richard
Get in touch

Engineering culture · 8 min read

How software engineers really feel about AI taking their jobs

Richard Raduly
Richard RadulyFractional CTO · 25 July 2026

For the past six months I’ve been talking with dozens of software engineers about AI: what they make of it, and how much room it now takes up in their day. Something has changed, and for most of them it changed hard. I wanted to write down the emotion I keep running into, because it rarely makes the coverage, and to add my own read on it.

Before you read on, try something. From the coverage of the last six months, answer two questions: whether your job still exists in three years, and what you’d be doing in it.

  • Tech layoffs are running at about 1,115 a day, nearly double last year’s pace, and 54% of them name AI.

  • Only 9% of hiring managers say AI has actually replaced a role at their company.

  • Employment for developers aged 22 to 25 is down nearly 20% since 2024, and new computer science graduates are now more likely to be unemployed than new art history graduates.

  • Engineering hiring is down 11% since 2019 while tech hiring overall is down 25%, and engineers were 55% of new hires at big tech in 2025, up from 46%.

  • Asked in their own words why they lost their job, 1% of laid-off workers said AI. The top answers were restructuring, budget cuts and the economy.

  • 59% of hiring managers admit their company plays up AI when it explains cuts, because that story plays better with stakeholders.

  • Google said in April that 75% of its new code was AI-generated and approved by an engineer, up from 50% late last year.

  • The head of Claude Code at Anthropic announced he hadn’t written any code himself in over two months.

  • Measured across 4.2 million developers, AI-authored code came to 26.9% of production code.

  • 32.7% of AI-generated code gets through review unmodified. For human code it’s 84.4%.

  • In a controlled trial, experienced developers took 19% longer to finish tasks with AI, and came away believing they had been faster.

Now write your answers down. In three years, will software engineering still exist, and what will engineers be doing?

Whatever you wrote, you chose which lines to believe. And you chose them based on how you already felt before you started reading. Nothing in the first list settles the question. Every line is sourced, recent, and about the same job. Some of it is marketing, and we know that for sure, because the people selling it answered a survey and admitted it. Some of it is real. But nothing here tells you how much is which.

The second list is worse, and it’s easy to miss why. The first question can be answered one day: hiring numbers are real, and sooner or later they’ll settle. The second question can’t be measured at all. “AI-written” means one thing at a company where the AI is a fancy autocomplete, and something completely different at a company where the AI builds whole features on its own. So the 75% and the 26.9% don’t contradict each other. They just aren’t counting the same thing. Nobody can tell you how much of your job is still yours, and that includes you. Remember the trial where developers were slower but felt faster? It’s the same problem, pointed at you.

So this is where people are stuck. Maybe it’s decline. Maybe it’s replacement. There’s no way to know which one, or what’s left of the job if it’s neither. The second question is the harder one to live with. If the work is going away, you can retrain, move, get angry, or leave. But if the work stays and slowly turns into something you can’t picture, and maybe wouldn’t have chosen, there’s nothing to do but keep showing up and find out.

Every engineer I spoke to had already read all of this, most of them more closely than I had. Not one of them could tell me where they stood.

Andrej Karpath, cofounder of OpenAI has as good a view of this as anyone, and he sounds just as lost. On 26 December he wrote:

“I’ve never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year, and a failure to claim the boost feels decidedly like a skill issue.”

What nobody disputes

In a little over a year the job changed under people’s feet. We need a lot less hands-on coding skill now, and a lot more supervising: you set up the problem, hand it to the model, and check what it gives back. The closest analogy I have is telling a blacksmith he now runs a metallurgy line. He still knows metal. But almost nothing he loved about the daily work survives the move.

A lot of engineers came to this because they like the craft. They like sitting with a hard problem and writing their way out of it, and they’re the ones taking this the worst. It’s worth being honest about why.

Evil Martians described the mechanism well. The old loop went planning, then building, then result, and the building was the part you enjoyed. AI squeezes the planning into a prompt and swaps the building for review. In their words, “we remove the part that we like, and replace it with a part that we really don’t.”

