How many agents can one person actually manage?

Elliot PadfieldOct 3, 2026

I’m using agents for more and more of my work, and it’s very easy to end up with a bunch of things running at once. I’m less convinced I’m good at keeping up with all of them. One needs a decision, another says it’s finished but I need to check what it actually did, and then there’s a chat I haven’t looked at for a while and can’t remember where we left things. They’re still saving me a lot of time. But I sometimes get to the end of the day having spent most of it moving between chats, reviewing things and giving the next instruction. There’s always something else I could set going. For people doing this regularly, how are you managing it? Have you found a way to keep more work running without spending the whole day keeping track of it?

Discussion

Zach Davidson Oct 3, 2026, 6:25 PM UTC

ive found that much of my work is shifting from initiating new tasks to verifying / evaluating work thats in flight. there's a constant ebb and flow, but "babysitting agents" is increasingly what my work looks like. i'll often ask older agents to gimme a quick status update on what they completed / whats in flight / what decision is needed from me, and that helps simplify the triage

Wojciech Kulikowski Oct 3, 2026, 7:28 PM UTC

One thread agent spinning up subagents is the only UX that feels right for me. Dot, instinct and muse got it right. Grokbot introduced the primary agent now too.

Dhruv Gupta Oct 3, 2026, 8:53 PM UTC

I usually work with between 4-7 concurrent terminals working on different parts of the codebase! I think this is all about your own context management ability, and trying your best not to drop any balls + ensuring that your agents don't get in each other's way. I think it's definitely hard, but is so much quicker/less downtime because you always have something to do, whether its planning one of your agent's next missions, or auditing one of their's work

Drew VolpeOct 4, 2026, 3:42 AM UTC

I've found having them write software artifacts allows them to solve something once and then they've automated it and don't need your input to keep running it (or don't need as much). Agents are very good at writing and using code.