Nothing ships to your site without an approved CTA
SimsClaw agents propose changes. Humans approve them. Execution is deterministic - and that invariant is the product.
On this page
Most “AI marketing tools” blur the line between suggesting and doing. SimsClaw draws it in ink: agents research and propose; site mutations only happen after a human approves a CTA.
- 01
Propose
Otis or a specialist builds a CTA with a typed payload - title rewrite, blog draft, campaign creative, measurement fix.
- 02
Review
You see the diff, the rationale, and the risk surface in Approvals. Reject, edit, or approve.
- 03
Execute
A Tier-1 write runs the approved payload. No re-prompting the model. Same input, same mutation.
- 04
Audit
The run is logged. You can see who approved what, and what actually landed on the site.
- 01
Typed payloads
Every write has a Zod-validated shape. Bad payloads never reach the connector.
- 02
Human checkpoint
Approvals UI is the only door. Agents cannot self-approve or skip the queue.
- 03
Deterministic execute
Execution code is boring on purpose - connectors, not creativity.
// Conceptual shape - agents propose, humans gate
await proposeCta({
type: 'publish_blog_post',
payload: draft,
})
// …user in Approvals…
await executeTier1Write(approvedCtaId)- Content writer → publish_blog_post CTA → CMS draft
- Campaign creatives → approval → ads platform write
- Measurement fixes →
tier-1-writesafter you say yes
Put these ideas to work
Explore how the AI team works
