Apologies, this post is about AI

And also about a bicycle wheel.

So, screenshots of this bad post have been going around on Bluesky:

And there are actually some good thoughts coming from it – mainly from people who are actually going, okay, but what about the second or third order effects? Here's some I saw:

And, now I'm thinking about the front wheel of my bicycle. Let me talk about that for a while.

I bought this bike because I loved my other bike too much to lock it up on the street. But also I live up some stairs, and having to carry a bicycle down some stairs stops you from wanting to pop on your bicycle for a quick trip. And yet I live a slightly annoying distance from a tube station, and from the pottery studio, and from the nice cinema, and and. So: the solution is to buy a newer cheaper bike which I lock up on the street and which is easy and accessible by my front door.

And then of course the bike I bought was a lovely pink vintage Dutch bike. Cheap, secondhand, a little rusty - but also the right size and I love it.

And I've ridden it for a few years, but recently the front wheel has had an alarming amount of play. The rubber gaskets are brittle and failing. You can sometimes see some bearings... I think that's a bad sign?

And I know that the wheels are a weird size, because I've had to replace the tyres before. So rather than try to order some online, I go to the local bike shop down my road, and ask them for a wheel. The man takes a look, says "yep, you definitely need a new wheel". But I'm tired (no pun intended) when I buy it, so when the man insists that what I need is a 700c wheel, I buy the wheel and take it home to fit.

So: this was the task today. To fit the 700c wheel, prove that it doesn't actually fit, and then return that wheel and instead get a 28 x 1½ wheel, also known in ISO standards* as a 40-635 wheel.

And in a move that reaffirms my faith in local bike shops, the (other) man happily accepts the return, introduces himself in a very pleasant soft voice, and says I must need a 28 x 1½ wheel. He then says he'll check his stores to see if he has some spare second hand wheels of the right size, and call me in a few days to let me know.

So now I have a few different ways I have tried to get the right wheel. I have looked online, searching for "40-635 wheel". No good results. I searched for "635 wheel" – this got better results – some shops had it, all I think based in Europe (well they were priced in Euros), but generally sold out and with unknown shipping costs. A lead if this doesn't pan out. I also searched for "700b", based on a forum thread that said that that might be an equivalent. That search autocorrected to "700c" and returned useless results. I tried to search for "28 1 1/2 wheel", but Google didn't really parse the fraction in a useful way. For the purposes of this blog post, I tried Amazon directly, to no joy at all. I have tried to buy from a local shop, and they didn't have the right thing in stock. I have tried again at the local shop, and tapped into a possible source of a new wheel which is not visible to technological systems at all. A "dark pool" of bike parts, if I was being pretentious. And I have mentioned this whole situation to my flatmate, who works as a cycle courier, who made a note of the size and said he'd have an ask around.

And this feels kind of emblematic of the way difficult tasks get solved these days. There are computer systems, and they can solve common tasks at scale. For more specific tasks, they handle ambiguity badly, and they take patience and skill to extract the correct answers from. Filtering through listings, refining search terms, picking up on context clues to try to understand. The computer systems do contain good contextual knowledge, but in increasingly marginal spaces - and the motivation for creating this knowledge and making it public increasingly isn't worth the cost. And then there are the smaller scale human networks which you can tap into, if you can find them and be trusted by them, and those are inefficient but ultimately the most powerful ways of solving these problems.

So, I guess the way this wraps up is – in this postulated agentic AI future, do I think that an AI will be able to solve the problem of sourcing a replacement for the front wheel of my bicycle? No. I think that AI agents will make the task for searching for a rare and specific thing online harder and scammier. They will be optimised for the common case, and they will be ever more enmeshed in an influence arms race. I think that the most fruitful avenues for solving these problems are ones which are explicitly hostile to interacting with automated systems in this manner. I think that, to the extent that AI agents are able to access these systems, the systems will be ruined and useless for solving the problems they are currently equipped for solving. Right now I could call the bike shop and ask about a specific wheel size, and they would pick up and have a conversation. If the bike shop starts getting 50 calls a day from an almost human voice asking about random sizes of bicycle wheel, they will stop answering the phone. And if my flatmate starts getting those calls, he might give up on telephony all together.

(God – has thinking about AI too much made me an anarchist??)

[* also known as ETRTO standards. Bicycle wheel size labelling is a fantastic rabbithole]