I don’t believe AI is making engineers obsolete. I do believe it is raising the premium on the ones who know what they’re doing, and that the stakes are very high if they’re pointed at the wrong thing.
The AI conversation I’m seeing is very much centred around the question: who gets replaced? But we are still so early in agentic coding. The engineers who can hold a system in their head, the ones who can see security risks before they happen, the ones that can make AI-augmented software actually work… these engineers matter more now, not less. Anyone can vibe code, but not everyone can vibe code something you’d put in production (without potentially huge risks to quality and security). The people we need to trust to direct agentic workflows, and to ship agentic-written code, are the ones who’ve already lived through the consequences of bad ones. I’m firm on that.
In this new landscape, ideas have become cheap and infinite. Senior engineering talent is still finite, though. The engineers I see speaking out against vibe coding on LinkedIn, they’ve been around, they’ve watched things go wrong, and they’re the ones who’ll have to pick up the pieces. And it’s harder than ever to protect them from the firehose of ideas and requests, because everyone’s brain has been set on fire by AI. It’s not a bad thing or a good thing, it’s just the new thing. Anyone with a Slack account can now generate a credible-sounding feature request. The torrent of what if we just… ideas is bigger than it has ever been, and meanwhile, the people who can actually steer the build, the ones who know which corner of the system will break under load and which abstraction will rot, they are being seen as the bottleneck. But sometimes the bottleneck is the point. Right now, the real bottleneck has moved upstream to problem definition and judgement, and misdirecting those people could be a pretty expensive mistake for a software company.
A few years ago I joined a small startup in a complex, traditional industry. The kind of industry where a single project has twenty-plus stages and dozens of stakeholders touching it. I was hired as a front-end engineer to work on project management software. The project statuses we were showing users were confusing, and somehow, the data wasn’t making sense. So I worked on the front end, but I realised the data coming out of the API was wrong. When I went looking, I realised the API logic itself was wrong. The logic had never actually been written down in plain English, and what was in the codebase didn’t match how the industry worked. The founder, who knew the industry inside out, and the lead engineer, who was solving for technical elegance, had never fully agreed on what the software was supposed to do. The founder explained the domain, the engineer translated it into a technical solution. Nobody had codified the logic in between and that misalignment had calcified into the architecture.
The consequence wasn’t abstract. Project managers were showing up on site with an app that didn’t reflect reality. It couldn’t do the thing it was promising to solve, and it made them more confused. And so… why would they continue their contracts with the company? I spent the next stretch sitting between both sides, making sure we could build a product that users would love, so that their procurement teams would keep paying our contracts. I was figuring out how to translate all of it into something the engineering team could actually build against. And so, the system stabilised. The company won a meaningful contract and we got asked to prototype the next product. Cue the sales bell ding!
That was years ago, before AI. Imagine the same misalignment now, with engineers building the wrong thing ten times faster. The cost of a bad spec has never been higher.
The PM’s job in this agentic coding environment is not slowing things down with process, and it is not simply coordinating tickets. It is absorbing the firehose so engineers don’t have to. It is holding the bridge between what’s being built and how it lands commercially. It is directing a scarce, expensive, increasingly powerful resource toward the things that actually matter.
And I think technical PMs have an edge here, because many of us have been engineers. We understand the cost of a misdirected engineer instinctively. We can read the architecture and notice when it stops making sense. We can tell whether an AI-assisted output is actually good or just confidently formatted, and that goes both ways: code being shipped by a dev, and “feature ideas” arriving from customer-facing teams. The translation matters more than ever, because AI is fluent in the surface of things, but not necessarily the substance.
Much of the AI coding conversation now is still fixed on who gets automated away. I’m more curious about which roles get redirected, and how a company’s energy moves when code and ideas are cheap, and judgement is expensive. Engineers still matter, possibly more than ever. As a technical PM, I’m paying attention to the seam between them and the rest of the business, because to my mind, that is where the work is going to compound, or quietly come apart.