The Two Clocks

When the rate of change inside an institution becomes slower than the rate of change outside, the end is in sight.

Jack Welch said that decades ago, watching industrial conglomerates ossify around him. He meant it as a warning to managers who confused the stability of their hierarchy with the stability of their world. It was a useful warning then. It is a much more dangerous one now.

The external clock

The external clock has stepped up by an order of magnitude. The cost of producing a competent first draft of almost anything, a market analysis, a working prototype, a customer response, a campaign, has fallen by two orders of magnitude in three years. A competitor who used to need a quarter to ship a feature now needs an afternoon. A customer who used to wait six months for a response now expects six seconds. Categories of work that defined entire functions are being repriced continuously, and the repricing does not stop between quarters to let anyone catch up.

This is not a wave that crests and recedes. The rate itself has changed.

The internal clock

Now look inside the average enterprise.

Strategy lives in slide decks. OKRs cascade through three layers of managers before anyone touches a keyboard. Quarterly planning cycles assume the world will hold still long enough to execute on the previous one. The state of the company, what it knows, what it has decided, what it is doing right now, is not stored anywhere coherent. It is distributed across email threads, Notion pages, individual heads, and the tribal memory of people who may have left six months ago.

This is not because the people inside are slow or stupid. The internal clock is structurally slow. It is built out of human coordination overhead, and human coordination overhead has a floor measured in days and weeks. Some of that latency is even useful. Slow institutions are how civilizations protect themselves from bad decisions made quickly. The problem is not the absolute speed of the internal clock. The problem is that its architecture has not meaningfully changed since the postwar era, while the external clock has accelerated by an order of magnitude.

The deeper problem: the company cannot read itself

The widening gap between these two clocks is the foundational source of enterprise risk today. Not AI disruption. Not geopolitical fragmentation. Not regulation. Not even competition in any classical sense. The latency of your own organization.

And here is the part most discussions miss.

The reason the internal clock is slow is not that decisions take too long. It is that the company has no coherent representation of itself. It cannot read its own state. It cannot tell you, in any structured form, what it currently believes about its market, what it has committed to, what work is in flight, what was learned from the last quarter, or what the operating consequences of a new decision would be. All of that exists, but only as scattered artifacts in human heads and human-readable documents.

You cannot operate at machine speed against a substrate that only humans can parse. Agents need state to act against. Loops need state to close. Without a machine-readable representation of the company, every AI initiative degrades into a faster way to produce more documents that no system can read.

This is why most AI initiatives in large companies produce so little. Leaders are buying copilots, appointing Chief AI Officers, running pilots, layering tools on top of an organization that still cannot read itself. None of it changes the clock speed of the institution. A genius locked inside a bureaucracy is just an expensive bureaucrat.

The error is categorical. Leaders are treating AI as a productivity input, when the real question is whether their organization is the kind of object that can metabolize AI at all.

The shift required is not more AI. It is computational legibility. The company has to become an object that can read and write to itself in something other than meetings.

What that looks like, concretely

Consider an ordinary decision. A competitor launches a new product on Tuesday morning, and the company needs to respond.

Today: someone notices on Slack. A manager calls a meeting for Thursday. The meeting produces an action item. The action item is assigned to a team that is already overloaded. The team books a workshop for the following week. The workshop produces a slide deck. The slide deck goes up the chain. Two weeks later, a decision is made. Three weeks after that, something ships. By then the competitor has launched two more things and the original analysis is stale.

In the version that works: the launch is detected automatically and written into the company's market state. Existing positioning, current bets, and recent customer signals are already represented in machine-readable form, so the first-order implications are computed in minutes. A response brief is drafted by an agent operating against that state, surfaced to the relevant human for direction-setting and ambiguity resolution. The human pushes back, narrows the scope, kills two of the four proposed angles, and sets intent on the remaining one. From there the work is decomposed and dispatched across the agents and humans best suited to execute it. Progress is written back into shared state continuously. Each step, from decision through execution and evaluation, advances against the same substrate, without requiring a meeting to move.

It is not frictionless. The human is still the bottleneck on judgment, and judgment is still slow. But the difference is not that the same activity happens faster. It is that an entirely different category of activity becomes possible. The org chart is no longer a hierarchy of approvals. It is a working graph of people and agents operating against shared context.

The honest part

This is hard, in ways that the current discourse around AI adoption almost completely fails to capture.

It is hard because every existing business process is encoded in slide decks and tribal memory rather than in machine-readable state, and translating one to the other is the work of years, not quarters. It is hard because the people who hold the most context are also the people most threatened by encoding that context. It is hard because the governance models we have for organizations were designed for a world where decisions were rare and expensive, not for a world where the system can take ten thousand of them an hour. It is hard because we genuinely do not yet know what the right balance is between automated action and human judgment, and we will only find out by running the experiment.

But the difficulty of the transition does not slow down the external clock. The external clock does not care about your change management plan.

Where this ends

Not every company whose internal clock lags will die. Some are protected by regulation, distribution, capital structure, or switching costs deep enough to absorb a decade of accumulated latency. Welch was not literally correct about every conglomerate.

But the protection is finite, and the latency compounds. Each year the gap widens, the cost of closing it grows, and the set of companies whose substrate is fast enough to even attempt the transition shrinks. At some point, the gap stops being a competitive disadvantage and starts being a structural impossibility, the way a 1995 mainframe cannot become a cloud-native system no matter how much money you throw at it. The architecture decides the ceiling.

The companies that come out the other side of this decade will not be the ones with the most AI, the most data, or the most impressive partnership announcements. They will be the ones that did the slow, unglamorous, deeply political work of becoming legible to themselves, building a substrate where the company's state, intent, and execution all live in the same continuous loop.

Everyone else will spend the next ten years looking fine from the inside. The hierarchy will be intact. The meetings will be well-attended. The strategy deck will be freshly updated. And outside, the world will be moving in a way that the internal clock can no longer track.

That is the timeline Welch was describing.

It is already running.