Skip to content

The method we deliver with

The G.U.I.D.E. Method

The G.U.I.D.E. method is a five-phase approach to governing AI-assisted software delivery: Goals and Guardrails, Understand, Integrate, Deliver, Evaluate. Every phase ends with a person approving what happens next.

Where G.U.I.D.E. sits

An AI-native SDLC playbook: a governed method in which AI amplifies human judgement.

Vibe coding

Unreliable, unstructured, no governance.

G.U.I.D.E.

Governed. Production-grade.

The structured middle ground for real enterprise AI development.

AI fragmented

Narrow tool adoption, no SDLC integration.

Further reading

What Is an AI-Native SDLC? A 2026 PlaybookAn AI-native SDLC brings AI into every phase of delivery under governance. Here's why coding assistants alone stall, and what actually changes delivery.Read the article

What it is

G.U.I.D.E. is a harness for AI coding agents: the rules, templates and checks wrapped around the agent, so that it works the way your team does and every decision is recorded. It works with all the popular coding agents, including Claude Code, Cursor, Kiro, Google Antigravity, Windsurf and OpenAI Codex.

It works at two levels.

The engagement

Moves through five phases, from agreeing the goals to measuring the result, and a person signs off each one.

The five phases

Every change

Runs through the Tight Loop: AI Plans, Seeks Human Guidance, AI Implements, Human Clarifies. A typo fix moves fast; a change to a payment flow gets full requirements and design first.

The Tight Loop

Every prompt and every decision is logged with a timestamp, so an auditor can see who agreed to what, and when.

See it run:

Using G.U.I.D.E. to generate enterprise-ready BRDs, HLDs and LLDs with AI.
Greenfield build · G.U.I.D.E. end to end
Brownfield change · G.U.I.D.E. on an existing codebase
Brownfield change · G.U.I.D.E. on an existing codebase
A good harness should not necessarily aim to fully eliminate human input, but to direct it to where our input is most important.
Birgitta Böckeler, Distinguished Engineer, Thoughtworks
martinfowler.com, February 2026

The five phases

Each phase ends at a gate: nothing moves on until the named person signs it off.

  1. G

    Phase 1

    Goals and Guardrails

    What happens
    Agree the target outcome, the success metric and the baseline, and define what AI may and may not touch.
    Who is involved
    Your sponsor, your technical lead and our engagement lead.
    Deliverables
    • A one-page charter with baseline figures and guardrails

    Sign-off: Your sponsor signs the charter.

  2. U

    Phase 2

    Understand

    What happens
    Reverse-engineer the codebase or process, map systems, dependencies and risks, and rank candidate use cases.
    Who is involved
    Our engineers, working with your team.
    Deliverables
    • System map
    • Risk register
    • Ranked list of use cases

    Sign-off: Your technical lead confirms the system map is accurate.

  3. I

    Phase 3

    Integrate

    What happens
    Set up the harness in your repositories and pipelines, configure role-specific workflows, and run the first change through the loop.
    Who is involved
    Our engineers and your developers.
    Deliverables
    • Configured harness, with rules and prompts
    • The first merged change

    Sign-off: The first change passes your code review and your CI pipeline.

  4. D

    Phase 4

    Deliver

    What happens
    Ship backlog items through the governed loop, with tests, documentation and reviews generated and checked.
    Who is involved
    Your team, with ours alongside.
    Deliverables
    • Merged features
    • Test suites
    • Requirement and decision records

    Sign-off: A person approves every change, and releases follow your process.

  5. E

    Phase 5

    Evaluate

    What happens
    Re-measure against the baseline using DORA, SPACE and DevEx, the industry-standard measures of delivery speed, quality and developer experience, then decide whether to scale, continue or stop.
    Who is involved
    Your sponsor, your technical lead and our engagement lead.
    Deliverables
    • Impact report

    Sign-off: Your sponsor reviews the impact report, which opens the next cycle or the handover.

The Tight Loop

G.U.I.D.E.Tight LoopAIPlansSEEKSHuman GuidanceAIImplementsHUMANClarifies

Selective Intelligence

Executes only value-adding steps per request. No unnecessary automation — every AI action is deliberate and purposeful.

Code-Aware

Analyzes existing codebase, architecture patterns, and required complexity before proposing changes. Context-first, not prompt-first.

Risk-Aware

Deep handling for high-risk operations, fast handling for low-risk ones. The framework adjusts its governance depth based on blast radius.

Controlled Execution

Each phase runs only after human review and approval. No autonomous code merges, no unreviewed deployments. Humans stay in the loop.

See it on your codebase.

A 30-minute call to map the method to your current workflow and pick the first change to run through it.