Product managers
When you need to understand the users and find a solution that does not violate the product's constraints.
Scenario · Golden Trio
Research the audience. Invent the solution. Review the result.
Three AI agents work sequentially: CustDev researches the audience, Super Oracle finds the solution, Muse reviews the result. The human controls every stage.
The agents support local and cloud models; the selected configuration and data flows are agreed before launch.
CustDevNeeds map
Super OracleSolution options
MuseCritical audit
A demonstration using synthetic materials is available without processing customer data. Full deployment follows an approved estimate.
00Audience
Who the Golden Trio is for.
Four roles for whom the scenario saves time and reduces the risk of subjective decisions.
Startup founders
When the contradiction between "the user wants" and "we can build" blocks progress.
Researchers
When you need to collect observations, surface contradictions and check the conclusions for honesty.
Teams on a limited budget
When you need to test hypotheses before launching expensive development.
01Orchestration
Three agents work sequentially, each with its own task.
Each agent performs its own task and returns its own result. CustDev does not invent solutions. Super Oracle does not check clarity. Muse does not collect observations.
Original taskResearchInventionAudit
Context handoff
Each agent receives only what its task needs.
The next agent receives the approved result: conclusions, evidence, limitations and the human's remarks. The original materials do not become available automatically.
- The user chooses what to pass on
- The version and the author of the decision are recorded
- Ambiguities become questions, not hidden guesses
- Any stage can be repeated with new conditions
02Checkpoints
Four decisions stay with the human.
Automation does not skip over an ambiguous or sensitive choice.
After the stage: Research
1
Accept the research
Is the sample sufficient and are the conclusions correctly tied to the materials?
The human checks the conclusions against the original materials: whether the conclusions are too strong for the limited data, whether every link to a source is justified.
After the stage: Invention
2
Choose the contradiction
Is this the problem the team actually intends to solve?
The team confirms that the chosen contradiction is the root cause, not a symptom. Without this step the solutions will treat the consequence.
After the stage: Invention
3
Select the options
Which solutions are acceptable in terms of risk, deadlines and values?
Not all directions are equally safe. The human filters out the options that violate the constraints of the product or the team.
After the stage: Audit
4
Accept the audit findings
Which findings block the release and which require a test?
Muse sorts the findings by severity: blocking, requiring verification and cosmetic. The team decides what to fix before the release.
03Output
A single decision dossier.
The team sees not only the final recommendation, but also the path from the original observation.
- The original task and criteria — the problem statement and the completion criteria
- Signal map — a table of recurring needs with quotes from the interviews
- Contradiction model — the conflict of requirements stated without a compromise
- Options and tests — solution directions with verifiable experiments
- Critical audit — findings on clarity, honesty and dark patterns
- Open questions — what remains unverified and requires a human decision
Boundaries of the scenario
The scenario does not make product or investment decisions, does not promise market success, does not replace legal or industry expertise and does not make the conclusions more reliable than the source data.
04Example
How it works on a real task.
A typical scenario: from uploading interviews to a verified document with recommendations.
1
Upload
You upload 15 interviews with users of the product.
2
Research
CustDev identifies 8 recurring needs and records the open questions.
3
Contradiction
Super Oracle formulates the main contradiction "Simplicity vs security" and suggests 3 solution options.
4
Review
Muse checks the conclusions and finds 3 hidden assumptions in the wording.
05Questions
Short answers to frequent questions.
Direct answers for quick reading.
What is needed to launch the scenario?
Materials for the research: interviews, notes, documents. A description of the task and the completion criteria. Access to data within the approved scope.
How long does one run take?
Timing depends on the material volume, selected configuration, and human review. An estimate is agreed before launch, and every stage is recorded in the log.
Can I use only one agent?
Yes, each agent can work independently. The full sequence is useful when the task requires all three stages: research → contradiction resolution → independent audit.
What if the result does not fit?
The human returns the artifact for rework with new conditions. Any stage can be repeated. The decision to approve or reject always stays with the human.
Are technical skills required?
No. You upload the materials and describe the task. The agents perform the analysis, you verify the result.
First step
Choose a ready-made solution or describe your process.
Pick one of the twelve agents and subagents, or fill in the questionnaire for custom development.