Skip to main content
Intempt
Coming soon

Human approval for AI actions

Holds riskier actions an AI workflow wants to take, such as a mass update or a send to a big account, in an approval gate and queue for human review before execution.

Authored by Sid Chaudhary

Founder & CEO at Intempt

Agentic workflow

Coming soon

9 steps Blu runs end to end. It drafts a plan and nothing runs until you approve it.

  1. 1

    Gate the risky AI actions

    Coming soon

    Blu builds a step

  2. 2

    Let other workflows queue one

    Coming soon

    Blu builds a step

  3. 3

    Send it to the right approver

    Coming soon

    Blu builds a step

  4. 4

    Ask in Slack

    Coming soon

    Blu builds a step

  5. 5

    Wait for an answer

    Coming soon

    Blu builds a step

  6. 6

    Act on the decision

    Coming soon

    Blu builds a step

  7. 7

    Tell the workflow to proceed

    Coming soon

    Blu builds a step

  8. 8

    Log what humans reject

    Coming soon

    Blu builds a step

  9. 9

    Publish and watch the queue

    Coming soon

    Blu builds a step

About this recipe

When to use it

AI-suggested actions are only safe to automate up to a point. This pauses the high-stakes ones - an email to a high-ARR account, a mass update, a large segment send - for a one-tap Slack approval before anything fires.

What it delivers

Routes high-value AI-suggested actions (auto-email to high-ARR accounts, mass account updates, large segment sends, AI-classified routing decisions) to a Slack-based approval step before execution.

How you prompt this

Describe the outcome in your own words and Blu Agent maps it to this recipe, or run it by name. It drafts a plan first and nothing runs until you approve it.

Tell Blu Agent what you want

“Holds riskier actions an AI workflow wants to take, such as a mass update or a send to a big account, in an approval gate and queue for human review before execution.”

recipe.md preview

Built from the published catalog. Blu Agent's full step prompts load in your workspace when you run it.

Sign up to read the full recipe.md and run it with Blu Agent.

Sign up to see the full recipe

Inputs & outputs

What the recipe needs before it runs, and what it leaves in your workspace.

Inputs

  1. Nothing declared beyond an Intempt workspace. Blu asks for anything else it needs as it goes.

Outputs

    How the recipe runs

    Blu follows the steps in order and shows you each one before anything goes live.

    1. 01Gate the risky AI actionsCreate a workflow 'AI action approval queue' that intercepts high-stakes AI-generated actions and requires human approval before execution. Goal: capture AI's speed while keeping human judgment on the consequential decisions. Configurable threshold for what triggers approval (ARR cutoff, segment size, action type, AI confidence).
    2. 02Let other workflows queue oneConfigure webhook trigger that other workflows call when they want to queue an action for approval. Payload schema: action_type (send_email / update_records / route_lead / make_decision), action_summary (one-line description), full_payload (the actual action data), requesting_workflow (originator), threshold_reason (why this needs approval: high-ARR / large-segment / low-confidence). Use the result of "Gate the risky AI actions".
    3. 03Send it to the right approverMulti-split by action_type: send_email to marketing lead approval; update_records to RevOps approval; route_lead to sales manager approval; make_decision to varies by decision type. Each routes to the right approver Slack channel. The approver matters: wrong approver leads to bottlenecks. Use the result of "Gate the risky AI actions", "Let other workflows queue one".
    4. 04Ask in SlackConfigure Slack step that posts an interactive approval message to the right channel/user. Message includes: action summary, originator workflow, threshold reason, action preview (e.g. for email: subject + first 200 chars), and Approve/Reject buttons (or Slack thread-reply 'approve'/'reject'). Tag the specific approver. Use the result of "Gate the risky AI actions", "Send it to the right approver".
    5. 05Wait for an answerConfigure wait-until step that pauses workflow execution until approval response received. Timeout: 4 hours (configurable: high-ARR sends might warrant 1hr, routine updates might allow 24hrs). On timeout: default-reject (safer than default-approve) + post 'auto-rejected after timeout' notification. Use the result of "Gate the risky AI actions", "Ask in Slack".
    6. 06Act on the decisionBranch step on the approval decision: APPROVED to execute action (webhook back to originator workflow with 'go ahead'); REJECTED to archive with rejection reason + notify originator workflow to abort + post reasoning to #ai-governance for learning; TIMEOUT to same as rejected, with timeout reason logged. Use the result of "Gate the risky AI actions", "Wait for an answer".
    7. 07Tell the workflow to proceedConfigure webhook call back to the originator workflow to proceed with the queued action. Include approval metadata (approver name, timestamp, any approver comments). Log to audit trail, every AI action that went through approval is fully traceable for compliance + later review. Use the result of "Gate the risky AI actions", "Act on the decision".
    8. 08Log what humans rejectOn rejected branch: write the rejection to a structured learning log: which AI suggestions are humans rejecting, what's the rejection reason. Over time this reveals which AI behaviors need prompt tuning or guardrails. The reject log is the feedback loop that improves the underlying AI workflows. Use the result of "Gate the risky AI actions", "Act on the decision".
    9. 09Publish and watch the queueValidate and publish. Monitor: approval volume per week, approval-rate (target: 80%+; below 60% suggests upstream AI workflows need tuning), median time-to-decision (target: under 30min during business hours), timeout rate (high = wrong approvers or low-engagement channel). The dashboard for responsible AI governance. Use the result of "Gate the risky AI actions", "Tell the workflow to proceed", "Log what humans reject".

    Install Intempt and run it

    1. 1

      Sign up

      Create a free Intempt workspace.

    2. 2

      Connect your data

      Add the sources this recipe reads from.

    3. 3

      Ask Blu

      Type /ai-action-approval-queue in Blu once this recipe is live there.

    4. 4

      Review and approve

      Blu builds each step and waits for your OK before anything goes live.

    Related recipes

    Sid ChaudharyDiscovery call qualification fieldsGenerates an AI summary for Discovery calls from meeting data using the qualification framework in the summary recipe.View recipe Sid ChaudharyPost demo nurtureKeeps a demo warm for 90 days with a playbook, a case study, an ROI calculator and a customer story, and pulls the AE in early if they bite.View recipe Sid ChaudharyReturn rate by categoryShows which product categories get returned most, and which ones have got worse since last quarter.View recipe

    More recipes by Sid Chaudhary

    Discovery call qualification fieldsGenerates an AI summary for Discovery calls from meeting data using the qualification framework in the summary recipe.View recipe Post demo nurtureKeeps a demo warm for 90 days with a playbook, a case study, an ROI calculator and a customer story, and pulls the AE in early if they bite.View recipe Return rate by categoryShows which product categories get returned most, and which ones have got worse since last quarter.View recipe

    Got a play worth sharing?

    Publish it to the Intempt Collective Marketplace and earn when other teams run it.

    Submit a recipe