We’ll see lots of people fired - not because an AI can do their job, but because an AI salesman can convince their credulous bosses to fire them and replace them with an AI that can’t do their job.
Cory Doctorow, The Reverse Centaur’s Guide to Life After AI.
Two ideas smashed themselves together in my head yesterday whilst attending the AWS Community Summit in Manchester (which is a top notch conference by the way). The ideas came from two brilliant people, Cory Doctorow and Dave Anderson.
Let’s start with Cory. I’ve recently read Enshittification, and then followed that straight up with his new book The Reverse Centaur’s Guide to Life After AI. I’d highly recommend both books. The title comes from an idea Cory uses a lot. A centaur is a human assisted by a machine: the human is in charge, the machine makes them better. A reverse centaur is the opposite: the machine is in charge, and the human is its squishy appendage, doing whatever the machine can’t (and taking the blame when it gets things wrong). Cory’s go-to example is the Amazon delivery driver. The machine tracks their every move: their route, how fast they need to deliver each package, and when they get to use the toilet (not that often). The human is fully controlled by the whims of the machine.
There is another point he refers to quite a few times, related to the opening quote of this article. The AI companies do not care if we think the tools are good or bad. Our opinions, as practitioners, are largely irrelevant. If one of us shitposts, says something bad, or loves it, it doesn’t really matter.
The main thing AI companies need to do is convince company executives that the tools they have built can do our jobs. And this is a subtle but important distinction. With the massive amounts of investment that have gone into AI, expertly covered by Ed Zitron, the labs need to return enormous amounts of money to even have a sniff at being profitable. You and I with our monthly Claude subscriptions aren’t going to cut it. For the finances to work out, they need some big enterprises to pay them an awful lot of money. Now most big enterprises I’ve ever come across aren’t suddenly going to start shelling out millions and millions of dollars because they feel like it. They need a return. Unfortunately for us, that return is a lower wage bill.
So it really doesn’t matter if the AI can do our jobs, only that enough executives believe it can.
Whenever I see any new AI feature, the cynic in me always asks: what’s the play? At one point, running agents in loops was all the rage, which seems like an awfully convenient thing for an AI lab to convince you to do when their billing model is based on how many tokens you consume.
Cory’s idea smashed up against the last part of the closing keynote yesterday in Manchester. The brilliant Dave Anderson briefly talked about the new crazy that’s sweeping the AI maximalists, and that’s the idea of a software factory. A quick Google Kagi search led me to this article by Cortex describing a software factory as:
An AI software factory is an organizational system that turns customer needs into shipped, reliable software, with AI agents doing the building while engineers move up a level to design and operate the system itself.
On the surface, it sounds appealing. The factory runs constantly, churning out features and bug fixes at a rapid pace. But it quietly assumes that high output means good output, and anyone who’s maintained a codebase knows that isn’t true.
Take that idea, alongside Cory’s thoughts around who needs convincing, and this becomes a little more dangerous. The factory analogy screams of the industrial revolution. A point in history where what was, at the time, highly skilled labour was replaced by machines. When a focus on craftsmanship, quality and value was replaced by a focus on output. On churning out as many widgets as possible with little regard for the people in the factory. Factory owners got rich, could extract more value, whilst the workers became replaceable cogs in a machine.
It’s worth remembering that the Luddites, many of them not far from where I was sat in Manchester, weren’t anti-technology. They were skilled workers who objected to how machines were being used: to drive down wages and quality, and to replace them with cheaper labour. Sound familiar?
A software factory doesn’t turn engineers into centaurs. It turns them into reverse centaurs: reviewing whatever the agents produce, at the pace the agents produce it, and then being the ones blamed when it breaks in production. If there’s one thing humans are bad at, it’s reviewing output over and over again and getting it right 100% of the time. Eventually, you just accept the machine’s output and get on with your day.
What better way is there to convince executives that the highly skilled labour they employ (engineers) can be replaced by machines that can shit out as many widgets as possible, than to start talking to them about building factories?
It doesn’t have to be this way, of course. Dave followed up the factory analogy with that of a garden. A garden is something you tend to, you cultivate, you look after for the long haul. As the gardener, your job is fulfilling. You tend to it with care, and at the end of it with some love and attention you end up with something that will stand the test of time. Your focus is on the process, as much as it is the output.
The pushback to factories should be that of gardens. That’s not to say gardeners don’t use tools to make their jobs easier; technology moves on in gardening as much as it does in software. But the tools they use extend the work the human is doing to make them better/faster. The tools they use aren’t completely replacing them and tending to the garden without the gardener.
What does gardening software look like in practice? For me, it’s using AI to take care of the weeding, the boilerplate and the scaffolding. To be honest, I think the physical act of writing lines of code is probably dead anyway. But writing code was only ever part of the job. Offloading all of that to an AI means I can spend my time on the parts that need a human: understanding the system, making trade-offs, and deciding what shouldn’t be built. It’s also owning the code that ships, because I’ll be the one living with it. And most importantly, it’s growing the next generation of gardeners. Gardens get passed on. Factories only need operators.
Some of us will have very little control over the way AI tools are imposed upon us. Some of us will have bosses who believe the hype, and believe that AI can replace software engineers. That AI can replace the need for junior engineers who we nurture and train. I disagree with this sentiment, but I can see how it would sound like a nice story if I was in leadership.
Whenever I think about this, I think about a meme/quote I saw really early on when the AI bubble was just starting to take off. It was along the lines of ‘if your boss fires you and replaces you with an AI, don’t worry too much, because in 12 months you’ll be able to come back as a consultant to clean up the mess the AI has made, and charge 10x as much for it’.
It’s a nice thought, and there’s probably some truth to it. But it only works if there are still experienced engineers around to call. Fire the juniors today, and in ten years there’s nobody left to clean up the mess.
Nobody can predict how this is all going to play out; anyone who thinks they can probably has some kind of vested interest in the outcome. What we can control is how we work. Use the tools. Tend the garden. Grow the next gardeners. And don’t let anyone sell you a factory.
James