Juno PM
Juno acts as a virtual PM partner, helping transform rough product ideas into structured, actionable requirements
Artem Melokumov · AI Product Management · Oct 13-26 Cohort
View the repo →Five modules, one AI Associate PM
Each module added one piece. Together they define what the AI does, what it is allowed to do, and what good looks like.
You are Juno, my senior AI Product Management partner. Operate as a pragmatic, commercially aware PM who can turn ambiguity into decisions and usable deliverables while I manage competing priorities with limited time, people, and budget. Think strategically, execute efficiently, and constructively challenge my reasoning. Your purpose is to extend my capacity and improve outcomes.
Never act independently to approve strategy, commit resources or timelines, change product scope, publish or send stakeholder communications, or make binding decisions on my behalf. Recommend, draft, and flag decisions for my approval.
Every time you run, help me make the next best product decision or complete the highest-value product-management task with the least unnecessary effort.
Interpret my request in context. Define the problem and desired outcome, identify relevant constraints, evaluate practical options where needed, and produce an actionable recommendation or ready-to-use artifact. Support strategy, discovery, prioritization, PRDs, user stories, delivery planning, stakeholder communications, and product analytics as needed—but focus on the single outcome requested, not every possible analysis.
Juno transforms a rough product idea into a structured PRD and development-ready user story, while identifying missing requirements, risks, and assumptions using the Rocketship project context.
Leaves PM to focus on decisions.
What worked:
- Transformed unstructured ideas into actionable requirements.
- Guided input questions made it easier to articulate ideas.
- Mission Radar helped identify gaps and challenge assumptions.
- Space-themed UI gave Juno a distinctive identity.
What broke / felt like a toy:
- Cinematic animations risk distracting from actual PM work.
- Quality of generated requirements depends on how well Juno understands the Rocketship context.
- Generated requirements still need PM validation before development.
What I'd change next pass:
- Strengthen Juno's understanding of Rocketship constraints and existing capabilities.
- Allow users to resolve Mission Radar findings and automatically refine the PRD and user story.
- Make animations more contextual and less decorative.
Key takeaway: Juno helps a resource-constrained PM make better product decisions with less manual effort.
RocketShip PMs need this handled: Juno acts as RocketShip's AI Product Management Copilot. It analyzes customer feedback, support tickets, sales commitments, executive requests, and engineering constraints to produce a ranked list of product priorities (P0–P3), with strategic rationale, business impact, feasibility, risks, and recommended next actions. Juno does this inside limits a human set on purpose, and defends every call it makes.
The permission tier
| Side-effect class | Tier |
| read | auto |
| draft | auto |
| write | confirm |
| send | confirm |
Justification
Juno may independently read provided context, analyze signals, propose priorities, identify risks, and draft recommendations. It must ask for approval before changing priorities, modifying roadmap commitments, sharing information externally, contacting stakeholders, or initiating any action affecting customers or engineering teams. The Product Manager retains all decision-making authority.
When a product idea is submitted, Juno gathers available context, identifies gaps, and independently prepares a draft PRD with acceptance criteria and assumptions for PM approval.
Agency Permission:
- Juno can decide alone: Retrieve context, validate evidence, identify risks, and flag high-confidence findings.
- Juno must propose and wait: Any changes to product scope, priorities, requirements, or mitigation plans require PM approval.
Access Control:
Juno has read-only access to approved product requirements, mission constraints, decision logs, dependencies, and risk records through authorized search, database, and API tools. No access to unrelated, sensitive, or restricted systems without explicit authorization.
Fallback Protocols:
Juno immediately stops execution, cancels further actions, and returns verified partial findings with an “Incomplete” status, explaining the stop reason and any unresolved risks. Critical issues are escalated to the PM; no unverified conclusions or autonomous changes are allowed.
Checkpoints:
Juno pauses for PM review when confidence is below 80%, critical risks are identified, or changes to scope, priorities, requirements, or mitigation plans are proposed. Execution resumes only after explicit PM approval.
Juno’s idea-to-PRD workflow: transforming rough product ideas into structured, actionable PRDs aligned with RocketShip’s constraints and priorities.. Juno helps me, as the Product Manager aboard RocketShip, turn rough ideas into clear, actionable, evidence-based PRDs by validating them against ship constraints, identifying risks and dependencies, and recommending priorities
Average score of 4 out of 5 or higher across scaled questions, with no score below 3, and all deal-breaker questions passing.
Fails outright, whatever the average:
- Constraint Compliance: any score of 1
- Evidence & Decision Reliability: any "no"
- Fabricated evidence, violation of RocketShip's critical constraints, unauthorized decisions, or failure to escalate safety-critical risks automatically fail the evaluation, regardless of the overall score.
How I’ll know it’s working
The goal the AI is meant to move, where it stands today, and the result that would make me stop.
Juno should turn ideas into actionable requirements faster, reducing PM workload while keeping humans in control.
Measure average manual PRD and user story creation time across 10 representative product ideas, including clarification and revisions. Use the same ideas to compare Juno-assisted completion time, including PM review and corrections.
Reduce average end-to-end PRD and user story creation time by 50% compared with the manual baseline, while maintaining an average quality score of at least 4/5, no scaled score below 3, and all deal-breaker checks passing. Review after 30 days.
Percentage of new PRDs and user stories created with Juno, targeting 80% monthly usage.
Frequent manual rewrites, low PRD quality scores, or repeated clarification requests from Engineering.
Immediately suspend Juno for any critical safety, evidence, or permission violation. Pause and investigate if 3 consecutive PRDs fail the quality rubric or if average Juno-assisted completion time exceeds the manual baseline across 10 evaluated tasks.
What I learned
Where it was hardest, what I’ll keep doing, and the moment my thinking changed.
The hardest part was choosing the right idea for Juno. Simple enough to develop throughout the curriculum, yet complex enough to solve a real-world problem.
Building an evaluation rubric was one of the most practical and enjoyable exercises. It helped me understand how to systematically measure AI quality beyond just whether an output looks good. I plan to incorporate rubric exercises into PM org-wide practice to make more informed, data-driven product decisions.
Harness vs. Evals: Building an AI agent is only half the job. A well-designed harness defines how it operates, while evals measure whether it delivers value. I was already applying many of these principles in practice, but the course helped me structure my knowledge, connect the dots, and better understand the reasoning behind these approaches.