The House Article | What Software Engineering Can Teach Us About AI And The Future Of Work

(ronstik/Alamy)

10 min read18 May

Mr Bucket spends his days screwing little caps onto the tops of toothpaste tubes, until the factory closes and he loses his job, representing a financial blow for the family in Charlie and the Chocolate Factory. The fortunes of Charlie’s grandfather turn around in the 2005 film adaption, however: he later retrains and gets work again – repairing the robot that replaced him at the factory.

Software engineers hope that this best-case scenario is how the advent of AI will turn out for them. Right now, they are going through stage one – the painful bit.

After a hiring boom towards the end of the pandemic, the labour market is freezing over, with companies that once hoovered up junior developers pulling back. While increasing investment in AI, Meta is cutting around 8,000 jobs and Microsoft is offering voluntary redundancy to almost 9,000 of its American workers. It is part of what has been described as a “tech layoff tsunami”.

“AI is causing mass confusion and hysteria,” confirms Anna Brailsford, CEO of Code First Girls, which provides free coding courses to women and connects them with employers. “Three to four years ago, you could almost predict what a workforce plan would look like on an annual basis for an organisation… AI has thrown those workforce plans into disarray.”

Google has Gemini Code Assist. Anthropic has Claude Code. OpenAI has Codex. GitHub Copilot is another option; Cursor, offered by startup Anysphere, is also popular. These tools, which can overlap with each other in complicated ways, vary in capability – from AI ‘assistants’ that help developers code faster to autonomous AI ‘agents’ able to go off and complete tasks alone. Prompting AI tools to generate code for you is known as ‘vibe coding’. 

Very quickly, these AI tools have become integral to the everyday working lives of engineers. Advocates say they boost productivity and eliminate grunt work. But there is also widespread anxiety over them replacing developers – particularly junior ones, historically employed to undertake the basic coding that AI can now do easily. And the tools typically cost companies hundreds a month, rather than the thousands paid to workers.

In Britain, AI’s effect on software engineer jobs is being intensified by increased national insurance and new employment rights, says Brailsford, “making organisations more risk-averse when it comes to taking on particularly entry-level technical talent”. (Increasingly, Code First Girls are hiring the developers they train themselves, then deploying them to clients for a period of time, at the end of which the employers can convert them to permanent members of staff. “We take on all the risk of an early hire.”)

So, is this simply a painful transition – or are software engineering jobs under threat long term?

Economic historian Carl Benedikt Frey describes how job displacement depends on the nature of the tech (whether it enables new tasks or automates existing ones); elasticity of demand (demand for healthcare, for example, only grows as we get richer); and cultural, policy and regulatory factors.

Unfortunately for coders, he concludes that AI tools are largely automating existing tasks. The impact of this can be mitigated by other factors, such as professional qualifications being created to shield people from advances in technology. But the overall picture painted here is challenging for developers.

“You will probably have some managerial senior roles,” Frey says. “But you’re not going to need anything like the number of software engineers you currently have in order to do that particular part of the job.”

Brailsford is clear in her view that the notion AI will get rid of the profession entirely is a misconception. “There is a high level of confusion at the moment within organisations, and, yes, that is affecting the job market.” Yet the demand for software engineers will be higher than ever by 2030, she predicts – though the talent will look different.

“They will be replaced by a new profession of systems thinkers and systems architects – that is, people who have the ability to connect systems very effectively. It is not going to erase the need. I think the need will be bigger than ever.”

The kinds of engineers in demand are already changing – whether the employer is an enterprise dealing with ‘legacy tech debt’ (software solutions that have become outdated, thus inefficient and risky) that will take years to untangle, or a startup focused on frontier technology. For the former, big brains who can connect and translate old systems are needed; for the latter, engineers who can check the quality of AI-generated code and direct AI agents are most highly sought.

Code First Girls have adapted their teaching to this new environment. “It’s no longer about teaching what I call perfect syntax,” explains Brailsford, an English literature graduate herself. “All the emphasis previously was on making sure that the writing was perfect. Now the emphasis is on making sure the reading is perfect.”

AI is causing mass confusion and hysteria

Universities are having to enact a similar shift. Professor Aad van Moorsel is head of computer science at the University of Birmingham – which boasts, he says, “the cleverest students” of ‘comp sci’ after Oxford, Cambridge and Imperial. Graduate Llion Jones, for example, became one of the inventors at Google of the transformer (‘T’) aspect of ChatGPT.

“Universities have found it difficult,” he admits. “For a little while, we were discouraging use of AI tools, particularly in places where you had continuous assessment elements, coursework elements. We said, ‘Don’t use GPT, because then you don’t learn anything’. Now, we know you cannot say ‘don’t use GPT’. It just does not work like that.”

