The Everything Factory

Goldman Sachs didn't bury the headline. Their recent report about automating accounting laid out what many executives still treat as a distant forecast: a fundamental restructuring of how economic value gets produced. Not an upgrade cycle or a platform shift. A replacement of the production model itself.

Frontier AI labs — Anthropic, OpenAI, Google DeepMind, xAI — are building toward the same target. They are constructing scaffolding around their models designed to perform every task every knowledge worker can do. That is not a product roadmap, it's a declaration of intent with vast economic consequences.

And they are starting with software for reasons that should concern every business leader, not just CTOs.

The Beachhead

Software development is the first industry in the crosshairs, and the choice is not accidental. It’s strategic, in the same way a military campaign selects its landing point not for the beach itself, but for what the beach gives access to.

Software is an extreme knowledge task that is also verifiable in a fully automated way. Write the code, run the tests, measure the output. The machine knows whether it worked without a human reviewing the result. No other knowledge domain offers that closed feedback loop at the same scale.

Automating software development also feeds the labs' own research directly. Better code-writing models produce better scaffolding, which produces better models. The loop is self-reinforcing — each generation of tooling accelerates the next.

Most importantly, software is the substrate of every other knowledge industry. Automate the ability to build software and you unlock the ability to automate legal analysis, financial modeling, medical documentation, logistics planning, marketing strategy — every domain that runs on information and rules. Software is not one industry among many. It is the master key to all of them.

The goal was never a copilot, it was always a factory. But this factory builds knowledge workers not widgets.

From Assembly Line to Learning System

"I don't write any code anymore. I let the model write it — I just edit." — Dario Amodei, CEO of Anthropic

Amodei is not describing a productivity tool. He is describing a shift in the relationship between the worker and the work — the same shift that has reshaped physical manufacturing twice in the last century.

The first transformation was Henry Ford's assembly line. Ford did not invent the car. He invented a system for producing cars; and that system dropped production time from over twelve hours to ninety-three minutes. The resistance was enormous. Competitors dismissed it. Craftsmen resented it. But within a decade, every serious manufacturer had adopted the model or vanished.

The second transformation was subtler and, ultimately, more powerful. Toyota's production system did not just build cars faster. It built a factory that learned. Every worker on the line had the authority to stop production when they spotted a defect. Every stoppage became data. Every piece of data fed back into the system's design. Toyota's real product was not the car. It was the continuously improving process that produced the car.

That is the model the frontier labs are now replicating in software. Not tools that help developers write code, but systems that write, test, debug, and improve code — then use that improved code to make themselves better at writing code. Recursive development. The factory that redesigns the factory.

In Fremont, Elon Musk is pushing toward the physical version with Tesla's Optimus humanoid robots — machines that grab parts, adapt after failure, and feed performance data back into the next iteration. Musk and xAI have also launched Macrohard, a joint venture whose stated ambition is not productivity tooling but full replacement of any company whose product is information. Internally, xAI has already onboarded AI agents as named employees; listed on the org chart, receiving messages, attending calendar invites. A meaningful number of xAI's human employees reportedly did not realize some of their coworkers were not people.

The pattern is the same in every case. The work is not being assisted. The work is being absorbed into the system that produces it.

The SaaS Dependency Trap

For two decades, SaaS felt like the natural order. Software moved to the cloud, updates became continuous, and subscription revenue replaced license fees. The model had real structural advantages: near-zero distribution cost, centralized maintenance, and predictable revenue streams that Wall Street rewarded handsomely.

But SaaS scales linearly. Humans still maintain the systems. Vendors still control the improvement cycle. Customers still wait in roadmap queues for features they need now. Strip away the cloud infrastructure and what remains is an old arrangement in new packaging: vendor dependency.

The hard question — the one most SaaS customers have not yet asked — is this: if models can generate, refactor, and optimize software systems themselves, why would you architect your business around someone else's static platform?

The adoption curve for this shift looks like every major industrial transition. Long, flat progress that lulls incumbents into complacency. Then a vertical climb that rewards the prepared and punishes everyone else. Miss the inflection point and you do not fall behind. You disappear, like the coach builders in Ford’s revolution.

Your Factory or Theirs

This is not a problem reserved for Fortune 500 technology companies. If you run a mid-size firm, a consultancy, a services business, a financial institution, an oil & gas firm — any organization whose value depends on what your people know and how they apply it — you are directly exposed.

But here is the strategic insight that matters most: the frontier models themselves are a commodity. They will continue to improve. They will continue to compete with each other on price and capability. Treating any single model as your strategic foundation is like treating the electricity grid as your competitive advantage — it powers everything, and it is available to everyone.

Your data, your processes, your customer relationships, and your institutional knowledge; these are the raw materials that no lab can replicate. But raw materials sitting in a warehouse are not a business. They become a business only when you build a factory around them.

If you don’t build that factory, someone else will. And when they do, the economics are simple: you become a distributor or a reseller. The value: the margin, the insight, the competitive position accrues to whoever owns the production system, not to whoever feeds it inputs.

Build Your Factory

The Toyota analogy points to something specific about how you should act.

Toyota did not start by replacing its entire production system overnight. It started by giving one worker on one line the authority to pull a cord and stop production when something was wrong. That single change — a small feedback loop inserted into an existing process — transformed the company over decades.

Start the same way. Deploy an agent that adjusts pricing overnight based on market signals. Let it fail. Study why. Improve the logic. Redeploy. That cycle: build, test, fail, learn, rebuild. The cycle is the goal, it’s the factory beginning to learn.

Don’t spend the next year evaluating chatbot vendors. That is the horse-drawn carriage watching the Model T drive past.

Keep your model providers interchangeable. Build your intelligence layer — your agents, your pipelines, your decision systems — around the context and customers only you own. The model is the ingredient. Your factory is the product.

Phase2_ has been building exactly this for the organizations we partner with. Not theoretical AI strategy. Actual systems — agents, pipelines, and intelligence layers designed around your data, your processes, and your people.

The underscore is not decorative. It is a prompt. What comes after it is yours to build.

‍