A proposal: how to hire junior engineers in the age of AI

AI-assisted coding has changed software engineering from primarily writing code to product engineering: identifying the real problem, understanding constraints, and directing tools to build a solution. The shift has coincided with a steep decline in junior hiring: employment among software developers aged 22-25 has fallen nearly 20% since its late-2022 peak, and Anthropic reports that juniors’ share of its engineering hires dropped from 27% to 9%. Companies increasingly favor senior staff, but that approach is costly and cannot sustain the industry’s future workforce. Coding interviews test skills that AI now handles, while elite-school credentials, vague qualities such as “hunger,” and informal conversations are poor or biased substitutes. Large internship programs can reveal who performs on the job, but they are expensive and impractical for many employers.

The proposed interview is a structured, job-related simulation. Each candidate receives the same detailed scenario and uses one AI model to role-play a customer and another, standardized model to build software. Within an hour or two, the candidate must ask questions, clarify unstated needs, guide the AI, test the result, and revise or restart as needed; tricking the simulated customer into revealing hidden requirements is disallowed. The exercise could involve complex, domain-specific requirements and produce working, even deployed, software. An AI evaluator could test the output against the full specification, with a human making the final judgment and optionally discussing the result with the candidate. The assessment targets product sense, systems reasoning, and clear written communication - the capabilities needed to work effectively with AI.