Agentic
Software
Engineering
The discipline of designing, building, verifying, and governing software with AI agents across the entire development lifecycle.
What does AI-native mean?
AI-native engineers do not just prompt AI to write code. They decompose problems, design agent-ready tasks, provide context, evaluate outputs, build safety checks, review trade-offs, and use evidence to decide whether AI-generated work is trustworthy.
Agentic SWE is not about replacing engineers with AI. It is about becoming an AI-native engineer who can direct agents, verify their work, manage risk, and deliver trustworthy software faster.
Learn what changesDirect agents with clear goals
Specify intent, acceptance criteria, and constraints before an agent acts.
Verify with tests and evidence
Do not trust generated code because it looks plausible. Use CI, static analysis, and review.
Build safety before scaling autonomy
Guardrails, approval gates, and audit trails must come before increased autonomy.
The agentic-swe.dev ecosystem
Six interconnected streams covering every aspect of AI-native software engineering.
Book
Essential software engineering foundations for the agentic era — applying rigorous SE principles when AI agents are in the loop.
Research
Curated and original evidence on AI coding agents — productivity, code quality, safety, hallucination, developer experience, and educational outcomes.
Guardrails
Practical security, governance, and trust patterns for teams adopting agentic workflows. Threat models, catalogue, and CI integration guides.
EdCompass
Resources for educators, students, and organisations building AI-native software engineering capability. Curricula, rubrics, and portfolio guides.
Tools & Patterns
Opinionated patterns for prompting, workflow design, evaluation, review, CI, and governance in agentic software projects.
Manifesto
What we believe about the future of software engineering: human accountability, machine assistance, rigorous evidence, and continuous learning.
10 principles for working with agents
A practical framework — not just for coding faster, but for engineering better.
Think in outcomes, not prompts
Specify intent, acceptance criteria, risks, and dependencies before asking an agent to act. Do not start with "write me code."
Treat AI output as a pull request
Review every AI-generated change like a human contribution: requirement fit, test quality, edge cases, security, architectural consistency.
Engineer trust, not just speed
Know what to automate, what to inspect, what to test, what to reject, and what to own. Trust calibration is the defining skill.
Agentic Software Engineering is about more than using AI to code faster. It is about building trustworthy software systems through human accountability, machine assistance, rigorous evidence, and continuous learning.Read the full manifesto