Instead, they’ve had to lean in. “We are starting to change our curriculum,” van Moorsel says.

“Where previously we would use a whole year to learn to program, now all the programming is done by AI, and you don’t have to go through that process of actually programming from the bottom up every little piece.”

Students are still expected to do mathematical grounding; to understand networks and databases. But both the teaching and assessment have changed.

As “writing the dissertation is not something you can trust to be done by the students”, this has been reduced from 40 pages to 10. Examinations are now done in two ways: tracking weekly updates of software (“you can still fake it”, he says, but these changes should show progress); and an oral exam, in which they talk to students for half an hour about their projects, to ensure they have understood them.

The transition from writing code to checking and approving it “changes the whole mindset of what we teach”, van Moorsel says. From 2027, Birmingham will offer a new masters in high integrity software engineers. The idea came from the Defence Science and Technology Laboratory, a Ministry of Defence agency, asking the university to take a fresh look at the curriculum.

Trustworthiness is increasingly valued by employers. “If you go up to that level of approving the code that was 80 per cent generated by AI, then suddenly it becomes about, does the board trust this person? That relationship and being a person people trust is suddenly what’s important, and no longer ‘is your code good?’ alone,” van Moorsel says.

Social skills are at a premium too. “For computer scientists, potentially, there’s a challenge there, right? It tends to be an introverted group of people.”

This is where Code First Girls comes into its own: Brailsford describes career-switchers as “our secret weapon”. “They’re becoming bigger than graduate communities. Why is that? When they land, they have client-facing skills. They understand how business works.”

In the age of AI, domain expertise is key. Code First Girls receive applications from NHS workers and GPs, for example. They are taught data science, then placed as a “health and data insights consultant at one of the Big Four”, Brailsford says – bringing skills that cannot be delivered via a graduate scheme.

The CEO agrees that social skills are increasingly prized too: “The most well-rounded candidate is a humanities student that is technically brilliant. Their ability to articulate themselves, their ability to go client-facing – those women are absolute gold dust.”

Concern persists nonetheless – not only that junior developer jobs are under threat but also that cutting them out means the bottom of the career ladder will disappear. What does that do to the quality of code? Are there implications for the future of the internet?

Nick Myers is director of technology at educational platform Digital Theatre, and spent the 2010s leading on the creation of iPlayer for the BBC. He now routinely uses AI tools at work.

“I do get code generated for me by an AI agent. Thankfully, through the experience I’ve had of over 30 years of working, I can spot when it’s doing something wrong or it’s going to have some problems. I can get ahead of it,” he says.

Where previously we would use a whole year to learn to program, now all the programming is done by AI

“If I hadn’t had those 30 years of experience, I wouldn’t be able to do that, because it all looks entirely plausible. It sounds right. It has the right sort of shape.” AI tools can get developers 70 per cent of the way there – but “can’t take it the last 30 per cent of the way safely”.

“The main thing I worry about is my experience of AI so far is that senior people with a lot of systems design experience – a lot of knowledge about how to give clear instructions about what to do and how to validate that – get a lot out of this type of technology. But in order to become senior, you first have to be a junior,” Myers adds.

“I don’t really know, with lots of senior people tooled up with AI technology, where those more junior roles come into play, and how they get there.”

The fear among some is that the internet will be filled with AI “slop” – poor-quality code – as a result.

Frey, the University of Oxford fellow and author of The Technology Trap and How Progress Ends who directs the future of work programme at the Oxford Martin School, is surprisingly optimistic on this question.

“Tools like Mythos are going to make it so easy to expose those kinds of weaknesses that they will need to be fixed very quickly – or that business is going to be out of business. If you produce slop, you’re not going to be around for too long,” he predicts.

“I definitely think there’s going to be a lot of slop being produced in all sorts of domains, but I think the domain I’m least worried about that… would probably be software engineering.”

Because it will be jobs in, say, journalism that are more affected? “And academics,” he adds.

Myers is not convinced. Of Mythos, the new Anthropic model apparently so good at exposing unknown flaws in IT systems that its creator deems it “too dangerous” to release, he wonders, like many software engineers: “How much of that is marketing, and how much of it is real?”

“Can the AI agents doing the coding be made so good that there is less need for expertise on the user end of that, because the agent itself can be that safety handler? I haven’t seen that yet,” he says.

The best outcome for developers is that, like Mr Bucket, they move up the value chain – from writing code to checking and explaining it – and that these new roles emerge fast enough that the AI revolution is not too painful. Cyber security, for example, is considered a growth area for jobs, given the extraordinary abilities of AI-assisted hacking.

The gloomier alternative, however, would see the career ladder shrink, fewer overall jobs, and quality compromised in the long term.

Ultimately, industry will decide whether there will be enough jobs fixing the machines – or whether most engineers are left outside the factory gates.