Playbooks

Classify how a shopper replied to a recovery prompt, draft a personalized win-back email that addresses their reason, and QA it for tone and links before it goes out.

3 agents · 5 steps

Generate marketing ad copy from a brief, run deterministic QA on placeholders, links, and forbidden phrases, and reject variants that fail brand safety checks.

2 agents · 4 steps

Let an agent propose tool or function calls, validate names against a configured whitelist, and route blocked calls to human review before execution.

3 agents · 5 steps

Review urgent payment instructions, validate wire details, block or escalate to CFO.

4 agents · 6 steps

Score a marketing campaign idea for fit and risk. Approved ideas are expanded into a structured creative brief; weak or risky ideas are returned with a reason instead.

2 agents · 4 steps

Generate a patch, run security checks, route unsafe changes to human review.

3 agents · 5 steps

Triage incoming complaints by severity, escalate urgent cases, and draft a standard reply for routine issues.

3 agents · 5 steps

Classify a single review or NPS comment into structured JSON with sentiment and a primary topic from a fixed taxonomy. Invalid enum values are blocked deterministically — not silently mapped to the nearest tag. Run once per comment; batch ingestion loops on your side.

2 agents · 4 steps

Turn messy contact input into normalized CRM fields, then add audit metadata for downstream systems.

3 agents · 4 steps

Extract expense details, enforce policy rules, approve or reject with reason.

3 agents · 5 steps

Pull structured data from a document, confirm with the user when safe, or send a short failure message when extraction fails checks.

3 agents · 5 steps

Extract fields from input, validate required fields, route to a repair step when checks fail, then loop back to extraction (up to 3 visits).

3 agents · 5 steps

Score an incoming feature request for fit. Approved ones are checked for API/contract impact before being marked as planned; everything else gets a clear, polite reply explaining the decision.

3 agents · 5 steps

Generate a customer-facing financial answer, run compliance guardrails on guarantees and disclaimers, and route blocked output to human review instead of sending.

3 agents · 5 steps

Turn raw issue text into structured bug reports with required reproduction steps and logs, auto-reply when fields are missing, and draft triage responses when intake passes.

3 agents · 5 steps

Extract new-hire details from an intake form, enforce a strict schema so nothing required is missing, and send a personalized welcome only when the record is complete - incomplete records route to a human.

3 agents · 5 steps

Answer HR policy questions with citations and GDPR guard before sending.

2 agents · 3 steps

Pull the key error patterns out of a noisy log dump, propose a safe remediation, and automatically hand off to an on-call human when the fix can't be validated as safe.

3 agents · 5 steps

Extract first notice of loss, guard coverage language, escalate to adjuster when needed.

4 agents · 6 steps

Extract fields from an invoice, run a fraud-pattern check, then send an approval or rejection message.

4 agents · 6 steps