I’ve spent the last few weeks navigating two different kinds of grief. One is personal. The other is professional: watching the slow, agonizing death of the “Developer Who Knows Why Things Work.”
Grief does something useful, though. It strips away your tolerance for nonsense. When you’re running on fumes, you don’t have the energy for the fluff. You see the load-bearing walls and ignore the wallpaper. Turns out, that’s not a bad lens for architecture either.
As an Architect, my job used to be drawing rectangles and arrows that represented logical flow. Today, I feel more like a zookeeper trying to contain an explosion of AI-generated code that nobody actually understands. And right now, I don’t have the patience to pretend that’s fine.
The Post-Knowledge Developer
We are entering the era of the Post-Knowledge Developer. This is a world where “senior” engineers are essentially prompt-tuning their way through a sprint. On paper, velocity is up. In reality, we are building castles on someone else’s foundations without checking what’s underneath.
AI is fantastic at the what, but it’s historically illiterate regarding the why. It can give you a perfectly formatted LWC component, but it has no idea how that component will interact with a legacy database that hasn’t been patched since the late 2000s. It never sat in the architecture meeting. It doesn’t know about the workaround that Dave built in 2019 because the vendor API was broken. It can’t smell the code.
The Post-Knowledge Developer ships fast and debugs slow. They can generate a solution in minutes but can’t explain the trade-offs it’s making. Ask them why this pattern instead of that one, and the honest answer is “because the AI suggested it.” That used to be a junior developer trait. Now it’s becoming the norm across seniority levels.
When you’re grieving, you see this with painful clarity. The pretence falls away. And what’s left is a room full of people building things they can’t fully explain.
The TCO of “Free” Code
Stakeholders love AI because they think code is now “free.” It’s not. AI-generated code is free to create, but it is catastrophically expensive to own. In my experience, two costs are hiding in plain sight.
Maintenance Debt. If no-one wrote it, no-one feels responsible for it. I’ve seen this play out already. A generated service breaks at 2am, and the on-call engineer stares at code they didn’t write, can’t explain, and barely recognise. They don’t debug it. They regenerate it. The new version might fix the immediate problem, but it also quietly changes three assumptions the rest of the system depends on. Rinse, repeat. Within six months you don’t have a codebase. You have a palimpsest. Layer upon layer of AI-generated solutions, each one overwriting the last, with no human being holding the thread of why it was built that way in the first place.
(That’s the Contextual Decay problem I wrote about last year, except now the context isn’t just decaying. It was never there to begin with.)
Architectural Erosion. AI doesn’t care about your long-term roadmap. It cares about satisfying the current prompt. It will happily introduce a circular dependency if it makes the function work right now. One component at a time, each one technically correct, the overall system drifts away from any coherent design. It’s death by a thousand pull requests. No single one is catastrophic, but the cumulative effect is an architecture that nobody designed and nobody can explain. I wrote about this in Escaping the Feature Factory, but at least in a Feature Factory the humans were making the bad decisions. Now we’ve automated it.
The Amber Trap, Accelerated
Here’s what keeps me up at night. Everything I wrote about the Amber Trap, about how documentation fossilizes into something that actively misleads, is about to happen at ten times the speed with AI-generated code.
Think about it. Traditional code had at least the ghost of a developer’s intention behind it. Variable names that hinted at decisions. Comments that occasionally explained the why. A git blame that led back to a human who remembered the context (or at least remembered the argument that led to the context).
AI-generated code has none of that. It arrives fully formed, syntactically perfect, and contextually hollow. In eighteen months, when the model that generated it has been deprecated and the developer who prompted it has moved on, that code becomes amber. Perfectly preserved. Totally misleading. And nobody in the room can tell you whether it’s still the right solution or a fossil from a world that no longer exists.
The traditional Amber Trap took years to form. The AI version takes quarters.
The Human Side of Architecture
Being an Architect is often seen as a purely technical exercise. But systems are built by people, and people are trying to keep up with a machine that doesn’t sleep, doesn’t grieve, and doesn’t need a coffee break.
Lately, I’ve realised that my value as an Architect isn’t in my ability to embrace the new. It’s in my ability to be the sceptical, slightly grumpy adult in the room. That is probably why Captain Haddock is my spirit animal. (Billions of blue blistering barnacles, who approved this pull request?)
My advice for fellow Architects? Focus on System Boundaries. If you can’t control the quality of the code going into the boxes, you better make sure the boxes are damn near impenetrable. Define the contracts. Enforce the interfaces. Make the blast radius of bad code as small as possible. Because bad code is coming faster than ever, and the only thing you can control is how far the damage spreads.
Operating at Lower Capacity
I’m currently running at a lower capacity than usual. I’m distracted, I’m tired, and I’m grieving. But ironically, that’s made me a sharper Architect. When you don’t have the energy for the fluff, you only focus on the core of the matter. You stop nodding along in meetings and start asking “but why?” again. You stop accepting “the AI suggested it” as an answer.
Grief is a brutal lens. But it’s an honest one.
I’ll be back to my usual “visionary” self eventually. For now, I’m just making sure the boxes hold.
Leave a Reply