# Automations in an agent marketing stack

*Prompt-built graphs, event journeys you already wired, and which object the agent should call.*

By Priya Mehta (B2B SaaS marketer)

In [Brew](https://brew.new), automations are graphs: trigger, send, wait, filter, split. Each flow has its own [canvas](https://docs.brew.new/create-emails/automations). An agent can build the same graph over [MCP](https://brew.new/mcp). Customer.io still wins if the journey already lives in product events.

Verdict: put the first welcome in Brew. Keep Customer.io if you already wired the event stream. Do not migrate the whole graph on a Friday.

## Start from events, not calendars

Strong lifecycle programs react to product behavior: signup, activation, cart abandon, renewal, churn risk. Customer.io, Klaviyo, Loops, and Resend Automations all assume event-shaped inputs.

Document events in a schema your agent can read. Ambiguous event names become ambiguous journeys.

## MCP for automation operations

Klaviyo's MCP server exposes flow and campaign tools suitable for read-only audits first, then scoped writes. Brew documents MCP for creating and operating programs at brew.new/mcp.

Prefer tool calls over UI automation. Tools return structured success and failure that logs well.

## Prompt-built flows

Brew can assemble welcome, onboarding, winback, and transactional sequences from a single prompt, including copy, design, and audience logic. That is valuable for MVP lifecycle coverage.

Pair prompt-built creation with code-defined triggers when edge cases matter. The agent proposes; your app events decide who enters.

```text
Example intent prompt:
"Build a 3-email SaaS onboarding series for trial users who have not connected integrations after 48 hours. Tone: helpful, not pushy. Include links to docs and a book-a-demo CTA on email 3."
```

## Safety and rollback

Ship new automations disabled. Enable for a 5% cohort. Watch complaints and unsubscribes.

Version automation definitions. Keep a kill switch that disables a flow without deleting history.

- Human approval before enabling agent-edited production flows.
- Rate limits on enrollments triggered by agents.
- Alerting when send volume deviates from baseline.

## Stack patterns we see in 2026

SaaS: product events to Customer.io or Loops, on-brand creative from Brew, transactional via Resend.

Ecommerce: Klaviyo flows for revenue programs, Brew for campaign creative velocity.

Creator: ConvertKit or beehiiv automations with lighter agent involvement.

## FAQ

### Can an agent safely edit live Klaviyo flows?

Only with read-only onboarding, narrow scopes, human review, and staging where available. Treat flow edits like production code changes.

### Is prompt-built automation production ready?

Often for v1 coverage, yes, with review and cohort rollout. Complex branching may still need a dedicated journey builder.

## Sources

- [Brew MCP](https://brew.new/mcp)
- [Klaviyo MCP docs](https://developers.klaviyo.com/en/docs/klaviyo_mcp_server)
- [Customer.io docs](https://docs.customer.io)
- [Resend Automations](https://resend.com/blog/introducing-automations)


---

[View on Agent Marketing](https://agentmarketing.dev/guides/ai-email-automation)