Contra - A professional network for the jobs and skills of the futureDesigning Human-in-the-Loop AI Workflows That Fail Safely
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Yonatan's avatar
Yonatan Tesfaye
pro
• Nov 3
Your AI workflow doesn't need to be smarter. It needs to know when to stop and ask a human.
Most automations break in one of two ways: they auto-send something wrong, or they fail silently and lose data. Both are design problems, not model problems.
5 rules I use for human-in-the-loop workflows:
→ Route by confidence: auto-run above a threshold, everything else goes to a person
→ Use a real approval queue: show the draft, the input and the reason, then approve, edit or reject
→ Make retries idempotent: key every action on a unique ID so nothing sends twice
→ Alert on failure, loudly: send the error, the record and a retry link to a named owner
→ Keep an audit log: input, output, model, approver, timestamp. Every rejection makes the prompt better.
Where do you draw the line between auto-send and needs-approval in your workflows?
AI AutomationN8NClaude
Post image
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started