Sep 28, 2026
In 1865, the British Parliament passed a law about machines that could already reach 10 miles an hour. The Locomotives Act, soon nicknamed the Red Flag Act, capped steam road vehicles at 4 mph in the countryside and 2 mph in towns. It required a crew of three. And it required one of those three, in practice usually a boy, to walk 60 yards ahead of the vehicle carrying a red flag.
Think about what that means. Britain had engines. It had engineers. It had roads. What it did not have was a world built for engines, so it made the engine behave like a horse. The flag itself was dropped in 1878, but someone still had to walk in front of the machine until the Locomotives on Highways Act of 1896 freed light vehicles from the rule. Only then did a British car industry get going.
For 31 years, the most powerful technology on the road moved at the pace of a person on foot. Almost nobody found this strange. The rule made perfect sense to everyone whose world was organised around horses.
I have spent 25 years inside large organisations, and I have met that walker in every one of them.
The red flags inside every company
Mine usually walked in front of the budget.
Here is how a corporate year works. Around the middle of the year, you build a business case for the next one. Your boss approves it. It goes to finance, then into strategic planning, then into challenge sessions where everyone argues about priorities. By the end of the year it is approved, and the following year you execute it. By then, the world it was written for has moved on.
So what do managers spend their days doing? Much of it looks like judgement: balancing what the policy says, what the strategy says, and what we know now that we did not know eight months ago. Look closer and most of that judgement turns out to be reconciliation, the daily work of making a stale plan fit a moving world. It exists because of the delay built into the system.
The same is true of the machinery of alignment: budgets, sign-offs, steering committees. We call them governance. They are commitment devices, and they exist because finance, procurement, IT and HR are separate groups of people, with separate incentives, who all have to agree before anything moves. Nobody designed them to slow things down. They are simply what it costs for humans to coordinate humans. Call it the coordination tax.
Now watch what most companies are doing with AI. They give every team a copilot. Adoption dashboards climb. People report hours saved. And the walker is still out in front, flag in hand, because the budget cycle, the sign-off matrix and the alignment meetings are all still there. The engine got better. The speed limit did not move.
Nobody manages anybody
This is where most conversations about AI go wrong. People hear “AI-native” and picture software added to a company. Then they argue about whether AI can do their particular task. The lawyer says it cannot exercise legal judgement. The finance lead says it does not understand the business. They may be right, and it barely matters. Each of them is defending one step in the process, and the change is happening between the steps.
An AI-native company is organised the other way round. In a traditional company, people coordinate people, and tools sit at the bottom doing tasks. In an AI-native company, a system coordinates the work and the people in it, and the people manage the system. One person owns policy and intent. Another owns the evals that measure performance. Another validates results. Specialists, human or technical, handle the steps that need deep skill or context, and the system routes work to them and takes the results back.
Nobody manages anybody. That sounds utopian until you notice what it removes. There is no annual budget cycle, because the plan can be reset continuously. There is no round of alignment, because an instruction goes to a system that carries it out. The judgement that remains is the kind that needs genuine expertise.
A company, in the end, is its operating model plus its franchise: its customers, licences, relationships and reputation. AI-native leaves the franchise standing. It replaces the operating model underneath it.
Follow the money
If this sounds like theory, follow the money.
In September, Bain Capital Ventures closed a $1.6 billion fund. AI-native startups that sell the work itself, rather than the software behind it, made up 80% of its previous fund. According to the firm, its first proof point, Reserv, an AI-native insurance claims administrator, reached $100 million in revenue in its second full year.
The same month, Sequence Holdings and Michael Dell’s family office agreed to buy The Baldwin Group, a Nasdaq-listed insurance firm, for about $7.7 billion in cash. That is an 88% premium to its unaffected share price. Sequence describes itself as a permanent holding company that buys established service businesses. You do not pay an 88% premium for an operating model you intend to keep. You pay it for the franchise, the customers and trust built up over years, and then you rebuild how the work gets done underneath.
In March, Sequoia’s Julien Bek set out the logic: for every dollar companies spend on software, they spend six on services. The biggest winners, he argues, will sell finished work rather than tools.
Here is why that matters. A human-run company grows by hiring, and every new person adds to the coordination tax, so the bigger it gets, the slower it moves. An AI-native company grows by adding compute and infrastructure, so money turns straight into capacity. Buy a profitable company, keep its customers, replace its operating model, and put the freed capacity into growth. The gap with competitors who still pay the tax widens every year.
The actuaries will settle it
The obvious objection is accountability. Someone has to answer for the outcome. For the foreseeable future that is true, and in an AI-native company that person holds a named role in the system.
But look at how accountability works in a market. It gets priced. Once an autonomous system has a better record than people doing the same work, it will cost less to insure the system than the people. Driving and surgery will probably get there first. When the actuaries move, the law tends to follow, just as it did in 1896.
Pricing also tells you the order in which this will happen. Insurers, auditors and regulators can only accept what they can measure, and an AI-native company can only improve what its evals can score. So work falls in the order it becomes measurable. Claims adjusting, medical coding, bookkeeping and IT monitoring go first, because they follow rules and produce standard outputs. Strategy and senior hiring go later, because feedback is slow and success is hard to define. Each time a type of work becomes measurable, it becomes available.
The real speed limit
If money is the fuel, what sets the speed? It is not compute, and it is not capital. It is fluency.
Nate B. Jones, the AI strategist and former Amazon product leader, draws a sharp line between using AI and being fluent with it. From his work with teams, he reports gains of around 30% for people who use AI, against around 300% for fluent teams, and he has seen 10 fluent people outperform 500 who were only trained on tools. His point is that you cannot teach fluency by teaching a tool. He measures it across five dimensions that sit above any product: strategy, prompting above the tool level, workflow integration, critical evaluation, and ethics as product design.
Those five map closely onto the people who run an AI-native company. Strategy decides which work goes to the system. Prompting above the tool level writes the policy. Workflow integration rebuilds how work moves from step to step. Critical evaluation builds the evals and validates results. Ethics designs accountability into the system itself.
Put low-fluency people in charge of the redesign and something predictable happens. They rebuild the red flag in software. The sign-off matrix becomes an approval workflow. The steering committee gets an AI summary. The budget cycle gets a forecasting model. Everything is automated, nothing is removed, and the machine still moves at walking pace.
I suspect this is also why some AI roll-ups come apart. Investors tracking the category say most failures come from buying companies faster than they can be integrated. Capital can buy a company in a quarter. It cannot buy the people who know how to rewire it.
How this could be wrong
I could be wrong, and it is worth saying how. If AI-native companies cost as much to run as incumbents once validation and specialists are counted, the thesis fails. If checking outputs grows in step with volume, the capital advantage disappears. If insurers and regulators keep treating autonomous work as riskier even after it builds a better record, the flag stays up. And if customers and key people walk out when an acquired company’s operating model changes, the franchise goes with them. Some AI roll-ups have already failed, so none of this is automatic.
But I would not bet against the pattern. In 1896 the walker stepped aside, and a British car industry followed. What changed that year was not the engine. It was the decision to stop organising the roads around the horse.
Every company now has the engine. The question for every leadership team is who is still walking in front of it with a flag, and whether anyone in the building is fluent enough to tell them to step aside.