Templates, not client work
Every template in the library is a process pattern. Client information never leaves the engagement; what we publish is generic, reproducible, and yours to take.
Methodology Library · 94 templates · Updated monthly
The AI Methodology Library
Most AI agencies say “we use AI” and stop there. We publish the prompts, the models, the phases of the build they apply to, and the human time each one replaces. Take any template into your own team.
0
prompt templates published, updated monthly
0.0×
faster vs 2022 baseline
0%
senior-reviewed
Four principles
How the library is built
Model named, phase declared
Each template lists the model it was tested against and the phase of the build it belongs to. No "AI did it" handwave.
Human time replaced, measured
Every template carries an estimate of human time it replaces. Where the estimate is fuzzy, we say so.
Updated monthly, with changelog
Models evolve. We re-test the library each month, retire templates that no longer beat their replacements, and publish the diff.
Categories
Seven categories, 94 templates
Each category lives in the library with its own changelog. The largest are Engineering and Design, because that is where AI moves the needle most.
12 templates
Discovery
Templates that turn an idea into a scoped build before contracts.
9 templates
Research
Competitor matrices, ICP synthesis, market sizing.
26 templates
Engineering
Code review, test generation, migration drafts, doc first-pass.
14 templates
Design
UX copy, layout exploration, design-system audits.
11 templates
Process
Retro synthesis, weekly Loom scripting, decision logs.
8 templates
Operations
Status pages, incident drafts, customer comms.
14 templates
Sales-ready
Pitch decks first-pass, proposal scaffolding, audit reports.
Featured templates
Six examples from the library
The full template includes parameters, tested model, expected output shape, and the human time it typically replaces. Click through to the full library to see all 94.
DiscoveryClaude Sonnet 4
Discovery session structureRuns before every kickoff
You are a senior product discovery facilitator.Your goal is to surface {N} risk areas for {product_type}before kickoff. Use the {framework} method.For each risk: 1. Restate the user-visible symptom 2. Hypothesize the underlying cause 3. Score likelihood × impact (1–5)
ResearchGPT-4o
Competitor feature-matrixSynthesises 12+ competitor pages
Compare these {N} competitors on {axes}.Output a markdown table where rows are featuresand columns are competitors.For each cell, mark: ✓ has it, ✗ doesn't, ~ partial.Cite the source URL for every ✓ claim.
EngineeringClaude
Code review checklistFirst-pass on every PR
Review this PR against our checklist: - Security: input validation, authz boundaries, secret leaks - Performance: N+1 queries, unnecessary re-renders, blocking I/O - Maintainability: naming, dead code, test coverageSurface only issues. Not style nits.
ProcessClaude Sonnet
Sprint retrospective synthesisTurns 45 min of notes into 8 min
Synthesize this retrospective transcript.Cluster issues by theme. Each cluster: - Theme name - Severity (1–5) - Concrete next-action with ownerSkip rumination. Output ready-to-ship Notion doc.
DesignClaude + designer
UX copy generationFirst draft, designer approves
Write UX copy for {component} for {persona}.Constraints: - Voice: {voice_token} - Max 6 words headline, 16 words body - No jargon, no superlatives - Every CTA verb-led
EngineeringGPT-4o
API documentation first passFrom OpenAPI to readable docs
Convert this OpenAPI spec into developer docs.For every endpoint: - One-line purpose - 3-line minimal cURL example - Common errors with HTTP code + reasonNo marketing fluff. Engineers reading at 2am.
Where AI accelerates our work
- Boilerplate code (components, CRUD, tests). 40% of lines written with AI assistance.
- Discovery prep (competitor research, ICP analysis). Saves 2 to 3h per client.
- First-pass code review. Catches style issues, obvious bugs, missing edge cases before human review.
- Documentation drafts. 60% time reduction. A senior edits, doesn't write from scratch.
- Sprint retro synthesis. 45 minutes of notes turn into structured actions in 8 minutes.
Where we don't use AI, and why
- Load-bearing architecture decisions. A senior engineer makes these, documented in writing.
- Security-sensitive code (auth, payments, encryption). Human-only, no AI suggestions merged without full manual review.
- Final client-facing copy. AI drafts, a human approves every word.
- Anything with your confidential business logic. It stays in your codebase, never in a shared AI context.
FAQ
Common questions on the library
Why publish your AI methodology?
A claim without receipts is a pitch. The methodology library lets a prospect verify the work pattern before signing a contract, and lets a client reuse the templates after delivery without contacting us. It is the most reliable signal of how we actually build.
Are these templates safe to use with client data?
Yes. The templates are generic process patterns. They do not contain client data and they describe how to structure a prompt, not what to put inside it. We use the same templates across engagements with different data.
What's in the changelog?
Model swaps (e.g., a code-review prompt migrating from one Claude version to another), new templates added, retired templates with the reason, and process changes (e.g., regression prompts moving to every PR rather than release branches).
Can I take a template into my own team?
Yes. Every template is published under a permissive license. Clone, fork, adapt. We track adoption signals to know which templates need polishing or replacement.
How does this differ from a typical "AI playbook"?
Typical AI playbooks are slide decks listing tools. This library names prompts, the model used, the phase applied, and the human time replaced. It can be diffed quarter over quarter.