Scheduled · 0 */2 * * *
Support ticket triage
Status
ActiveTrigger
Scheduled · 0 */2 * * *
Owner
john
Steps
4
Definition
Steps, top to bottom.
-
Pull unanswered tickets
Open or unresponded HubSpot tickets touched in the last 48h.
Fetch -
Classify
Sort into: regulatory-sensitive (escalate to Cara), product question, account/billing, hardware, other.
Decision -
Draft replies
One draft per ticket. PHI-redacted, citation-aware, never sends — only drafts.
Agent → basal -
Surface for review
Drafts surfaced in HubSpot for John's review. Sends only after explicit approve.
Human approval
Why we run this
Support triage is the canonical “agent drafts, human sends” workflow. It runs every two hours during business hours and never auto-sends — the human-approval step is non-skippable.
When a ticket is regulatory-sensitive (anything mentioning adverse events, prescribing, or specific patient outcomes), the workflow escalates immediately and does not draft a reply. That’s a guardrail at the workflow level — Cara, the support agent, is configured the same way.
The output discipline is the same as the morning briefing’s: short, structured, and oriented toward what the human is going to do next.
Run history
Mock dataLast six runs.
Durations, outcomes, and costs below are mock data — every number on this table is fabricated for the demo. The real run history will come from the workforce audit log + fleet-api Worker once Blocks 3 and 4 land (see /roadmap). The table renders the same; the source of truth changes.
| When | Status | Duration | Triggered by | Version | Cost |
|---|---|---|---|---|---|
| 2026-05-10 07:00 | Success | 14.2s | cron | v3 | $0.084 |
| 2026-05-09 07:00 | Success | 12.8s | cron | v3 | $0.079 |
| 2026-05-08 07:00 | Success | 19.4s | cron | v3 | $0.094 |
| 2026-05-07 07:00 | Failed | 11.7s | cron | v3 | $0.012 |
| 2026-05-06 07:00 | Success | 13.1s | cron | v2 | $0.081 |
| 2026-05-05 07:00 | Success | 16.4s | cron | v2 | $0.090 |
Designer · Phase 4
Edit, version, A/B promote.
The in-UI designer lets you edit step lists, swap agents, adjust prompts and parameters, save a new version (immutable history), and promote a candidate when it wins on cost, latency, or quality. Wired with Phase 4. Today's edits go through a platform-team PR.
See /roadmap for sequencing.