Skip to main content

Agentic Specification-Driven Development

A framework for building software with AI agents — governed by humans, executed by agents, continuously improved by production learning loops.

✦ v5.0 · Agentic Era 2026✦ 10 Specialized Agents✦ 8-Phase Lifecycle✦ Human-First Governance

Why ASDD?

Traditional agile was designed for human-only teams. ASDD is designed for human-AI squads.

📋
Specification-First
Every feature starts as a machine-interpretable spec. Agents cannot execute on ambiguity — specs eliminate it before a single line of code is written.
🤖
10-Agent Pipeline
Specialized agents handle discovery, validation, design, implementation, QA, security, and learning — each with confidence scoring and human override authority.
🛡️
Human-First Governance
Every phase gate requires human sign-off. Agents propose; humans decide. The Dissent Protocol lets any team member formally reject an agent artifact.
📊
Cumulative Confidence Score
The CCS is the product of all agent confidence scores in a pipeline. If it drops below 0.65, the pipeline halts automatically for human review.
🔄
Production Learning Loop
The Knowledge Agent monitors telemetry, identifies patterns, and proposes steering rule updates — closing the loop between production and your specifications.
Parallel Wave Execution
Implementation tasks run in parallel waves with context-fresh sub-agents, dramatically reducing cycle time while maintaining architectural integrity.

Start where you are

ASDD documentation is organized by role. Jump to what's relevant to you.

CTOs & Engineering Managers
Business outcomes, team model, adoption roadmap
Tech Leads & Architects
Lifecycle, governance, agent pipeline, security
Engineers & Juniors
Your role, how to work with agents, spec writing

The Human–Agent Agile Manifesto

14 principles for the agentic software era. A deliberate evolution of the 2001 Agile Manifesto — written around AI agents, keeping humans in authority.

Validated intentover over generated code
Human authorityover over agent autonomy
Continuous trust loopsover over periodic reviews
Adaptive executionover over rigid planning
Read all 14 principles →