Set up the overnight agent

Situation

You want Specky to work while you sleep. Every morning, instead of triaging raw notifications across 6 tools, you want to open one inbox with confidence-scored drafts ready to approve.

What you need

• At least one integration connected (Slack or Jira recommended)
• A workspace with some signals already indexed
• 5 minutes to review the agent's output settings

Steps

1. Connect your integrations

The more signals the agent has, the better the morning output. Connect in priority order:

1. Slack — highest signal density; threads surface the most opportunities
2. Jira — gives the agent context on open tickets and sprint state
3. Gong — adds call transcripts for customer signal
4. GitHub — adds PR activity for execution context

Go to Settings → Integrations and connect via OAuth.

2. Understand what the agent does overnight

Every night, Specky's agent:

1. Reads all new signals since your last review (Slack threads, Gong calls, Jira updates, GitHub PRs)
2. Cross-references them with your OKRs and open opportunities
3. Identifies clusters — groups of signals pointing at the same problem
4. For each cluster above a confidence threshold, drafts a PRD, stages tickets, or queues an Alex campaign
5. Queues all of these in your PM Inbox with confidence scores

3. Review your first inbox

Open PM Inbox the morning after connecting. Each item shows:

• Type — PRD draft, ticket batch, research campaign, or competitive alert
• Confidence score — how strongly the agent believes this is worth your attention
• Evidence — the specific signals that triggered this item (click to expand)
• Action — Approve, Edit, Discard

4. Calibrate with feedback

The agent learns from your approvals and discards. If you discard three consecutive alerts about a particular signal type, it reduces the weight of that signal. If you approve a particular pattern consistently, it raises confidence for similar items.

To give explicit guidance: open AI Chat and say:

For the PM Inbox agent: prioritise signals from #product-feedback Slack channel
and de-prioritise GitHub PR activity unless it's in the checkout service.

5. Set notification preferences

Go to Settings → Notifications and configure:

• Morning digest — email summary of the inbox at 8am your timezone
• High-confidence alerts — immediate push notification when confidence > 0.9
• Weekly synthesis — Friday afternoon recap of all approved items

Output

Every morning:

• 2–5 confidence-scored inbox items
• Drafted PRDs ready for review (not raw data dumps)
• Staged Jira tickets with acceptance criteria
• Queued Alex interview campaigns for open questions

Variations

Lean team? Set the confidence threshold higher (0.85+) so only the strongest signals surface. You'll get fewer items but higher quality.

High-volume? Set up a Custom Workflow to run a nightly synthesis of just your #support Slack channel and deliver a summary — separate from the main inbox.

Want to see the raw work? Open Product Graph and filter by created_at: last 24h to see every node the agent created overnight.