Agentic Software Engineering: Building Trustworthy Software with Stochastic Teammates at Unprecedented Scale

Ahmed E. Hassan argues that agentic coding tools are not the end of software engineering — they are the moment the discipline becomes genuinely essential. His central thesis: Agentic Software Engineering is the discipline of producing trustworthy software from stochastic contributors (AI and human alike) by engineering the full SE system — actors, process, tools, and artifacts — not by hoping for perfect agents.

The book is structured in four parts. Part I establishes the foundation: distinguishing "vibe coding" (fast, informal, useful for exploration) from structured agentic engineering, and framing AI as teammates rather than tools. Hassan introduces the SE 1.0 → 2.0 → 3.0 evolution: classical human-driven loops, copilot-augmented loops, and the agentic era where output explodes and human attention becomes the bottleneck. He catalogues four clusters of AI strengths (tireless/non-judgmental, strong communicator, wide knowledge, near-zero replication cost) mapped to repeatable interaction patterns, and four paradoxes (Eagerness, Context, Tunnel Vision, Learning) that make these teammates simultaneously powerful and challenging.

Part II covers assurance engineering: making stochastic teammates trustworthy through Mission Engineering (structured intent via Mission Briefs, Workflow Runbooks, and Mentorship Packs) and Context Engineering (treating context as a managed interface, not a dump). The artifacts — Mission Brief, Continuity Pack, Mentorship Pack, Workflow Runbook, Consultation Request Pack, Merge-Readiness Pack, Resolution Record — are not bureaucracy; they are the control surfaces that convert intent into trusted change at machine speed.

Part III addresses platform engineering for fleets of AI teammates: coordination engineering (async decision-ready packets, plan-first integration, collision avoidance), workbench engineering (separate human and agent workbenches — SE4Humans vs SE4Agents), capability engineering (role assignment, qualification exams, mentorship-as-code), trust engineering (governance at machine speed, layered verification, policy-as-code), and language engineering (why language choice is a governance decision when reading becomes the bottleneck).

Part IV speaks to specific audiences — developers, technical leaders, business executives, educators, and researchers — about what the SE 3.0 transformation demands of each. The afterword warns of the intensification trap: when the friction of slowness is removed, human cognitive load doesn't decrease; it silently accumulates as people take on more scope and more threads, until burnout.

The book's core refrain: a fool with an agentic tool is still a fool. Speed without an engineered system of constraints and evidence scales mistakes faster than value. The winning teams will not be those who generate the most code — they'll be those who convert stochastic output into deterministic confidence, change after change.