Laurie Voss, founder of npm, has proposed a new interview framework for hiring junior software engineers in the age of AI. The proposal, detailed in a recent blog post, suggests abandoning traditional coding tests in favor of a role-playing exercise where candidates interact with an LLM acting as a customer. Voss argues that the definition of software development has shifted from writing code to defining products, making old interview methods obsolete.

The Collapse of Junior Hiring

The context for this proposal is a stark decline in junior engineering roles. Stanford’s 'Canaries in the Coal Mine' research indicates that employment for developers aged 22 to 25 is down nearly 20% from its late 2022 peak. Anthropic reported that the share of junior engineers in their hiring mix dropped from 27% to 9% since 2025. This trend suggests that companies are increasingly reluctant to hire juniors because they cannot easily assess their ability to work with AI agents.

Why Old Methods Fail

Traditional coding interviews and credentialism are no longer reliable predictors of success. Voss notes that elite school hiring favors those with financial resources rather than skill, while unstructured interviews have a low predictive validity of 0.19 compared to 0.42 for structured ones. The core skills needed now are product sense, system-level intelligence, and the ability to explain requirements in natural language, which current interviews rarely test effectively.

The Proposed Interview Structure

Voss’s solution involves giving candidates an hour to interact with two LLMs. The first model simulates a customer with a complex, vague problem, while the second writes code based on the candidate's instructions. The candidate must extract requirements through natural language dialogue and guide the coding agent to build a working solution. This method tests the ability to define problems and communicate with AI, mirroring the actual job of a 'product engineer.'

Medical Roots and Legal Considerations

This approach is not entirely new; it mirrors the Objective Structured Clinical Examination (OSCE) used in medical schools since the 1960s, where students diagnose standardized patients. Voss emphasizes that while an AI judge can help score the output by running tests, human oversight remains critical to avoid algorithmic bias. Automated hiring tools face legal scrutiny in places like New York City, requiring annual bias audits that make fully automated decisions risky.

Key Takeaways

  • Junior hiring has collapsed due to the inability to assess AI-era skills with traditional coding tests.
  • The proposed interview uses LLMs to simulate customer interactions, testing product sense and communication.
  • Human judgment is required for final scoring to mitigate bias and comply with local hiring laws.
  • This method aims to identify candidates who can define problems and direct AI agents, not just write code.

The Bottom Line

The industry must stop treating coding interviews as the gold standard for junior hires; shifting to AI-driven role-play is the only way to fairly assess the product thinking that now defines the job.