Research
Curated and original evidence on AI coding agents. Follow the research to make evidence-based decisions about agentic workflows.
Evidence map
Key research categories we track, with annotated studies on the most important questions.
Productivity and speed
Studies on developer velocity with AI coding agents. How much faster, and under what conditions?
Code quality and defect rates
Evidence on whether AI-generated code introduces more bugs, security vulnerabilities, or technical debt.
Security and vulnerability introduction
Research on how agentic tools affect the security posture of software projects.
Developer experience and cognitive load
How agents affect developer flow, trust, skill development, and job satisfaction.
Educational outcomes and learning effects
What happens to learning when students use AI agents in software engineering courses?
Hallucination and reliability
How often do agents produce plausible but incorrect code, tests, or documentation?
Open research agenda
Questions that matter for practitioners but lack strong empirical answers today.
When do agents increase defect rates, and what factors predict this?
What verification strategies work at scale without eliminating the speed benefit?
How does agentic SWE affect the growth trajectory of junior developers?
What guardrail patterns improve security outcomes without killing throughput?
How should education assess AI-native engineering competency?
What is the long-term effect on software architectural quality under AI acceleration?
Working papers
Working papers and preprints will appear here as they are released.
Research digest
A curated summary of new evidence in agentic SE — delivered when there is something worth reading.
Newsletter signup coming soon.