We believe software engineering is entering an agentic era.
AI agents will increasingly plan, code, test, review, and maintain software. But software quality still depends on engineering judgement.
The future belongs to engineers who can collaborate with agents responsibly: giving clear goals, providing context, verifying outputs, managing risks, and building guardrails.
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.
Five core beliefs
Speed without trustworthiness is not engineering.
Producing more code faster has value only when that code is correct, secure, and maintainable. Velocity that erodes quality or introduces undetected risk is technical debt accumulation at scale.
Accountability belongs to engineers, not agents.
AI agents are tools, not authors. When agentic software fails — in production, in security, in correctness — the accountable party is the engineer who directed, accepted, and deployed that work.
Evidence replaces assumption.
Plausible-looking code is not verified code. Claims about productivity, quality, and safety of agentic workflows must be grounded in empirical evidence: tests, static analysis, benchmarks, user outcomes, and honest measurement.
Safety must be designed in, not bolted on.
Guardrails, permission models, audit trails, and human approval gates are engineering requirements for any system where agents take consequential actions. They must be built before autonomy is scaled.
Education must evolve alongside the tools.
Teaching students to ship code with AI is not enough. The next generation of engineers must learn how to decompose problems for agents, evaluate AI-generated work, design verification strategies, and exercise professional judgement in an agentic world.
Ready to put this into practice?
The Playbook turns these beliefs into 10 actionable principles.
Read the Playbook