Skip to content

AWS Working Backwards

Independent interpretation

This page is ProductBuildersHQ's own interpretation of Amazon's publicly described Working Backwards practice, adapted for AI-assisted spec authoring. It is not affiliated with, endorsed by, or sponsored by Amazon.com, Inc.

Amazon's Working Backwards methodology starts with the customer and works backward to the solution. It emphasizes writing as thinking, with the PR/FAQ and 6-Pager as core artifacts.

Profile Variants

Both profiles start the same way — with the human-authored Press Release, the founding artifact of Working Backwards. Amazon classifies decisions by reversibility (one-way vs. two-way doors), not by a product/feature taxonomy — so the profiles are named by the door. The door sets the ceremony; the scale picks the deepening tool (both are optional in both profiles):

Profile Founding Document Ceremony Typical For
aws-one-way-door Press Release 6-pager decision narrative required; formal decision meeting gates build spend Hard-to-reverse bets: new product lines, public commitments, major investments
aws-two-way-door Press Release Iterative PR/FAQ review is the only formal gate; 6-pager optional Reversible bets: features, experiments, incremental improvements

Optional post-FAQ deepening in either profile: MRD (market validation — typical at product scale) or OpportunitySpec (discovery canvas — typical at feature scale). For the latter, see OpportunitySpec Framework.

The Flow

One-Way Door Flow (PR/FAQ Start)

AWS One-Way Door Flow

Two-Way Door Flow (PR/FAQ Start)

AWS Two-Way Door Flow

Key Principles

  1. Customer Obsession: Start with the customer and work backwards
  2. Write First: Document the vision before building
  3. PR/FAQ: Write the press release announcing the product's launch
  4. Challenge Assumptions: Use FAQ to surface gaps and concerns
  5. Narrative-Driven: Use 6-pagers for decision-making
  6. Iterate: The PR/FAQ goes through many drafts — iteration is the method, not a failure of the first pass

AI-Native Working Backwards

The methodology's division of labor maps cleanly onto AI-native building — and gets stronger, because AI makes the expensive parts of authenticity cheap:

Working Backwards element Who does it Why
Press Release Human The vision seed — taste and intent are the inputs AI cannot supply. Written first, by hand, on purpose.
FAQ Agents synthesize, human curates Agents research the market, competitors, and constraints to stress-test the PR's claims; the human decides which answers survive.
Desirability evidence Agents build, humans watch users A clickable demo of the announced experience (hardcoded data, no backend) is minutes of agent work. Attach it to the PR appendix — "watch people try it" beats "we believe."
Feasibility evidence Agents spike The FAQ's "can we build it?" answers cite a vertical-slice proof-of-concept against real data — hours of agent work now, not weeks. Assertion-free feasibility.
Evaluation LLM-as-a-Judge Each document is scored against the profile's leadership-principle rubrics between drafts, so iteration has a feedback signal cheaper than a review meeting.
PR/FAQ review, decision meeting Human The gates stay human. The 6-pager review is the one-way-door judgment call — precisely the decision AI evidence can inform but not make.

Two disciplines keep this honest:

  • Label the evidence. AI makes artifacts look real for free, so every appendix item is marked by what it proves — a mockup illustrates, a demo tests desirability, a spike proves feasibility. Never let a facade cash a working-system claim.
  • Iterate before the gate, not after. Cheap synthesis and cheap evaluation mean the PR/FAQ can go through its ten drafts in days. Spend the iterations where Amazon always did — on the vision documents — so the one-way-door meeting decides on a hardened narrative, and the post-gate build (UXD, TRD, TPD) starts from settled intent.

VisionSpec Mapping

AWS Artifact VisionSpec Type Purpose
Press Release Press Founding artifact — human-authored vision announcing the solution; written first
FAQ FAQ Challenges claims, surfaces gaps; absorbs the business case
Business Case deep-dive MRD (optional) Post-FAQ market validation when the risk warrants it (Amazon has no MRD; the FAQ carries the case)
PRFAQ Combined PRD Product requirements derived from narrative
6-Pager Narrative Decision document for the one-way-door gate
Design Doc TRD Technical architecture
Test Plan TPD Test cases, automation, quality gates
Ops Review IRD Infrastructure and operations

Using the aws-one-way-door Profile

Initialize a Project

multispec init my-product --profile aws-one-way-door

Write the Press Release (Founding Artifact)

multispec draft press -p my-product

The press release is human-authored — it is the vision seed, not a synthesis product. Write it first; everything else works backward from it.

