Methodology for AI-assisted development

The antidote to vibe coding.
Minimal specs.
Real value.

$ micro-sdd init "stop detection"
→ Spec created: 3 acceptance tests
→ Vertical slice identified
→ Ready to implement

Micro SDD is a methodology for development guided by minimal specs, designed for teams using AI. You don't assume the solution — you discover it through short, verifiable, value-oriented cycles.

Read the article See the principles
The problem

AI generates code fast. It doesn't guarantee value.

Vibe coding produces the illusion of productivity: activity without direction, speed without destination.

✕ Vibe coding
✕ Prompt → code → see what happens
✕ No tests, no acceptance criteria
✕ Invisible technical debt from day 1
✕ Architecture nobody understands in 2 weeks
✕ Activity metrics, not delivery metrics
✓ Micro SDD
✓ Minimal spec → test → guided implementation
✓ Acceptance criteria before writing code
✓ Each cycle delivers verifiable value
✓ Emergent architecture, not presumed
✓ Progress measured in completed slices
Foundations

Five principles. One system.

Each principle resolves a real tension in AI-assisted development. They're not suggestions — they're invariants.

P0
The Minimum Viable Spec
The smallest possible spec that delivers verifiable value. You don't assume the solution — you define the criterion before writing a single line. It's the foundation all other principles rest on: without a clear minimal spec, there's no direction, only speed.
P1
Complete Vertical Slice
Each delivery crosses all layers: from UI to persistence. You never split by technical layer. Value is delivered whole or not delivered at all.
P2
Tests First, Always
ATDD defines the acceptance criterion. Unit TDD guides the implementation from the outside in. The test is the executable specification.
P3
Continuous Discovery
The solution emerges from the process, not from a prior design. Each cycle contributes knowledge that changes the next specs. You don't presume — you discover.
P4
Integrated Feedback
The cycle doesn't end at the merge. Feedback from the real environment feeds the next spec. The system learns from every delivery.
The cycle

From idea to verifiable value in short cycles.

Every Micro SDD iteration follows the same pattern — small, verifiable, complete.

01
Identify the slice
Define the minimal unit of value. A spec is a vertical slice, never a technical layer.
02
Write the spec
The spec defines the acceptance criterion before any implementation. Tests first.
03
Implement with AI
AI implements guided by the tests. It doesn't guess — it follows an executable specification.
04
Verify and learn
Tests pass. Feedback updates the map of candidate specs for the next cycle.
In practice
stop-detection.spec
// Spec: Stop detection in route
// Vertical slice — P0 minimum viable

Feature: Stop detection

  Scenario: Stop correctly detected
    Given a GPS with coordinates every 30s
    When vehicle stays < 50m for 5min
    Then a stop is recorded with timestamp

  Scenario: Continuous movement yields no stop
    Given average speed > 15 km/h
    Then no stop is recorded

// 2 acceptance tests · 1 slice · 0 assumptions
Spec anatomy
Small because the best path is discovered, not presumed.

This spec doesn't say how to detect the stop — it says when a stop is considered to have occurred. The implementation is free. AI has real context, not assumptions.

Two scenarios. One complete vertical slice. Clear acceptance criterion before writing a single line of production code.

ATDD vertical slice P0 applied no estimates executable criterion
Philosophy

"The small spec is not a limitation — it's a recognition of honesty: you don't know the solution until you discover it."

— Micro SDD · Principle of continuous discovery
Ready to start?

Stop guessing. Start discovering.

Read the founding article and understand why Micro SDD is the missing link between agility and AI-assisted development.

Read the founding article See the principles in detail