build and to count.⁶

A framework: reading the road

1. Decompose the demand: substitution, induction, extension. Which existing journeys will simply move to a cheaper provider (substitution, your margin at risk)? How much more of the journey gets made once it is cheap (induction, the growth on your own road)? And what was never done at all because the old cost priced it out (extension, the water)?

2. Locate your moat, and check whether it is road-shaped. Defensibility made of implementation, integration or process complexity is road-shaped: it is the friction AI dissolves, and it will be paved away.⁷ Defensibility made of regulation, brand, proprietary data, accountable judgement or genuine switching cost is not road-shaped, and it endures.

3. Read four dials for the speed of change. Where the moat lives (road-shaped is fast; regulated or brand is slow). Cost-structure inversion (fixed assets bolted to the old channel die fastest, as the new road turns them into liabilities).⁷ Distribution piggyback (if the capability ships inside a product customers already trust, there is no moment at which you can compete for the sale). Procurement friction (if you still own the buying relationship, you keep control of the pace).

4. Name the unbundling target. Find the profitable, repeatable, scalable slice that subsidises everything else: the classified ad, the associate's research memo, the standard audit. That is what gets pulled out and run at near-zero cost first.

5. Find your water. Ask what your cost structure has always priced out. Who never became a customer because the bespoke, hourly, expert version was too dear? That non-consuming market is the extension, the bridge no incumbent is defending. It is often larger than the market you serve today, and it is where the additive growth is.

6. Choose the move, and it is rarely just defence. Defend the moat that is not road-shaped. Let the unbundled slice go without a fight, because bans and detection buy less time every year.⁸ Pave your own road to capture the induced demand, a larger prize than the slice you are losing. Re-price from hours to outcomes, so induction shows up as revenue rather than lost margin. And decide, deliberately, whether you build the bridge to the water or cede the far bank to whoever does while you are busy defending a margin on the near side.

The road-building clock and the displacement clock are the same clock seen from both sides: every quarter you spend waiting for a profit the mud road was never going to produce is a quarter you hand to whoever is laying tarmac. But the bridge clock is a different, more hopeful instrument, and it is the one that actually grows the economy. Whoever builds the road takes the traffic, and there turns out to be far more of it than the mud road ever carried. Whoever builds the bridge gets the traffic that never existed at all.