Fleet 1.17.0 is out.See what's new →
FleetFleet
Use case

AI Workflows for Marketing Content Governance

Marketing teams adopted AI drafting faster than anyone — and created a new bottleneck: review. Drafts are now instant; brand review, legal review, and factual checking still run on Slack threads and goodwill. The result is either a review queue that erases the drafting speedup, or — worse — content that skips review and ships a claim legal would have killed.

The risk isn't hypothetical: an unsupported product claim in a launch post is a retraction; a compliance miss in a regulated industry is a fine.

How it works with an agent fleet

A Fleet workflow makes review the pipeline instead of the favor: drafts run through a brand-and-claims review against your actual product facts, then a named approver gates publication.

genflows:
  - name: launch-post
    steps:
      - {name: draft, prompt: "Draft the announcement per the brief in the trigger ticket. Every product claim must be supported by the corpus.", corpus: ["marketing/brand-voice.md", "docs/product/**/*.md", "marketing/claims-allowed.md"], kind: report, out: post.md}
      - {name: brand-claims, prompt: "Flag off-voice passages and any claim not supported by the product docs or the approved-claims list.", depends_on: [draft], kind: review, out: flags.md}
      - {name: marketing-lead-ok, depends_on: [draft, brand-claims], kind: approval, out: decision.md}
      - {name: publish, depends_on: [marketing-lead-ok, draft], kind: publish, out: published.md}

The approved-claims file is the quiet hero: a maintained list of what you're allowed to say about the product, applied by the review step on every draft. Claims drift gets caught at review, not at retraction.

The fleet pattern

Brief → draft against brand + claims corpus → review flags unsupported claims → lead approval → publish. AI speed on drafting, institutional rigor on what ships.

Guardrails that matter here

  • Publication is gated on a recorded approval — 'who approved this claim' always has an answer
  • The claims review traces statements to the product docs and allowed-claims list, the structural fix for AI-drafted overclaiming
  • Bounded revision loops: flagged drafts route back a fixed number of times, then escalate rather than ping-pong

Who this is for

Marketing teams shipping product content at AI speed who need brand and legal review to keep up — especially in regulated industries where a wrong claim is more than embarrassing.

Frequently asked questions

Does this slow down publishing?

It speeds up the part that was actually slow. Drafting was already fast; review was the queue. The review step does the claim-tracing legwork and presents flags, so the human approval is minutes of judgment on a pre-checked draft.

Can legal be a separate gate from marketing?

Yes — chain two approval steps with different recipients. Each records its own decision; the post publishes only past both.

Run your first agent fleet

One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.