Stuart Winter-Tear

<start of post>

I’ve been thinking about the difference between 'using AI' and 'being an AI-augmented organisation'.

Most people are currently stuck in the former. They treat the LLM like a faster intern—a way to get the same tasks done in less time. But that’s just efficiency. It isn't transformation.

The real shift happens when you stop looking at the tool and start looking at the workflow.

If a system can now perform the synthesis, the drafting, and the initial critique, why is the process still designed around a human doing those things? We are essentially putting a jet engine on a horse-drawn carriage and wondering why the wheels are falling off.

It’s an operating model problem.

I caught a glimpse of this in a recent paper on 'Agentic Workflows'. The researchers found that when you give an agent the ability to iterate on its own work, the quality doesn't just improve—it compounds.

But here is the catch.

To make that work in a company, you have to change who has the authority to sign off on the 'first draft'. If the human still has to check every single step, you’ve just created a new bottleneck. You’ve traded a production problem for a supervision problem.

How much 'unsupervised intelligence' can an organisation actually tolerate before the governance structure breaks?

Perhaps the goal isn't to automate the work we do now. Perhaps the goal is to design work that only becomes possible when intelligence is abundant and cheap.

Told you it was going to be a big deal.

Give it a think.

<end of post>