Today I stumbled upon Don't paste the AI:
When someone asks you something, they want your answer. Not a wall of ChatGPT text. A short answer from a human will always be worth more than one from a robot, even if the robot is right.
This quote from the article really sums it up:
And to be clear, I don’t care whether you typed the code yourself. I care whether you understood it before you shipped it. I care whether you can explain why the bug happened, why this fix is the right fix, what the model might have missed, and what would make you roll it back.
This quote on the abstraction tower really stood out for me:
I saw someone on LinkedIn recently — early twenties, a few years into their career — lamenting that with AI they “didn’t really know what was going on anymore.” And I thought: mate, you were already so far up the abstraction chain you didn’t even realise you were teetering on top of a wobbly Jenga tower.
They’re writing TypeScript that compiles to JavaScript that runs in a V8 engine written in C++ that’s making system calls to an OS kernel that’s scheduling threads across cores they’ve never thought about, hitting RAM through a memory controller with caching layers they couldn’t diagram, all while npm pulls in 400 packages they’ve never read a line of.
But sure. AI is the moment they lost track of what’s happening.
The abstraction ship sailed decades ago. We just didn’t notice because each layer arrived gradually enough that we could pretend we still understood the whole stack. AI is just the layer that made the pretence impossible to maintain.
Instead, most organisations don’t tackle technical debt until it causes an operational meltdown. At that point, they end up allocating 30–40% of their budget to massive emergency transformation programmes—double the recommended preventive investment.
I can very much relate to this statement. Many contracts I've worked on in the last few years, have been transformation programmes, where an existing product is rewritten and replatformed, often because of the level of tech debt in the legacy system.
I am not surprised that there are parallels between vibe coding and gambling:
With vibe coding, people often report not realizing until hours, weeks, or even months later whether the code produced is any good. They find new bugs or they can’t make simple modifications; the program crashes in unexpected ways. Moreover, the signs of how hard the AI coding agent is working and the quantities of code produced often seem like short-term indicators of productivity. These can trigger the same feelings as the celebratory noises from the multiline slot machine.
I think the most interesting, and also concerning, point is the eighth point, that people may become busier than ever.
After guiding way too many hobby projects through Claude Code over the past two months, I’m starting to think that most people won’t become unemployed due to AI—they will become busier than ever. Power tools allow more work to be done in less time, and the economy will demand more productivity to match.
Consider the advent of the steam shovel, which allowed humans to dig holes faster than a team using hand shovels. It made existing projects faster and new projects possible. But think about the human operator of the steam shovel. Suddenly, we had a tireless tool that could work 24 hours a day if fueled up and maintained properly, while the human piloting it would need to eat, sleep, and rest.
In fact, we may end up needing new protections for human knowledge workers using these tireless information engines to implement their ideas, much as unions rose as a response to industrial production lines over 100 years ago. Humans need rest, even when machines don’t.
This does sound very much like what Cory Doctorow refers to as a reverse-centaur, where the developer's responsibility becomes overseeing the AI tool.
Thank you! I've added the image to the post as well.
I use AI coding tools, and I often find them quite useful, but I completely agree with this statement:
And if you think of LLMs as an extra teammate, there's no fun in managing them either. Nurturing the personal growth of an LLM is an obvious waste of time.^___^
At first I found AI coding tools like a junior developer, in that it will keep trying to solve the problem, and never give up or grow frustrated. However, I can't teach an LLM, yes I can give it guard rails and detailed prompts, but it can't learn in the same way a teammate can. It will always require supervision and review of its output. Whereas, I can teach a teammate new or different ways to do things, and over time their skills and knowledge will grow, as will my trust in them.
My understanding of how this relates to Jevons paradox, is because it had been believed that advances in tooling would mean that companies could lower their headcount, because developers would become more efficient, however it has the opposite effect:
Every abstraction layer - from assembly to C to Python to frameworks to low-code - followed the same pattern. Each one was supposed to mean we’d need fewer developers. Each one instead enabled us to build more software.
The meta-point here is that we keep making the same prediction error. Every time we make something more efficient, we predict it will mean less of that thing. But efficiency improvements don’t reduce demand - they reveal latent demand that was previously uneconomic to address. Coal. Computing. Cloud infrastructure. And now, knowledge work.
Based on my own experience of using Claude for AI coding, and using the Whisper model on my phone for dictation, for the most part AI tools can be very useful. Yet there is nearly always mistakes, even if they are quite minor at times, which is why I am sceptical of AI taking my job.
Perhaps the biggest reason AI won't take my job is it has no accountability. For example, if an AI coding tool introduces a major bug into the codebase, I doubt you'd be able to make OpenAI or Anthropic accountable. However if you have a human developer supervising it, that person is very much accountable. This is something that Cory Doctorow talks about in his reverse-centaur article.
"And if the AI misses a tumor, this will be the human radiologist's fault, because they are the 'human in the loop.' It's their signature on the diagnosis."
This is a reverse centaur, and it's a specific kind of reverse-centaur: it's what Dan Davies calls an "accountability sink." The radiologist's job isn't really to oversee the AI's work, it's to take the blame for the AI's mistakes.
This quote from the article very much sums up my own experience of Claude:
In my recent experience at least, these improvements mean you can generate good quality code, with the right guardrails in place. However without them (or when it ignores them, which is another matter) the output still trends towards the same issues: long functions, heavy nesting of conditional logic, unnecessary comments, repeated logic – code that is far more complex than it needs to be.
AI coding tools definitely helpful with boilerplate code but they still require a lot of supervision. I am interested to see if these tools can be used to tackle tech debt, as often the argument for not addressing tech debt is a lack of time, or if they would just contribute it to it, even with thorough instructions and guardrails.
Good question, though wouldn't that be a feature of the browser? It wouldn't be a good user experience asking people to manually enter coordinates.