Synthesize Working Backwards Documents

# Generate FAQ challenging the Press Release
multispec synthesize faq -p my-product

# Optional: deepen the FAQ's business case with market validation
# (or `synthesize opportunity-spec` for the feature-scale discovery canvas)
multispec synthesize mrd -p my-product

# Generate PRD from Working Backwards artifacts
multispec synthesize prd -p my-product

# Generate 6-Pager Narrative for the decision gate
multispec synthesize narrative-6p -p my-product

# Generate technical specs (post-gate)
multispec synthesize trd -p my-product
multispec synthesize tpd -p my-product  # Test plan
multispec synthesize ird -p my-product  # Infrastructure

Evaluate Documents

# Evaluate Press Release against Leadership Principles
multispec eval press -p my-product

# Check readiness
multispec status -p my-product

Templates

Press Release Template

The Press Release template includes:

  • Headline: One sentence capturing customer benefit
  • Subheadline: Target customer and key value
  • Problem: Customer pain point
  • Solution: How the product solves it
  • Quote (Spokesperson): Internal vision
  • Customer Journey: How to get started
  • Quote (Customer): External validation
  • Call to Action: Next steps

FAQ Template

The FAQ template covers:

  • Customer Questions: What customers will ask
  • Internal Questions: Stakeholder concerns
  • Technical Questions: Implementation challenges
  • Business Questions: Market and financial viability

6-Pager Template

The 6-Pager template follows Amazon's structure:

  • Introduction and context
  • Goals and tenets
  • State of the business
  • Lessons learned
  • Strategic priorities
  • Appendix with data

Rubric Categories

Each document type carries its own LLM-as-a-Judge rubric (structured-evaluation format), grounded in the Amazon Leadership Principles — all 16 are carried in the profiles' methodology metadata, and the most judgeable teachings are encoded as rubric categories: economic sustainability in the press rubric (Customer Obsession includes economics), disconfirmation rigor in the FAQ rubric (Are Right, A Lot: work to disconfirm; Earn Trust: vocally self-critical), the ownership horizon in the 6-pager rubric (the personal-money test), and door-specific decision-reversibility criteria in the PRD rubric (Bias for Action). Cross-cutting themes:

Category Weight Description
Working Backwards Fidelity 20% Starts with customer, not technology
Customer Clarity 20% Specific customer segment and problem
Decision Reversibility 15% Two-way vs one-way door decisions
Bias for Action 15% Speed of execution considered
Long-Term Thinking 15% Sustainable competitive advantage
Frugality 10% Resource efficiency
Deep Dive 5% Data-driven analysis

Example Workflow (aws-one-way-door)

# 1. Initialize project
multispec init checkout-redesign --profile aws-one-way-door

# 2. Human writes the Press Release — the founding artifact
multispec draft press -p checkout-redesign
# ... iterate on the announced customer experience ...
multispec eval press -p checkout-redesign
multispec approve press -p checkout-redesign

# 3. Synthesize FAQ to stress-test the press release
multispec synthesize faq -p checkout-redesign
multispec eval faq -p checkout-redesign
multispec approve faq -p checkout-redesign

# 4. Optional: deepen the business case with market validation
#    (or `synthesize opportunity-spec` for the feature-scale discovery canvas)
multispec synthesize mrd -p checkout-redesign

# 5. Synthesize PRD from Working Backwards artifacts
multispec synthesize prd -p checkout-redesign
multispec eval prd -p checkout-redesign

# 6. Generate 6-Pager for the one-way-door decision meeting
multispec synthesize narrative-6p -p checkout-redesign

# 7. Post-gate: human authors UXD
multispec draft uxd -p checkout-redesign

# 8. Synthesize TRD
multispec synthesize trd -p checkout-redesign

# 9. Check status
multispec status -p checkout-redesign

Reference Materials

For deeper understanding of AWS Working Backwards methodology, see:

See Also