agent·sit·ting
noun, verb/ˈeɪ.dʒənt.sɪt.ɪŋ/
verb forms: to agentsit · agentsat · agentsitter (n., one who sits)
The practice of supervising AI agents while they work — watching the stream, approving the steps, catching the mistake before it compounds. Like babysitting, except the toddler types 400 words a minute and has your production credentials.
The ambient, low-grade attention state of having an agent running somewhere: not quite working, not quite free, unable to fully look away.
"Can't do lunch — I'm agentsitting a refactor."
"I agentsat three deploys yesterday. I wrote nothing. I have never been so tired."
"We didn't hire a junior this year. We hired six agents and made everyone a sitter."
↑ live demonstration. you are now agentsitting.
Why it needed a name
Somewhere in the last two years, a new kind of work appeared without anyone announcing it. Millions of people now spend part of their day watching software work: reading an agent's running commentary, granting permissions, spotting the moment it confidently heads somewhere wrong. It's not coding. It's not managing. Job descriptions don't mention it, calendars can't represent it, and time-tracking software has no category for it.
Work that has no name can't be discussed, priced, or designed for — you can't say "this tool reduces agentsitting by half" until agentsitting is a word. So: a word.
The name is affectionate on purpose. Babysitting is what you do for something you're responsible for but can't fully control — something genuinely useful, occasionally brilliant, and liable to draw on the walls the moment you check your phone.