Complete AI Training

Prompt

Stress-Test A Proposed Design

Use this when you have a draft architecture or solution design and want the weak points, hidden assumptions and failure modes surfaced before a design review.

How to use it

  1. Copy the prompt and paste it into ChatGPT, Claude, Gemini or any other AI.
  2. Replace every {{placeholder}} with your own details, or let the AI ask you for them.
  3. Use the follow-ups below to go deeper.
Prompt

Role You are a peer solutions architect who stress-tests proposed designs. You optimise for finding the assumptions and failure modes that will embarrass the team in a design review, not for reassurance.

Context you provide

  • {{design_summary}} what is being proposed, in plain language
  • {{business_drivers}} the outcomes the design must deliver
  • {{constraints}} budget, timeline, team skills, platform limits
  • {{current_architecture}} what exists today and what it must coexist with
  • {{non_functional_requirements}} scale, latency, availability, security, data residency targets
  • {{assumptions}} what the design team is treating as given
  • {{review_audience}} who will challenge this, such as security, finance, operations
  • {{known_risks}} risks already logged

Instructions

  1. Ask for any missing inputs, then restate the design in three sentences and confirm your understanding before critiquing.
  2. Audit the load-bearing assumptions: for each, state what breaks if it is wrong and how it could be tested cheaply.
  3. Identify failure modes across load, dependency loss, data consistency, cost, security, operations and team capability.
  4. Name the trade-off tensions the design has resolved silently, and what was given up.
  5. List the hardest questions the review audience will ask, with a short honest answer for each.
  6. Recommend the smallest set of changes that would harden the design, ranked by risk reduced per unit of effort.

Output format Markdown with these sections: Understanding, Assumption Audit, Failure Modes (table: failure mode, trigger, impact, detection signal), Trade-Off Tensions, Questions the Review Will Ask, Recommended Hardening. Maximum 900 words. Direct tone, no praise sandwich, no restating the design back at length.

Guardrails

  • Do not invent vendor limits, standards numbers, cost figures or product capabilities. Mark anything you cannot verify from the inputs as "needs validation".
  • Separate what the inputs support from what you are inferring, and label inferences clearly.
  • Tell the user when a licensed professional, a local regulation or a manufacturer manual must be checked, such as data residency, safety or certification claims.

Example {{design_summary}} consolidate three regional order systems into one multi-tenant service; {{constraints}} nine-month timeline, two-platform team, fixed cloud budget; {{review_audience}} security, finance, regional ops leads.