Thank you for opening up the latest article in 'The Intelligent Commit' newsletter, my thoughts, research and analysis on Artificial Intelligence and how AI is reshaping the software development lifecycle. I hope you find it insightful, subscribe, and engage.

At a recent Executive Edge dinner, technology leaders were debating which rank of software engineer is most at threat from AI, if any at all, whether the juniors, the middle, or the seniors. Opinion was mixed, with compelling arguments for every scenario, each backed by credible witness evidence from within their own organisations. Curious as usual, I went digging into the data and research, and took a side.

You know I like an analogy. Last November, I had a new boiler installed and on day one the apprentice didn't show. Typically, a mid-morning water leak crisis ensued and I reluctantly entered into the narrative (as anyone that has seen me put up a shelf would attest). Fortunately I found the right set of dongles and widgets in the van, and together we avoided disaster. Anyway.

Think about how anyone becomes a master plumber in the first place. They didn't start out installing a new boiler on their own. They started as the apprentice, fetching the widgets from the van, carrying the bags, bleeding the radiators, watching someone who knew what they were doing, and getting corrected a hundred times a week. Running to the van was never the point, but it was how you earned a spot close enough to the work to absorb the craft. Twenty years later, that apprentice is the one out installing boilers, and there's a new apprentice running to the van, learning the trade.

Software development works the same way, or it always used to.

The difference is that the junior tasks, the ones our engineers learned on, are exactly the ones we've started handing to AI.

Within the last fortnight, Stanford's Digital Economy Lab published "We Must Act Now", organised by the economist Erik Brynjolfsson and signed by 16 Nobel laureates and over 200 economists and AI researchers. While warning that the pace of change could be much faster than many expected, it is careful rather than alarmist, and is deliberately non-prescriptive.

The damage is concentrated at the bottom of the ladder.

The statement is a signpost to Brynjolfsson's earlier related analysis, "Canaries in the Coal Mine", which found that for roles most exposed to AI, those aged 22 to 25 have seen employment in the US fall, while their older colleagues held steady.

Within this data, software development is the clearest case. In a widely shared piece, Laurie Voss laid out the numbers under a blunt title, "AI has torched the market for junior programmers". The UK picture is more startling. Adzuna data shows entry-level IT vacancies are down 55% since ChatGPT launched in November 2022.

If fewer juniors ever become capable middles, today's gains become tomorrow's shortage.

That answers part of the dinner debate. Of the three ranks, the middle may fare best for now, soaking up the biggest productivity gains. But the risk sits upstream, and if fewer juniors ever become capable middles, today's gains become tomorrow's shortage.

Growth at the top can hide collapse at the bottom.

How does it feel out there? Some readers may not recognise this story at all, especially when total developer employment is up 4.4% from late 2022, and job postings have risen for the last thirteen straight months (up 10% year-on-year).

But those stats are the totals, and juniors account for only a relatively small share of the developer workforce. The bottom rung of the ladder can vanish entirely without either figure flinching, which is roughly what's happening. The total rose, while the 22-to-25 cohort fell.

The tasks we're automating are the apprenticeship itself.

Crucially, a rising headcount tells you nothing about whether juniors get the work that turns them into seniors. Apprenticeship was never just employment; it was repeated exposure to low-risk, reviewable work under supervision. In Voss's words, you got hired to write ‘mediocre code’, a senior reviewed it, you absorbed judgment through repetition and correction, and a decade later you were the senior doing the reviewing. AI is increasingly writing that routine, reviewable code, and the rung you used to stand on is being pulled away before the future leaders can get a foot on it. Unless we deliberately build new ways for juniors to earn that engineering instinct, the apprenticeship disappears faster than any replacement arrives.

Displacement is happening at the level of tasks while growth happens at the level of firms.

Those aren't contradictory. A company can grow, hire more engineers, and quietly stop producing the seniors it will need five years from now.

The senior doesn't get automated away with the junior.

Here's a question I heard at the Executive Edge event. Will AI eventually remove the need for seniors too? Some feared more for the seniors; I'm not so sure. Ask me again in six months. My view today is that the more code the machines write, the more somebody has to read, integrate, secure and keep that code running in production, and that requires judgment and human responsibility.

