AI Jobs in 2031: Who Pays to Train the Next Senior Developers?

On November 4, IT Pro reported that 63% of developers in an Adaptavist survey said AI had reduced what they learned through hands-on coding. The survey covered 1,000 developers across five countries; it records perceptions rather than measured changes in competence. [1]

My five-year worry is a software economy that produces more software while failing to develop enough people who understand it. By 2031, employers could complain about scarce expertise while beginners struggle to find anyone willing to train them.

That is a conditional scenario. I think the decisive question is who pays for the experience future vacancies will demand.

Completing work and learning are different outcomes

In its January 29 research account, Anthropic describes a randomized experiment involving 52 mostly junior engineers learning an unfamiliar Python library. The AI group averaged 50% on an immediate quiz, against 67% without AI. That is a 17-percentage-point gap. The AI group's small speed advantage was not statistically significant. [2]

The company's small experiment measures immediate comprehension. It leaves long-term learning and career progression unresolved. [2]

Still, I think it exposes a serious weakness in judging training through delivery alone. An assignment can be finished without the person completing it gaining the understanding a future assignment will require. If employers reward only fast completion, learning could become something workers must squeeze in elsewhere.

I would treat that possibility as a business problem. A company planning to depend on human oversight should also plan how people become capable of providing it.

2031 if employers keep buying experience

CNA's October 27 report, updated November 6, says Tata Consultancy Services' graduate intake in financial year 2026 was less than half its level four years earlier, citing company reports. Interviewees also describe employers expecting greater technical depth from new graduates. The reporting does not isolate AI's contribution to the recruitment decline. [3]

Now imagine that narrower entry route persisting across successive graduating classes. In this first 2031 scenario, firms use AI effectively and recruit experienced engineers when difficult work arises. Each employer avoids some training costs. Collectively, they could become increasingly dependent on expertise they are doing too little to replenish.

Recruiting from a competitor moves experience between employers. It does not, by itself, create another experienced person. I think this is the flaw in treating experienced hiring as a complete workforce strategy.

The resulting job market could be cruel: vacancies demanding judgment alongside applicants denied opportunities to develop it. A shortage of qualified people would not automatically help someone who still cannot qualify.

I would include experienced workers in this risk. IT Pro reports that 45% of experienced respondents perceived fewer learning opportunities because of AI. [1] In my view, renewing expertise includes helping existing engineers learn unfamiliar systems.

2031 if automation comes with an apprenticeship

A different choice is visible in advertised job design. Enfield Council posted an AI and Automation Apprentice vacancy on November 3: a full-time contract paying £31,290–£33,129 annually. The duties combine process analysis, supervised implementation, testing and apprenticeship reviews. [4]

This advertisement establishes intended work. It supplies no evidence that someone has been hired, received effective supervision or progressed afterward. Its public-sector automation role also cannot stand in for every software engineering job. [4]

Nevertheless, I think it illustrates a practical alternative: pay a beginner to help automate work while learning how to assess the result.

In a second 2031 scenario, employers fund successive groups of learners and give experienced staff time to teach. Beginners use AI, investigate defects and gradually take responsibility for more consequential decisions. Productivity gains help support that development.

That arrangement costs money and attention. It could produce less immediate output than assigning everything to the fastest available combination of experts and software. I would accept that tradeoff where it develops people the organization expects to need later.

CNA also reports that TCS and Infosys say they have revised graduate training to include generative AI. Those are company claims about provision; their effectiveness remains unproven in this reporting. [3]

Better AI could change both futures

Neither scenario proves a looming shortage. If AI becomes reliable enough to handle more advanced decisions, employers might need fewer human experts too. That could make the employment outlook harsher even as the software economy expands. I would not promise beginners that senior positions must eventually await them.

Conversely, AI could improve learning. Anthropic found that requests for explanations and conceptual questions were associated with stronger immediate results among some participants. Those exploratory comparisons did not establish causation. [2]

My preferred future therefore leaves room for different ways of learning. Preserving opportunities to build and test independent understanding matters more to me than preserving every old coding exercise.

There is also a possible access problem. If meaningful practice increasingly happens before employment, people able to spend longer preparing without pay could have an advantage. That is a risk I infer, rather than an outcome these sources measure. Employer-funded learning would help keep that burden from falling entirely on applicants.

Choose a route that creates experience

For someone planning a career, I would widen the search beyond job titles to examine the work offered. An automation apprenticeship that includes supervised investigation may provide a more credible learning route than a junior title attached to repetitive acceptance of generated output. The actual supervision and progression matter more than the label.

I would track whether responsibilities grow: whether you get to diagnose harder problems, make decisions and receive informed feedback. A stream of completed tasks can look impressive while leaving that progression uncertain.

By 2031, I expect employers' training choices to matter alongside improvements in AI capability. Software provides a bounded example; these sources cannot forecast the whole economy. But I have little faith in a workforce strategy that keeps demanding experience while assuming somebody else will pay to create it.