Adding a Human-in-the-Loop Approval Step Before the Agent Acts

Full automation is a fantasy for most workflows. The pragmatic version: agent does 95%, then pings a human for one click of approval before it sends, writes, or spends money.

Some examples

• A reply-drafting agent posts to Slack with an Approve button instead of sending the email itself.
• An invoice-paying agent emails you the proposed payment and only pays after you click confirm.
• A code-merging agent opens the PR and asks for review instead of merging directly.
• A scheduling agent pre-fills the meeting invite and waits for one tap before sending.

Try it!

Take any agent that sends or writes. Add an approval step — Slack message, email, or web button. Test the gate.

autonomy is a slider, not a switch; pause before destructive actions

Some examples

• Asking for confirmation before deleting files
• Showing a diff before committing
• Pausing before sending the email

Try it!

Open your favorite AI tool and try one of the examples above. Pick the one that matches what you are actually working on this week. Spend 10 minutes, no more. Notice what worked and what did not — that's the real lesson.

Human-in-the-loop means the agent stops at risky steps (sending email, paying, deleting) and waits for your OK.

Some examples

• Mark which tool calls require approval.
• Make the approval prompt show what will happen (not just 'continue?').
• Don't approve without reading.

Try it!

Configure any agent to require human approval before it sends an email or makes a payment.

Understanding "Human-in-the-loop: the kill switch you actually need" in practice: AI agents don't just answer questions — they can do things, like looking things up, writing files, or talking to apps. Even autonomous agents should pause and ask before doing irreversible things — and knowing how to apply this gives you a concrete advantage.

• Design clear agent goals before adding tools
• Define permissions and scope before deploying any agent
• Build in human-approval checkpoints for high-stakes actions
• Understand when to use an agent vs. a simple chat prompt

1. 1Design an agent spec: goal, tools, permissions, stop condition
2. 2Run a simple web-search agent in a sandbox environment
3. 3Instrument an existing workflow to identify where an agent could save time