An exobrain is only as good as the context you give it

In August 2022, NetDragon Websoft appointed an AI named Tang Yu as the general manager of one of its subsidiaries. A Chinese gaming company, not a startup looking for attention. In the following six months, its stock rose by around 10%, while the Hang Seng index dropped by 3%.

What makes this case more interesting than most AI examples is what it does not explain. Tang Yu was not making decisions in a vacuum: it operated inside a real company, with real operating data, years of product and market history, and a human team that had built something coherent enough to be handed to an external system.

The AI was the surface. The strategy was human.

The HustleGPT case

In March 2023, Jackson Greathouse Fall gave GPT-4 100 dollars and asked it to build a business. The project was called HustleGPT, and the broader challenge was later collected in a public HustleGPT repository. The model chose affiliate marketing for eco-friendly products, registered a domain, generated a logo, and allocated budget to social ads. Interest arrived immediately. The project attracted investment pledges at a theoretical valuation of around 25,000 dollars, but no actual sales were recorded. The concrete result: around 1,378 dollars in cash and commitments, mostly powered by hype.

The comparison with Tang Yu is not completely fair. One had a real company behind it, the other had a 100-dollar prompt. But that is exactly the point.

The gap between the two cases was not model capability. It was the context the model had available. Tang Yu operated on data built over time. HustleGPT had a prompt.

What we actually compete on

The idea of an exobrain comes from the extended mind theory in philosophy: the idea that cognitive processes do not stop at the edge of the skull. A notebook that externalizes memory becomes part of how we think. A calculator is not separate from reasoning, it extends it. The argument is that AI is a more powerful version of the same mechanism.

That lens is useful. It moves AI from “a tool I use sometimes” to “an extension of how I work.” But it opens a question the concept does not solve on its own.

Today, almost everyone has access to roughly the same AI capabilities. The models are available. The integrations are available. If the exobrain is the competitive advantage, and everyone has one, that advantage has already disappeared.

So what do we compete on? The quality of what we put inside it.

Not the company values page. Not the mission statement. The real reasoning: the hard choices you made, the things you learned about your market after years of work, the mental models that live inside people’s heads because nobody has written them down.

That is not evenly distributed.

The practical version

For a team, adopting AI without making its strategy legible produces generic output at high speed. A model can write a positioning document in a few minutes. If the input is a vague brief, the output is a vague document that sounds convincing.

The organizations that get more from this technology are not necessarily the ones that adopt it first. They are the ones with enough codified strategy to have something real to put inside it.

This is not a new problem. It is the same problem good knowledge management has always raised: making explicit what is implicit. The difference is that now the gap between implicit and explicit has a direct effect on the quality of everything your AI produces.

The practical answer is not to create a bigger prompt dump. It is to decide which knowledge deserves to become part of the system: positioning, constraints, product choices, source hierarchy, examples, and rules for when the assistant should refuse. That is why the exobrain connects directly to tools versus assets and to AI-assisted delivery. The same model can produce generic work or useful work depending on the strategic material around it.

Tang Yu outperformed a stock index. HustleGPT generated 1,378 dollars. Both ran on some of the best models available at the time. One had a real strategy behind it.

That is the useful benchmark for an exobrain: not how much text it can produce, but how much real judgment it can reuse.