A tiny physical avatar for coding agents accidentally solved the 'can't look away' UX problem

jamropl · reddit · 2026-08-29

The author built a small physical avatar next to their monitor as a joke: reading → looks around, thinking → leans back, coding → types, done → rings a bell. It accidentally solved a real problem — you can stop watching the agent.

For a 10-second task you watch; for a 10-minute task you want to do something else while still knowing if it's progressing or needs you.
The classic reading → thinking → coding → done model carries too little info when attention is elsewhere. What you actually care about: Is it progressing? Stuck or retrying? Does it need me? Did something fail? Can I safely keep ignoring it?
They're considering moving to exploring → executing → validating → needs attention → done.

The robot has movement, a display, sound and speech, enabling feedback from subtle peripheral cues to explicit interruption. The author wants the next version to be a genuinely useful ambient interface and is asking the community what signals let them look away and what events justify interruption.

Original post →

A tiny physical avatar for coding agents accidentally solved the 'can't look away' UX problem

More from coding & agent