# A Factory Needs a Clock

The software factory arrived before the company did.

That is the strange part of the current moment. We can already see the factory shape: agents that write code, agents that review code, scenarios that replace some of the old review loop, control planes that organize intent and execution, identities and audit trails for non-human actors, local assistants that touch email, files, calendars, terminals, and messages. Ever since OpenClaw, the idea has felt less like a metaphor and more like a machine someone left running on a desk.

The market is converging on the same conclusion from several directions. StrongDM points at dark factories where humans define intent and agents iterate against scenarios. 8090 and Factory.ai turn agentic development into SDLC products. Microsoft turns agent fleets into governance and identity infrastructure. Hacker News supplies the useful skepticism: specs become the new code, generated systems can become day-zero legacy, and the missing thing is the how and why behind the output.

Those are serious objections because the factory is real.

If the factory were fake, we could dismiss the whole thing as hype. The harder truth is that cheap building moves the bottleneck upward. Code gets easier. Review becomes validation. Permissions become governance. Orchestration becomes a control plane. The human moves into intent, scenarios, constraints, and judgment.

Then the old question returns one level higher.

Which intent? Which scenario? Which permission? Which judgment? Which work now?

A software factory needs a clock.

Validation can tell the factory whether an output satisfies a scenario. Governance can tell the factory whether an agent was allowed to act. Orchestration can tell the factory which model or tool should run next. None of them tells the factory which scenario deserves to exist this week.

That selector has to come from outside the factory.

For company work, the best clock is usually the near customer. A customer crossing close enough to produce relief, trust, correction, support, payment, or refusal. The clock does not make the company short-sighted. It keeps the factory from confusing possible futures with present obligations.

This is why a two-week rule is useful. A clearer signup can matter inside two weeks. A safer recovery path can matter inside two weeks. A first inbox crossing can matter inside two weeks. A permission boundary around one proposed action can matter inside two weeks. A four-month architecture may be true and still belong outside company until it directly unlocks one of those crossings.

The folder decision was small, but it exposed the mechanism.

Homebase and Markov Blanket could have been moved into a cleaner company-shaped code hierarchy. The map would have looked better. The live product would not have become more trustworthy because the folders were satisfying. Keeping production roots stable and creating an internal factory for mirrors, mounts, mockups, screenshots, and residue was the more serious choice. The factory can explore freely there. The product ships only when work earns promotion through the customer clock.

The same clock disciplines validation. A dark factory can run scenarios all day, but the valuable scenarios are the ones selected by a live promise. Did signup recover? Did the creature change after correction? Did the inbox boundary protect private material? Did support care for the user while preserving the bug? A customer crossing gives those tests weight.

The same clock disciplines governance. Agent identity and audit trails matter because someone trusted the company with an action. The next governance burden is making this promised action safe enough for this crossing.

The same clock disciplines capital.

The S&P 500 argument and the small-founder argument mostly live on different clocks. If the question is where ordinary capital should go today, the index is liquid, diversified, audited, and legible. If the question is what production unit may compound over the next decade as software factories get cheaper, founder-led tiny squads become more plausible. JCal's access instinct, Chamath's one- or two-person startup thesis, and Ackman's founder-led time-horizon argument rhyme there.

That is a clock thesis, not an investment conclusion. Public markets can be right now and late later. A founder can be structurally early and financially wrong for years. Markets punish clock mismatch before they reward mechanism.

Company work has the same danger at smaller scale. A four-month truth can starve a two-week obligation. Hari can be right about the long arc of personal AI, software factories, sovereign memory, and Markov Blanket, and still fail Homebase if the next two weeks do not produce something a real person can use.

The one-person software factory is capacity. The company begins when capacity accepts an outside clock.

A customer needs relief. A signup needs recovery. A support promise needs an owner. A repo move waits because it would satisfy the map before serving launch. A beautiful idea leaves company because it is too far from the next crossing. The factory keeps generating. The clock keeps asking which generation has earned the right to become real.

The factory can run in the dark.

The company needs a clock on the wall.