What is ASDD?
Agentic Specification-Driven Development (ASDD) is a software engineering framework for teams that build with AI agents. It replaces informal AI-assisted workflows with a structured pipeline where specifications are the central artifact coordinating both humans and machines.
Version 5.0 · Agentic Era 2026 · Author: Edwin Encinas
The core problem ASDD solves
Most teams using AI in development have the same experience: the AI writes code quickly, but the code is wrong, inconsistent, or doesn't match the intended behavior. The root cause is almost never the AI — it's vague intent.
ASDD enforces that clarity happens before agents execute. A machine-interpretable specification, agreed upon by humans, enables agents to execute correctly at speed.
| Traditional AI-Augmented Dev | ASDD |
|---|---|
| Prompt → Code (hope for the best) | Intent → Spec → Validated spec → Agent execution |
| Hallucinations discovered in code review | Ambiguity eliminated at the spec gate |
| No audit trail of AI decisions | Every agent action is logged with a confidence score |
| AI replaces human judgment | AI proposes; humans govern; both learn |
The five layers
ASDD is organized around five structural layers:
- Product Layer — Where strategy, vision, and intent live. Owned by the Product Owner.
- Specification Layer — Machine-interpretable requirements and domain contracts. The handoff point between humans and agents.
- State Layer — The State Manifest: a centralized heartbeat tracking every slice's position in the pipeline.
- Architecture Layer — Design decisions and Architectural Decision Records (ADRs) synthesized by the Design Agent.
- Agent Execution Layer — 10 specialized agents executing in parallel waves.
- Platform Layer — CI/CD automation, security gates, and observability.
Key concepts at a glance
| Concept | What it means |
|---|---|
| Behavioral Slice | A self-contained unit of work (feature, bug, improvement) that flows through the pipeline independently |
| EARS format | A structured syntax for writing unambiguous, machine-interpretable requirements |
| Phase Gate | A validation checkpoint between lifecycle phases; requires human sign-off |
| Confidence Score | A 0.0–1.0 score emitted by every agent alongside every artifact |
| CCS | Cumulative Confidence Score — the product of all scores in a pipeline path; below 0.65 triggers a halt |
| Dissent Notice | A formal, structured mechanism for any team member to reject an agent artifact |
| State Manifest | .asdd/state/manifest.json — the single source of truth for pipeline state |
| Knowledge Agent | The system memory agent: maintains state, detects conflicts, and proposes improvements |
| Self-Healing PR | An agent-initiated pull request to align code with updated steering rules |
Who is this documentation for?
This site is organized by role. Jump to the section most relevant to you:
🏢 CTOs, VPs, and Engineering Managers
Start with For Leaders — executive overview, team model, ROI, and the maturity model for phased adoption.
🏗️ Tech Leads and Architects
Start with the Playbook for adoption sequencing, then Technical Reference for system architecture and governance.
💻 Engineers and Juniors
Start with Getting Started to understand your role, then the Agent Catalog to see what your AI teammates are doing.
Next steps
- Read the Manifesto → Human–Agent Agile Manifesto — 14 principles for the agentic era
- Understand the lifecycle → ASDD Lifecycle — 8 phases from discovery to production learning
- See the full agent pipeline → Technical Reference — how 10 agents transform intent into code