Claims to prove
Replace broad promises such as ‘safe’ or ‘accurate’ with an observable behaviour and a pass condition.
← Projects
AI Product Launch ReviewPublic prototype
Turn an AI idea into a launch decision you can test.
AI Product Launch Review translates a product description into testable claims, adversarial scenarios, evidence requirements, and launch gates. It is designed to make the reasoning visible before a polished demo creates false confidence.
AI Product Launch Review · Public prototype
Define what must be true before launch
01
What are you asking the system to do?
Start with the product claim. The review will test the boundaries around that claim rather than judging AI in the abstract.
About two minutes
Execution kit · CA$39
Turn the draft review into an owned launch process.
Use the editable workbook to manage risks, evaluations, evidence, launch gates, and a framework and regulation applicability screen. The guide and templates help a team run the review and record the decision.
Get the kit — CA$39See what is included →
One-time purchase · secure email delivery · no subscription
The method
Move from confidence to evidence.
A useful review separates what the team hopes will happen from what it can demonstrate. The first version follows a transparent set of rules, so every output can be challenged and edited rather than accepted because it sounds authoritative.
Boundary tests
Test ambiguity, excessive authority, adversarial input, tool failure, and recovery alongside the happy path.
Launch gates
Name the evidence, decision, and owner required before a consequential system reaches real users.
A review artifact a team can challenge
- System boundary and prohibited-use statement
- Claims with acceptance criteria and expected evidence
- Baseline, boundary, adversarial, and recovery tests
- Launch gates with decisions and suggested owners
- Markdown export for a repository, ticket, or review document
A first pass is not proof
The review uses only the description entered in the browser. It does not inspect the product, run the generated tests, validate evidence, certify safety, or make a legal determination. Its job is to make the next questions and tests concrete.
The next product milestone is evidence ingestion: attaching test runs, traces, and review records to each launch claim without allowing the model to grade itself.
Worked example
Inspect a complete review before starting your own.
The sample applies the current rules to a shopping agent that can complete a purchase within a user mandate. It shows the generated claims, adversarial cases, evidence requirements, launch gates, and open decisions in one portable artifact.
View the worked review →What the example establishes
- The rule engine produces stable output from the same inputs
- Money, browsing, and independent action trigger specific controls
- Every launch gate names evidence and a suggested owner
This is a worked prototype output, not evidence that the reviewed product is safe or ready to launch.
Newsletter
Follow the product and get The First Build.
I share occasional notes on AI Product Launch Review and the other things I am building. The download you receive is The First Build, my step-by-step guide to creating an AI morning brief.
Free, occasional, and written by me. Unsubscribe whenever you want.