Build an AI Product Manager Panel for Backlog Prioritization

Set up a Webhook trigger in n8n to listen for new issues or feature requests from your primary tool (Jira, Linear, or GitHub). This node acts as the entry point for the entire panel synthesis.

Configure the webhook to trigger on 'Issue Created' or 'Label Added' (e.g., 'needs-triage'). Ensure you capture the title, description, and any existing tags. This data provides the raw context for your AI panel.

Watch out: Webhook payloads can be massive. Use a Set node immediately after to filter only the relevant fields: title, description, and reporter, to keep your LLM context clean.

Use a Code node to sanitize the input. Often, ticket descriptions contain HTML, messy log dumps, or irrelevant internal metadata that can distract the agents or waste tokens.

This JavaScript snippet extracts the core problem statement and removes noise:

const rawDesc = $input.item.json.description || '';
const cleanDesc = rawDesc
  .replace(/<[^>]*>/g, '')
  .substring(0, 2000);
return [{ json: { 
  title: $input.item.json.title, 
  description: cleanDesc 
}}];

Watch out: If the description is too long (over 4,000 characters), truncating it might lose critical context. Add a logic check to flag 'Extra Large' tickets for manual triage.

Add an HTTP Request node calling the Anthropic API. This agent focuses exclusively on user friction, accessibility, and cognitive load. It ignores technical implementation details.

Use claude-3-5-sonnet-20240620 for its high reasoning capabilities. This agent evaluates how much the proposed change helps the user achieve their goals and identifies potential UX pitfalls.

Watch out: Ensure your system prompt explicitly forbids the UX agent from commenting on 'how easy it is to build'—that is the Engineering agent's job.

Add another HTTP Request node. This agent evaluates the request for technical feasibility, architectural debt, and potential scaling issues. It acts as the 'reality check' for the panel.

This agent identifies if a 'simple' request actually requires a major database migration or if it can be solved with existing components. It provides a 'Complexity Score'.

Watch out: Engineering agents tend to be pessimistic. Tune the prompt to distinguish between 'hard' and 'expensive' (time-wise).

Add a third HTTP Request node. This agent looks at the request through the lens of revenue, churn, and competitive advantage. It cares about LTV (Lifetime Value) and strategic alignment.

This agent determines if the feature is a 'must-have' for a big contract or a 'nice-to-have' that won't move the needle on company goals.

Watch out: Without specific company context, the business agent might hallucinate revenue numbers. Always include a short 'Current Company Priorities' block in the system prompt.

This is the most critical node. It takes the output from the UX, Eng, and Biz agents and synthesizes them into a final RICE score. It resolves conflicts between agents (e.g., high UX impact but extreme Eng complexity).

This node uses a complex prompt to weight the different perspectives. It produces the final 'verdict' that will be used for prioritization.

Watch out: Ensure this node has access to the justifications from previous steps, not just the scores, so it can explain why it reached its conclusion.

Final step: Send the synthesis to Slack for human approval and update the original Jira/GitHub ticket with the AI's analysis. This closes the loop and ensures transparency.

Use n8n's Slack node to post a formatted block with the three specialist viewpoints and the final PM verdict. Simultaneously, use the Jira/GitHub node to add a comment to the ticket with the RICE score.

Watch out: Do not automatically move tickets to 'Approved' or 'Backlog'. Use the AI output as a recommendation only. A human must still give the final 'Go'.