In February 2026 I walked out of a five-hour-long meeting that my CTO had named “Why should we still write code”. We discussed what it would take for us to never write a single line of code anymore and what that would look like. I remember coming out of that meeting feeling like we were on the precipice of a major shift. That my work would be drastically different before the end of the year.
I was not mistaken.
At my company we’ve set up a workflow enabling “agentic development”, an environment of agents and skills that you only have to feed a ticket describing a bug, improvement or feature. The ticket is analysed and a plan is written. A human reviews the plan and either forces changes to the plan or gives the green light to start implementation. After implementation, the human tests the work and reviews the code.
It’s terrible.
Not because of the code quality—which is surprisingly decent, though it’s hard to measure across so many changes—but because of the process. Reviewing code has always been part of the job, but it’s rapidly becoming all of the job. You hear people say that the work is now shifting from implementation to further up the pipeline, where you bring your technical expertise to refining the briefing statement. For some developers, the work might shift that way, but this isn’t the case for everybody. If you have 60 developers working on the software, maybe only about a quarter of them are included in the discussions where the feature description takes shape. The others are merely implementers, either not senior enough or simply not the type of people who think about the product that way. Those are the developers who especially love writing code, wrangling zeros and ones into a working product—and they’re the ones who will start to hate their jobs. They’re turned into reverse centaurs.
Reverse centaurs
I like how Cory Doctorow describes what’s happening. He mentions centaurs and reverse centaurs. Centaurs are workers that are enhanced by AI. They’re in control and use AI to augment their abilities and make their tasks easier. Reverse centaurs are not in control, they are in service to the AI—though not in the dystopian kind of way, not really. If all you do is check the work AI has done, you’re a reverse centaur. And being a reverse centaur sucks the fun out of work.
I wholly believe that making reverse centaurs out of your employees will have a negative effect on the quality of the software and ultimately your company. It’s too soon to confidently predict the fallout and it will likely be a combination of things. AI can generate a lot of code and seem very productive, but the number of bugs being introduced to the software will increase as well. Developers are struggling to keep up with the amount of code that needs to be reviewed. They’ll get lax, or they’ll fail to understand the code. Critical thinking is being outsourced and no developer truly owns the code that is being written. But it’s where the perceived monetary gain is at. If you can get AI to do most of the thinking, you can “right-size” (how I loathe that word) your engineering department by 30–50%—assuming that same 30–50% hasn’t already quit out of sheer frustration.
Centaurs
I’m not saying developing with AI is bad and will lead to bad code. Leaving aside everything wrong with the power-hungry data centres and the money-hungry companies financing it all, AI as a technical achievement is mind-blowing and it is the most powerful tool in a developer’s tool belt ever, full stop. But it’s only positive if it’s used like that: a tool. Something that enhances the abilities of a developer—augments them—makes centaurs out of them.
When used responsibly it can give junior developers wings. They will be able to do a lot more than previously would be thought possible. But junior developers would be wise to make sure they understand and own the code that they generate. When used responsibly it can make senior developers super productive.
Still fun
I consider myself fortunate. As a lead front-end developer, my role still has plenty of texture. I lead a platform team, improving our software’s shared library and underlying architecture. I align cross-functional teams, and spend a good chunk of my time executing work myself. When I use AI, I use it as a sparring partner, do analysis, or automate tedious work. I’m still learning, and I still like what I do.
But if you’re reading this and feel like you’ve been relegated to a reverse centaur, don’t accept it as the new normal. Talk to your CTO or engineering manager. In many cases, leadership doesn’t set out to make your job miserable; they just want productivity gains and assume “agentic workflows” are the best way to get them. If you can demonstrate that you’re a sharper, more effective, and happier developer when you’re driving the AI rather than babysitting it, most reasonable managers will listen.
And if they don’t? Well, you might want to ask yourself how long you want to keep holding the leash.