The autonomous company

One way to think about it is to try to imagine what an autonomous company would look like.

Take two examples: the ancient windmill and Instagram. The mill is a partially autonomous company. It does one workflow, milling, on autopilot. Its engine is simple and comprehensible to a skilled human.

Instagram is also a partially autonomous company. Its engine serves content, ads, collects feedback, and auto-adjusts. This engine is so complex that humans have no way to fully understand or directly control what it does: what content or ads every user sees. They can only rely on the top-level metrics and some narrow quality checks. They can influence it only indirectly, by tweaking the algorithm and fine-tuning the underlying AI models.

Certain behavior of the engine must be restricted, like serving illegal content, so humans set up the guardrails. The key here is that they are simple enough that it's easy to understand their effects. And they sit outside the engine, so it can't alter them.

Now we have the building blocks of the autonomous company: code defining the deterministic actions, AI running the intelligent ones, metrics for visibility, and guardrails for control. The whole organization is one integrated engine, running 24/7, turning inputs into outputs, interconnected with other engines.

What about humans? Humans make autonomous companies unique. Humans define purpose - the 'why' and 'what' - and set taste - what's considered good. No two teams share the same purpose or taste, so no two organizations will be alike.

Every autonomous company has a collaboration medium, a 'desk', for humans to participate in the work if they want to. It allows people to review the company's work-products, communicate with the company in a human-native way, and make manual contributions.

People also build relationships with one another, both with customers and with colleagues at other autonomous companies.

Building and running these complex engines used to be expensive and required a lot of highly skilled engineers. Firms like Palantir have been doing this, but only the biggest organizations could afford it.

Advances in AI make it much easier to build and run company engines. AI agents' resourcefulness and flexibility help simplify the engine's design. AI coding accelerates setting it up. We can now deploy engines at a much wider scale, soon - at the scale of the whole economy.