AGI and your business: surf the wave or get swept under

At the strategy offsite, someone puts up a slide with an exponential curve and the word “AGI”. One director asks whether that means hiring fewer people next year. Another asks whether this is all a bubble. Nobody in the room can answer either question, and the discussion ends in a vague agreement to “keep watching”.

The scene is a composite, but it illustrates a common discussion. It starts from a framing error: treating artificial general intelligence as an event that will happen on a given day. For planning today, it matters more to observe available capabilities, test their limits on your own tasks and be ready to adopt those that prove useful.

A term with no date and no consensus

There's no single definition of AGI. OpenAI's 2018 charter describes it as “highly autonomous systems that outperform humans at most economically valuable work”. Other researchers define it more strictly or more loosely, and the dates mentioned range from a few years to several decades. Planning a business around a date nobody knows is bad strategy.

Recent models have expanded what is possible on some tasks, but no improvement transfers equally to every process. For a company, a model's performance, cost and risk on its own cases matter more than its label.

The lesson of electricity

In 1987, economist Robert Solow wrote a line that became famous: “You can see the computer age everywhere but in the productivity statistics.” Three years later, economic historian Paul David published an essay, “The Dynamo and the Computer”, that offered an explanation: the same had happened with electricity. Factories took decades to benefit, because at first they simply swapped the steam engine for an electric motor. The leap came when they redesigned the whole plant around the new technology.

General-purpose technology doesn't improve a company by being available. It improves it when the company reorganizes to use it.

Something similar is happening with AI. Plugging a chat into current operations is the equivalent of changing the engine without redesigning the factory. The value appears when the company organizes what a model needs to work: its data, its definitions, its rules and the permissions to act.

Who surfs and who gets swept under

When a more capable model appears, preparation changes the starting point. A company with defined data, documented rules and controlled interfaces can test sooner whether the new capability improves a process. Another will need to organize that knowledge before exposing it to the model. Both need quality, security and cost tests before going live.

Access to a model alone does not determine the outcome. Integration, permissions, oversight and the ability to measure whether operations actually improve matter too. That is where a more durable advantage can be built across model versions.

What preparing actually means

Preparing isn't betting on a vendor or guessing the future. It's building pieces that many business uses of AI need: traceable sources, shared definitions, written business rules, interfaces to query and act, permissions, records and human approval where appropriate. This season walks through each of those pieces.

That preparation can pay off even if AGI takes its time: defined data and documented processes also help teams review decisions today, with or without agents.

The scenes in these episodes are composites of common situations, not customer cases. Any figures in them are illustrative.