Code generation and the work of software
Code generation, program synthesis, vulnerability detection and software development appear on 75 recent papers, mostly in AI journals, and 35 of them touch productivity, quality or security. Coding assistants are the most widely adopted generative AI tool at work. What happened to developer output, code quality and the junior developer's job?
What prompted this
Since October 2025, 75 papers carry a code-generation, program-synthesis, coding-assistant, software-development, code-quality or vulnerability-detection keyword (60 in AI journals, 8 in management): "code generation" on 18, "program synthesis" on 6, "smart contract vulnerability detection" on 5, "software development" on 4, "code quality" on 3. Thirty-five of them also carry a productivity, developer, quality, security, bug or labour keyword, but almost none uses data on real teams.
Questions
- In firms that adopted coding assistants, did output per developer rise, and did defect rates, security vulnerabilities and code churn rise with it?
- Did hiring of junior developers fall relative to senior ones, and did pay for review and architecture skills rise?
- Do open-source projects show more contributions and more low-quality pull requests since 2023?
Data that could help
Firm engineering metrics around assistant roll-outs, open-source repository histories, job postings for software roles by seniority, and developer surveys.
Next step
Measure pull request volume, acceptance rates and post-merge bug fixes in a large sample of open-source repositories before and after 2023.