The Methodology

Specification First
Agentic Development

The Operating Model for Successful Agentic Projects.

“SFAD starts with the use case — not the code. You define the problem, validate the constraints, lock the specification. Only then does the AI touch it. The human's job is to approve, not to guess. The AI's job is to execute within what's been agreed, not to fill in the gaps.”

— Brendan Small

Think of SFAD as the engine room of an AI Lab — it takes a raw business problem, runs it through a structured incubation process, and hands a governed, production-ready solution out the other end. Use cases don't just inform the build. They become it.

The full SFAD workflow

Five phases, three automated feedback loops, human gates at every critical decision point. Click any blue Specification Phase node for full detail — including artefact examples. Deeper phases available on request.

Specification PhaseAgent Planning & Code GenerationCI/CD PipelineFeedback & OptimisationReview & ProductionLoop 1Loop 2Loop 3🔒Contact for detail

Click any blue Specification Phase node for full detail · All other phases locked

Why SFAD over vibe coding?

Vibe coding moves fast at the start. SFAD moves fast at the start and keeps moving fast. The difference is the gate.

Development Methodologies Compared

Dimension
Vibe Coding
Spec-First Agentic Development
Specification
Loose, evolves during coding
Precise markdown, defined upfront
Documentation
Often written after (or skipped)
Source of truth, guides execution
Implementation
Developer intuition, ad-hoc
Agentic automation within guardrails
Testing
Manual, inconsistent coverage
Automated (unit, integration, UI)
Debugging
Manual investigation required
Agent debugs, fixes, re-tests
Validation
Developer review before deploy
Automated + human review gate
Scalability
Struggles as complexity grows
Scales with clear structure
Time to Market
Fast initially, slows over time
Fast initially + sustained
Developer Role
Writes & reviews all code
Specifies & validates output

Full methodology detail available on request

Contact Brendan →

The evidence

20x faster than traditional development.

Not a prototype. Not an MVP.

An AI production scheduling platform — ingesting all production inputs, generating and optimising schedules using a CP-SAT constraint solver, and surfacing AI-driven production recommendations — delivered by a single developer in under 4 weeks.

A conventional development team would estimate the same system at 6–12 months. The 20x figure is conservative. The system is live in production.

See all case studies →

Want to implement SFAD in your organisation?

Get in touch to discuss adoption, team training, or senior AI leadership roles.