Method
How SDD Works
From waterfall to specification-driven AI execution — the complete technical breakdown.
Evolution
How Software Development Evolved
Four eras of building software — from rigid waterfalls to specification-driven AI agents.
Waterfall
Sequential phases: requirements, design, implementation, testing, deployment. Each stage completed before the next begins. Changes are expensive; feedback loops are measured in months.
Predictable but slow. A change request after the design phase could reset the entire timeline.
Agile
Iterative sprints, user stories, daily standups, retrospectives. Teams deliver incrementally and adapt to change. But velocity is still bounded by team size and coordination overhead.
Faster feedback, but scaling requires more people — and more people require more coordination.
AI-Assisted
Code completion, AI copilots, chat-based code generation. Individual developers become faster, but the development process itself remains unchanged. AI assists; humans still do the work.
Productivity gains of 20–40% per developer — but the process, roles, and bottlenecks stay the same.
SDD
The specification becomes the executable contract. AI agents implement, test, document, and deploy — orchestrated by a senior engineer who controls the entire lifecycle through formal specs.
Not faster developers — a fundamentally different process. One engineer, specification-first, AI-executed. Core Concept
What Is Specification-Driven Development?
A methodology where the specification is the source of truth and AI agents are the workforce.
Specification-Driven Development (SDD) is a methodology where every feature, service, and integration begins as a formal, machine-readable specification — before any code is written.
Unlike traditional approaches where developers interpret requirements and make implementation decisions on the fly, SDD front-loads all architectural and business decisions into the spec. The specification is not documentation — it is the source of truth that AI agents execute against.
The result: predictable output, consistent quality, and a development velocity that scales with specification depth — not team size.
Specification is Code
Every feature starts as a formal spec: API contracts, data models, business rules, acceptance criteria. The spec is precise enough for AI agents to implement without ambiguity.
Agents Do the Work
AI agents handle implementation, testing, documentation, CI/CD configuration, and localisation. Each agent operates within a defined role and strict context boundaries.
Human Stays in Control
A senior engineer writes the specs, reviews all output, makes architectural decisions, and validates quality. AI amplifies expertise — it does not replace judgement.
Architecture
Three Layers of Context
Every AI agent operates within a strict context hierarchy that ensures consistency and quality.
System Context
Architecture rules, technology stack constraints, coding standards, naming conventions, and project-wide invariants. This context is loaded into every agent session and never changes mid-project.
Example: "Java 21, Spring Boot 3.2, PostgreSQL. All services use hexagonal architecture. REST APIs follow OpenAPI 3.1 spec. No ORM — raw SQL with jOOQ."
Feature Context
Requirements, acceptance criteria, API contracts, data models, and business rules for a specific feature. This context is scoped to the current task and defines what the agent must build.
Example: "Certificate issuance service: POST /api/v1/certificates. Validates warehouse capacity, checks duplicate serial numbers, emits CertificateIssued domain event."
Execution Context
The current file, function scope, test expectations, and immediate dependencies. This is the narrowest context layer — it tells the agent exactly where it is and what to produce next.
Example: "Implement CertificateService.issue() method. Input: IssueCertificateCommand. Output: CertificateDTO. Must pass: CertificateServiceTest lines 45–78."
Lifecycle
The SDD Lifecycle
Seven steps from specification to production — with human checkpoints at every critical stage.
Write System Specification
Define architecture, tech stack, coding standards, API conventions, and project-wide rules. This becomes the immutable system context for all agents.
Define Feature Requirements
For each feature: API contracts, data models, business rules, acceptance criteria, and test expectations. The spec must be precise enough for unambiguous implementation.
Human Review: Specification
The engineer reviews the spec for completeness, consistency, and architectural soundness. This is the most critical checkpoint — errors here propagate everywhere.
AI Agents Implement
Agents execute the spec: write code, create tests, generate documentation, configure CI/CD. Each agent works within its role and context boundaries.
Automated Validation
All generated code is compiled, tested, and linted automatically. Test coverage, type safety, and coding standards are enforced by CI pipelines — not by humans.
Human Review: Output
The engineer reviews generated code for correctness, edge cases, security, and production-readiness. AI output is never deployed without human validation.
Deploy to Production
Approved code is merged, pipelines run, and the system is deployed. Full traceability from spec to commit to deployment.
Ready to Start?
Let's Build Something Real
NDA first. Then a clear specification, fixed price, and a working system — delivered in weeks, not months.