Microsoft Execs: Companies Must Continue Entry-Level Hiring

Even if it means an initial productivity hit.

Senior figures at Microsoft have warned about AI’s potential to erode the foundations of software engineering.

In a new paper, Redefining the Engineering Profession for AI, CTO of Microsoft Azure Mark Russinovich and VP of developer community Scott Hanselman argue that AI-powered coding assistants are disproportionately helping experienced engineers while making life more difficult for those at the start of their careers.

The pair say agentic AI coding tools can give senior engineers a productivity boost, but create an "AI drag" for early-in-career (EiC) developers, who must spend time steering, verifying and integrating the systems' output.

Speaking previously on a company podcast, Russinovich described the issue as a "hot topic" in discussions with customers, saying many firms report the same skewed productivity phenomenon.

Russinovich and Hanselman’s paper highlights a range of problems they say are common in AI-generated code, including “significant” bugs, inefficient algorithms, duplicated code scattered across projects and debug routines left behind in production systems.

In some cases, the tools may declare success despite underlying faults. The authors cite an example of an AI agent attempting to fix a race condition – a type of concurrency bug – by inserting a Thread. Sleep delay.

This simply masks the problem and could introduce new risks.

"Only an engineer familiar with synchronization code has the confidence to point out the agent's mistakes," they write.

The concern is that companies observing these patterns may conclude it is more efficient to hire fewer junior staff and instead rely on experienced engineers who can better supervise AI systems.

Research from Harvard University, cited in the paper, found that firms adopting genAI saw sharp declines in junior hiring compared with non-adopting companies, while senior employment remained largely unchanged.

Industry data appears to point in a similar direction.

Venture capital firm SignalFire, in its 2025 State of Talent Report, found that new graduate hiring has fallen significantly across the tech sector.

According to the report, recent graduates accounted for just 7% of hires at large technology companies in 2024, down 25% from the previous year.

Calls For Mentoring

Russinovich and Hanselman argue against this short-term approach. They argue that large companies must continue hiring early-career engineers, even if they initially reduce productivity, and make mentorship an explicit organizational goal.

They propose a "preceptor-based" model, formally pairing senior engineers and junior colleagues to guide both them and the AI systems they use.

Another suggestion is the development of a special "EiC mode" in coding assistants, designed to provide coaching rather than simply outputting solutions.

However, given the types of errors identified in the paper, the authors acknowledge such systems may not always be reliable mentors.

The paper represents the personal views of its authors rather than official Microsoft research. It is also unclear how far Microsoft has implemented the recommendations.

Last year, the company announced job cuts that included reductions in software engineering roles. On the podcast, Russinovich said Microsoft was "starting a pilot" based on the paper's ideas.

The debate comes amid broader concerns about AI's impact on white-collar work.

Earlier this month, tech entrepreneur Matt Shumer, who founded OthersideAI, warned that AI could wipe out as many as half of all entry-level white-collar roles within the next five years (we disputed that claim last week).

In his essay Something Big Is Happening, he compared the current moment in AI to early 2020, just before the Covid-19 pandemic reshaped global economies.

Shumer said AI systems at his company can now independently write, edit and debug code – tasks that previously required teams of engineers and hours of back-and-forth collaboration.

This article originally appeared on MES Computing’s sister site Computing.