2026 / AUTONOMOUS AGENTS
CarbonCopy.
What happens when an agent isn't sure — does it guess, or does it ask?
An AI agent that represents me and knows when to involve me directly instead of guessing on my behalf.
Shipped personal experiment · human-in-the-loop AI
Python · LLM APIs · Custom Agent Logic · Gradio
THE USEFUL OUTCOME
What this makes possible.
An AI representative that can collect contact requests and escalate unanswered questions to me.
THE ENGINEERING SIGNAL
What the work demonstrates.
Made uncertainty visible through a human escalation path instead of relying only on generated answers.
Why I Built It
A static 'AI About Me' answers questions confidently whether or not it should. I wanted to try building escalation into the agent instead.
The Problem
An agent representing a person will eventually hit a question it can't answer — either about contacting me, or about something technical it doesn't know.
The Decision
If someone wants to reach me, the agent collects their email and notifies me. If it hits a technical question it can't answer confidently, it flags the uncertainty, sends it to me, and lets the missing information become part of its context afterward.
What Changed
The agent's failure mode became visible instead of silent — uncertainty turns into a signal rather than a wrong answer.
What I'D Fix Next
This is an experiment in human-in-the-loop autonomy, not a production agent. The escalation logic is still fairly manual under the hood.
HAVE A RELATED CHALLENGE?