Product feedback routing agent

TL;DR: Nick Hill walks through building an agent that pulls product feedback from Slack, forums, and support, clusters it into recurring issues, then auto-files prioritized Linear tickets for the right teams. ## Overview Nick Hill demonstrates how to build a “product feedback routing agent” that: 1. Ingests feedback from Slack, public forums, and support channels 2. Clusters the input into recurring problems and pain points 3. Sends a daily digest to product leadership 4. Creates or updates Linear tickets with rich context The entire flow is configured with plain-language instructions; ChatGPT turns the description into an executable plan, provisions the necessary connectors, and drafts the agent’s operating instructions. ## Step 1 – Describe the Agent in Natural Language Inside ChatGPT, Nick opens the “Create an agent” flow and types: > “Build an agent that reads product feedback, groups repeated issues, and generates follow-up tasks assigned to the correct teams.” ChatGPT converts this sentence into a structured execution plan and immediately surfaces the required integrations. ## Step 2 – Grant Tool Permissions The agent needs three capabilities: - **Web search** – to crawl public forums - **Slack integration** – to read internal channels and post summaries - **Linear integration** – to open or update tickets Nick reviews each permission and explicitly enables only the scopes the agent will use. He stresses that “permissions are critical—the agent can only access data you explicitly authorize.” ## Step 3 – Confirm Operating Instructions ChatGPT auto-drafts detailed instructions covering: - Data sources to poll - Clustering logic (“group by recurring problem or pain point”) - Output format for the Slack digest (bullet list, customer quotes, frequency count) - Linear ticket template (title, description, customer evidence, suggested owner) Nick skims the draft, makes no changes, and clicks “Create agent.” ## Step 4 – Choose Trigger Mode Two options are shown: 1. **Manual** – run on-demand inside ChatGPT 2. **Scheduled** – run daily at a set time, or trigger via a Slack slash command For the demo Nick keeps it manual. ## Step 5 – Run the Agent On first execution the agent: - Pulls the latest 100 Slack messages from designated feedback channels - Scrapes the top 20 forum threads containing product tags - Queries the support ticketing system for “product-question” and “bug” labels It then embeds the text, clusters semantically similar items, and produces a concise summary. ### Example Slack Digest Output > **Daily Feedback Snapshot – 17 May** > - **Export timeout (23 mentions)** – Enterprise users hitting 60-second gateway timeout when exporting >500 k rows. Suggested fix: paginate export endpoint. > - **Mobile offline sync fails on iOS (15 mentions)** – Reproduced on v4.3.1; logs show “Network unreachable” although Wi-Fi is stable. > - **New spreadsheet UI scroll lag (9 mentions)** – Reported by Chrome 124 users on retina screens; disabling smooth scrolling flag mitigates. The same digest is posted to `#product-leadership` with a thread for discussion. ## Step 6 – Sync Findings to Linear Nick taps “Sync to Linear.” The agent: 1. Searches Linear for existing tickets whose titles or descriptions match the clustered issues 2. If a ticket exists, appends the new customer quotes and increments a “report count” field 3. If no ticket exists, creates one, prefilling: - **Title** – concise problem statement - **Description** – full context, screenshots links, frequency metrics - **Labels** – `customer-reported`, `agent-scoped` - **Owner** – mapped via a simple keyword-to-team table (e.g., “export” → Data Platform) In the demo the agent opens three brand-new tickets: - DATA-1823 – Export gateway timeout - MOB-724 – iOS offline sync failure - FRONT-455 – Spreadsheet UI scroll lag Each ticket contains a “Customer evidence” section with anonymized quotes and a count of unique reporters. ## Closing Notes Nick highlights that the entire setup took under five minutes and required zero code. He recommends switching to the scheduled trigger so the Slack digest and Linear sync run unattended every morning at 08:00 Pacific, giving product teams a ready-made backlog aligned with real user pain. Source: [OpenAI on YouTube](https://www.youtube.com/watch?v=bk2H8WfHZZk)