I've spent about 23,000 hours writing sofware. Last Friday, for the first time, I started to see my net value as a software developer start to trend down. I saw the machine's ability to code without me, begin to outpace its ability to code with me. Some of that was because Claude got better. Some of it was because of the agents I've been building to replace me. But whatever the reason I started to realise my influence on the machine is ratcheting across from boost to drag. Building software at the quality-level I require, has historically meant either writing it myself, or overseeing Claude like a hawk. Because as smart as Claude is, its mistakes are still bad enough to undermine the long term quality of an application. But then I built agents to correct those mistakes. Agents that mean I don't need to watch for them. Agents that mean I can spend my time doing other things instead. For every 50% of my work I automate, Claude moves 100% faster. Which is empowering. And weird. Because it represents my own obsolescence. The mantra of the AI-maximalist crowd is that humans will find other work to do. That we always do. That we move on up the value chain. And that was true when revolutions affected just some of the jobs we did. But what happens when they affect all of them? Because I'm not just automating my code, I'm automating everything I do. I'm even automating the process of automating. I'm writing agents to watch my interactions with Claude and look for additional jobs it can take over and automate for me. Ways to automate another 50% of what I do. Double again. My belief to-date has been that my value as a software developer would be retained in my taste, my experience and my ability to drive the machine. But something changed on Friday. I got a sense that the machine I'm building may not even need a driver. When algorithmic trading arrived, the gap between humans and machines wasn't incremental - it was a category change. Algorithmic trading happened so fast humans coudn't even perceive the game being played, much less participate. Out-trading the machines was like out-cutlassing a tank. The game had evolved. And while past technology revolutions brought floods of change they also came with high-water marks. There was a point where you just couldn't add a steam engine to any more jobs. And so humans moved up into those jobs instead. This time feels different. Because everywhere we go to escape the AI, we find more AI. Helping us, accelerating us, accelerating itself. Becoming faster and better until our only influence is to hold it back. Even as a fully AI-pilled developer this feels peculiar. Seeing my edge erode and becoming automated. Feeling the water lap up against my chin. Seeing just how quickly, quietly and undramatically it all happens. And since I am both a developer and an entrepreneur there is usually another door and another floor for me to climb to. But I rather wonder for how long.
Really disagree with this one Peter Nixey. As someone who is interviewing dozens of top-tier AI engineers every week, it is very obvious that we need the drivers. I don't even know where to start explaining how it would feel to live in a world where enterprises are run on claude-written software. It's all happy times when you start a project, but as you build on it, as you need to streamline, debug, refactor, LLMs are for from being efficient, or even intelligent for that matter.
to see this from YOU makes this post feel extra edgy! I am not a dev, but I have been deep into crypto for 5 years and the emerging Agentic AI Meta (2.0) is giving me tingles. Most of my corporate clients are stuck between dismissing the hype (there is hype ofc) and feeling fearful to "do sth." beyond Copilot. Red tape. The way most orgs approach AI is like another software tool. That's why I am building Challenger House, for leaders who act on AI differently. Who rethink work, and not just re-tool it. As an org dev expert of 20 years, I see entire new org designs emerging once AI agent teams replace and augment the remaining talent. I advocate for an intentional transition of the stale employer-employee divide towards a new model: make all contributors co-owners; via tokenising the specific value they co-create, on chain. I would love to get your view on this Peter Nixey if you can spare some time.
I would argue your value hasn’t plummeted just yet I would argue we are seeing the piano-ificstion of software engineering Everyone knows how to play it but only the experts will take the time to learn how to PLAY it Agents still require a level of prompting that requires prior knowledge “the clever horse effect” And as a manager you need a way to verify the output
Thank you for sharing that. Might be a good time for the two of us to reconnect.
What an amazing post! So thought-provoking and evocatively written at the same time 🤩
Awesome post Peter! Curious about your thoughts on teaching other humans to do what you have learned to do, and then automating that teaching.
Very nicely put, Peter. Totally agree. Been a while, eh!
I recognise the trajectory and yet I'm optimistic. Not necessarily because I see myself infinitely rising "above in the value chain" but rather because automating everything will push me to do things that only humans can do — and we're all discovering what they are. I really struggle to believe that a human being like you, with agency, good education, good character, strong network can run out of doors to open. Will they all be economically super-valuable? Who knows, but I'm sure they're infinite.
I share your "uncomfortable excitement" with the progress in AI coding, but I still think humans have a pretty essential role figuring out what to build in the first place, which is something you're really good at! Understanding unmet market needs and what's going to be better enough but also easy enough to adopt and play well with existing services etc. still seems like the harder job than building to spec once you know what you want. Or do you feel like even that is something where AI is edging you out?