I may be biased, having spent much of my career on that side of the line, in reliability and security engineering. It gets harder as the volume climbs; doing continuous delivery at scale is a buzz, but it needs strong leadership to do well, safely, securely, and sustainably. The scarce resource is the people who can recognise when generated code is wrong, unsafe, brittle, or solving the wrong problem altogether. AI increases the volume of software, but it doesn't eliminate the need for engineering judgement.

Of course, I'm not neutral on this, because I climbed this exact ladder. I started as a junior developer back in 1997, writing plenty of forgettable code that a senior picked apart in review. Being the apprentice and receiving the feedback mattered. Take away the experience of getting schooled on design patterns, best practices, and yes, my poor code, and I'm not sure what promotes the next me.

I remember a conversation less than a year ago about whether AI was a pair programmer, and whether it wrote the code or reviewed it. The question now isn't whether it does the junior's work, it's what that does to the person who used to.

You can't buy a senior in 2031. You have to grow one.

Regular readers of this newsletter will be used to me asking if the measurement is broken, rather than the technology. Same again. Headcount is easy to count, but offers little. So is autonomy, as a junior with an agent can still ship things nobody should trust. The outcome that matters is judgment, whether your two-years-in engineers produce work that holds up on quality, security and reliability, and can spot when it doesn't.

If we agree the need to keep hiring juniors, here are my key takeaways to protect the pipeline of seniors for tomorrow in an AI-augmented world of software development.

  • Ring-fence apprentice work. Deliberately reserve a slice of the now-automatable tasks for juniors to do by hand, even when an agent would be faster. The point of that work was never the output, it was the learning.
  • Put juniors on the review side, not just the generation side. Have them critique and correct the agent's output and explain why it's wrong, rather than only prompting for more. Judgment comes from catching mistakes, not accepting them.
  • Pair and mob on purpose. Human pairing still transfers judgment fastest, with the usual caveats about psychological safety, and cross-seniority pairing takes skill. Code review is the cheapest apprenticeship you have, and the first thing to go when everyone's reviewing the machine instead of each other.
  • Measure judgment, not just output. Not how much your early-career engineers ship, nor how little help they need, but whether the work holds up to scrutiny. Does it pass review without rework, survive production, stand up on security and reliability? Who can take on materially harder problems than they could six months ago?

The craft of software development itself isn't dying. It's arguably never been healthier, with millions more now building real software without ever claiming the job title. What is at risk of dying is the on-ramp to doing it well.

And it isn't only our problem. Every knowledge profession, law, medicine, accountancy, makes its experts the same way, by letting beginners do the simple things under a watchful eye. AI is increasingly capable of taking on many of the simple things. Do we let it?

You can take the bottom rung off a ladder and it'll stand there looking perfectly sound, right up until the morning you look up and realise there's no one left who knows how to climb it.

Thank you for reading my latest article. Please share your experiences and engage with the discussion. I am not always right, but I try to be insightful and thought-provoking.

  • Juniors, middle, or seniors, which rank do you think is most at risk from AI?
  • When you last let an agent clear a task instead of handing it to a junior, did you weigh the code you saved against the engineer you didn't grow?
  • Can your organisation measure whether its early-career developers are genuinely more capable on hard problems, or only whether they're getting more productive?
  • If you stopped hiring juniors tomorrow, where does your first senior engineer of 2031 come from? Do you think that even matters?
  • Is your code review still a place where judgment gets taught, or has it become a rubber stamp on the agent's output?

Please reach out if you'd like to continue the discussion.

Sources

  1. Stanford Digital Economy Lab, "'We Must Act Now': Sixteen Nobel Laureates Join Leading Economists and AI Researchers", 13 July 2026 — https://digitaleconomy.stanford.edu/news/wemustactnow/
  2. Brynjolfsson, Chandar & Chen, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence", Stanford Digital Economy Lab, November 2025 — https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
  3. Laurie Voss, "AI has torched the market for junior programmers", seldo.com, 4 July 2026 — https://seldo.com/posts/ai-has-torched-the-market-for-junior-programmers/
  4. Adzuna research on UK entry-level vacancies, reported in Business Matters, "Entry-level jobs slump 32% amid rise of AI and growing employer costs", 30 June 2025 — https://bmmagazine.co.uk/news/uk-entry-level-job-losses-ai-chatgpt/