I think the biggest factor is fear. When we first started discussing LLMs a few years ago, it was easy to be amused and reassured when looking closer at all the inaccuracies. The models have improved at an alarming rate. Some of us have spent decades automating the roles of others, and now it turns out our own medicine doesn’t taste very nice. The timing is also terrible because the rise of these tools is coinciding with a dramatic post-COVID tightening of the jobs market.

On the other hand, it’s frustrating to see the trend and tools go towards vibecoding and fully agentic development. Many of us have also been in the business of inheriting (and supporting) code, so it goes against the grain when non-developers produce something without even attempting to understand how it works. Because we’ve been there in the trenches trying to diagnose and debug a serious problem out of hours under pressure, and know how essential it is to have a decent mental model of how it is meant to operate.

As a personal anecdote, I was working in an area with someone business-focussed and not particularly technical. There was some functionality we had been discussing, and one day they wanted to discuss it on a call. They then went on to demo something that seemed to have a working implementation of all the features we had been discussing, and more. I was curious, so asked them to screen share the code… at which point they started to get a bit cagey. I managed to get them to show it to me, and it turned out they had vibed the whole system as a single massive React component. And had no clue how any of the code worked. I told them I couldn’t possibly integrate that massive ball of spaghetti, and we agreed to treat it as a throwaway demo prototype and develop any production system properly. That sort of mess is inevitable when banging Accept All like the LLM is a casino slot machine.

So personally, I have spent the last few years trying to amplify my skills and experience with these tools rather than bury my head in the sand. Three decades as a freelance consultant developer (with several stack pivots) has taught me that new technology trends don’t simply vanish if they are providing at least some business value. Don’t wait for all this to go away, because that day will never come.

I also think this technology makes us all generalists, unless somebody’s specialist knowledge and skills are very deep and highly unusual. It won’t be so easy to be a backend-only guy who doesn’t touch frontend, when your project stakeholder thinks he could have a stab at it with his free Claude account.

On my journey with this tooling I’ve struggled to find the line between how much I write versus how much I generate, and tried to maintain the balance between velocity and quality. I gravitate towards a workflow of insisting that I understand and approve every diff, and use the knowledge ingested by the LLMs to keep learning, and try to be sensitive to when the convenience has become laziness and there’s a danger of de-skilling. I combine my experience and instincts with the new powertools and feel that we are greater than the sum of our parts. I just have to hope that when this all settles down a bit, there’s a solid market for that type of work rather than being replaced by non-technical vibecoders with only velocity to offer (back in the day we used to refer to them as cowboy coders, but same idea).