On Friday I had to nip into town and do a run of mildly annoying admin. I gave GPT-5.6 a three-minute voice prompt as I walked, explaining the whys and wherefores. It saved me about forty-five minutes. A procedure at the bank had changed, which I would otherwise have discovered by joining the wrong queue and failing.

A little later I walked out of Boots with a pharmacy product a model had suggested asking the pharmacist about. None of this felt especially futuristic. I was still the one walking around, speaking to people, showing ID and carrying the bag home. We are obviously still in the useful tools stage, although I’m becoming less sure what that phrase excludes.

At work it already feels professionally irresponsible to design or implement software without wheeling out a frontier model. Increasingly I don’t arrive with a plan for it to execute. I describe the situation, ask what it would do and pull on whichever thread looks best. A post that caught my eye described the same shift, from telling agents what to do to asking them.

What looks like a small change in prompting moves more of the planning to the other side of the conversation.

Hayek’s account of the price system is partly an account of people being coordinated by something they cannot see. A price changes and thousands alter what they buy, make or substitute, usually without knowing why. The system combines their local knowledge without giving any one of them a view of the whole.

Another post described meeting a man in a Seoul dive bar who had devoted his life to micro-optimising the fifteenth most important inefficiency in a superconducting supply chain. It is funny because the economy really does produce people like this, dropping them into absurdly specific problems that matter to a system nobody can see whole.

Nick Land’s more feverish version treats capitalism itself as an emergent intelligence, with machine intelligence as its continuation rather than a separate invention. I’ve never known how literally to take that. It is easier to see the attraction now that a model can absorb a few minutes of context, form a better plan than I would have formed alone, then send me around town to execute the physical bits. It did not need a robot because I was still handling that layer.

This sounds more dramatic than it felt at the time. There was no point at which I handed over control. Every suggestion was optional, local and useful. I checked the bank advice, and we still had to decide whether the product from Boots made sense.

At work I reject agent suggestions constantly. The ratio just keeps moving, and every few months there is a little less reason to begin from my own plan.

A friend wondered whether we will miss the singularity, in the sense that it will happen next year while we are still wandering around chatting about it. The usual picture gives us something obvious to notice, with a model crossing some threshold and taking command. A distributed transition would look much duller from inside. Millions of people ask first, choose among generated plans, then carry out the parts that still require hands, accounts, signatures and physical presence.

Humans remain useful throughout this. We look a bit like bees moving pollen between flowers, doing real work and responding intelligently to what is nearby without needing a theory of the field. The shift from useful tool to something more agentic is scalar from here, rather than a clean change of state. The early sign may simply be that asking the machine becomes the sensible first move, until ignoring its advice is the choice that needs explaining.

I wrote in January that crossing a horizon need not feel unusual locally. Missing the singularity may be the same thing. We live through each reasonable step and keep waiting for the global event.