Picture two engineers with the same four-hour task. One works it by hand and finishes with something to show for it. The other lets AI do it in two hours, ships twice as much, feels oddly empty, and keeps going anyway. “We compensate for a lack of satisfaction with work quantity.” What they’re describing is stranger than ordinary burnout: a kind of overload that looks like productivity from the outside.

The trust numbers back it up. In the 2026 Stack Overflow survey of more than 49,000 developers, AI use hit 84% while the share who highly trust what it produces was 3%. The top complaint, from 66% of them, is code that’s “almost right, but not quite,” and 45% now spend more time debugging the model’s output than they’d have spent writing it themselves. Aruna Ranganathan at Berkeley gets to the same place from the other side: AI doesn’t reduce the work, it intensifies it. Whatever time you save goes into doing more, not into resting or thinking.

So the daily mental load hasn’t dropped. It changed shape, and for a lot of people it weighs more than the old one. Underneath it, plenty of people in tech are quietly asking what their work is even for now.

“AI slop”

“AI slop” is the new word for content a model churned out that adds nothing and reads fake. It’s landing next to words like “spam,” “junk TV” and “clickbait,” and sociolinguists have started studying it alongside “enshittification” as one of those words that finally lets people name a decline they can feel online.

Inside teams, though, the anger about slop is rarely about quality itself. It’s usually about what sits underneath: whether people still feel safe, whether they’re being treated fairly, whether the job still feels like theirs. When AI arrives with no clear rules, they feel judged by systems they can’t see inside. They stop being sure their own judgment counts for much. And they’re caught both ways, because using the tools can make you look lazy and skipping them can make you look slow. Recent research ties AI adoption to more stress and lower mood exactly where psychological safety is thin and leadership is weak: people stop speaking up and quietly resent the whole thing. “AI slop” becomes an acceptable way to say the harder stuff out loud:

“Management doesn’t care about quality, just output metrics.”

“My craft is being cheapened by auto-generated code.”

“We’re drowning in garbage and nobody owns cleaning it up.”

See the phrase as a mask over those fears, and you can start dealing with what’s really behind it.

So what do we do?

For everyone’s clarity and sanity, an organisation needs a clear position on AI, a real one people can repeat, not a slogan on a slide. The following few questions can help you and your team to approach this:

  • What problem are we actually solving with AI, and can we say it in plain language?
  • Where do we want AI to support human judgment, and where do humans keep the final call?
  • What work are we deliberately taking away from people, and what do they get to do with the freed-up time? Without an answer, that time just fills back up with more work.
  • What can we honestly promise, and keep, about jobs, evaluation and fairness as we roll this out?
  • How do the people doing the work get a say in where AI is used and where it isn’t?
  • What do we want our people to be better at because of AI, and are we investing in it?
  • Where do we make it safe to talk about the AI-specific stuff, the fear of being made redundant, the embarrassment about skills, the slop fatigue, before it turns into burnout?

I have it easier than most. I work in healthcare, where quality isn’t negotiable and where the nature of the work has already answered a lot of these questions for us. But wherever you are, these need to be clear, for you and for the people around you.

Name it: this is grief

I think it helps to be precise about the word. What a lot of engineers are going through is grief. They’re mourning a craft, and that brings up real fear, real anger, and a harder search for meaning than ever before.

The managers and leaders steering this owe it to their teams to take that grief seriously, instead of trying to bury it under metrics or push it out of sight. Name it, give people room to actually feel it, and the team has a real chance of reaching whatever the job becomes next.

Sources

#ai#engineering-culture#burnout
Richard Raduly

Written by

Richard Raduly

Fractional CTO for regulated healthtech. I build engineering teams that ship, pass audits, and stay, then hand them back stronger than I found them.

Follow on LinkedIn

Field notes by email

Get the next note when I publish it.

One email per note. Nothing else.

Get in touch