Eric Siu ran an AI agent for 48 hours straight to finish building an app, burning through their entire token budget. The key question: is the outcome worth what you spent? Neil breaks down when long-running agents like Codex and Claude Code actually pay off versus when they drain resources for nothing.
You're talking about building all this stuff. Who else is using it right now? Number one. Number 2, what value are they getting from it? And I find that question is so important because it's so easy to build right now. Like we've had someone on the team spend 4 or 5 weeks building and there hasn't been much movement until like we, you know, someone else stepped in. So it's really important to ask those 2 questions. So number 3 from my side. So number 3, this is kind of a multi-set of tools, but it's more so a command than a tool here. If you use /goal, in Codex or Claude Code or Hermes or OpenCLAW. You know what /goal does, Neil? Imagine if I worked for you and I didn't get tired. You say, hey, /goal Eric, build this house, okay? And I kept working until I built the house for you. How would that feel? If you kept working— And it was a nice house. If it was what I wanted, yeah, I'd be happy. Yeah, or you say, hey Eric, build me an app that makes me $1 million. Obviously it's not that easy, right? But if you were very specific with your asks, right? And it just went through. It didn't have to go back and forth with you. Not like you're having to go back and forth on like the bathtub and all that stuff right now. My hesitation is what is the cost for it to achieve that outcome? Because if it's double or triple that I wanted to end up paying, then no, I'm not happy. It's not as bad as you think. So my point of saying this is that /goal is with all these things now. And you basically, like I've seen someone work, like they had the Codex work for like 48 hours and I'm like, okay, so what did you yield from it the other day? Right. And he is like, I got my app completed for me. I was like, was it worth you spending all your tokens? He exhausted his token budget for the week and he's like, absolutely it was. And so I've been running it over and over and I just didn't want to wait on a couple things and I got 'em all done, I think a couple days ago. So it's, I think using that command judiciously will be able to get you a lot of yield.
