Tell me the riskiest thing your AI agent can do. I'll map where the human check goes.
A couple of weeks ago I asked here how you prove a human approved what your AI agent did, and the replies turned into the most useful thread I've had on DEV. Two patterns stood out: most approvals happen in a chat or an inbox, and almost nobody could list the actions that ran without one. So here's an offer, in public: Reply with one action your agent takes (a refund, an email, a deploy, a…
One of the riskiest actions your AI agent can perform is a database migration. In many cases, the approval process for such a task simply doesn't exist. The AI agent simply initiates the migration without any human oversight or verification.
When asked to identify where the human check should be placed, the AI agent would recommend that a human review the proposed changes before approving or rejecting them. This could be in the form of a chat message or email asking for confirmation. The screen should display a clear summary of the changes being made, including any potential risks or impacts. This allows the human to review and understand the modifications before giving their approval.
The AI agent would also log the approval decision, along with the specific details of the migration and who approved it. This provides an auditable trail that demonstrates who was involved in the decision-making process. It also allows for future reference and verification that the approval actually took place.
The AI agent would test the effectiveness of the human check by simulating a migration and observing whether or not a human intervened. If the check works as intended, the human should be prompted to review and approve the changes before they are executed. If the check fails, the AI agent should flag the issue and alert the human responsible for oversight.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.