AI employees got a 3D office; some call it The Sims with token-based billing

Robert Scoble has recently been pitched a lot of agent products. Skydive is one of the somewhat different ones: it gives AI agents a 3D office where you can watch them work.

It’s not complicated to use. Build a team of agents, assign tasks through Slack, email, iMessage, or the web, then open the 3D interface and watch each of them busy at their desk. After you claim your company domain, the office can also be placed on a map. Free this week.

Scoble’s assessment is restrained. He says there will be more and more of this kind of thing, but he’d rather keep using GrokBot, for a straightforward reason: GrokBot already knows him well enough, and switching to a new set of agents is like re-onboarding a whole team.

The comments section is more interesting than the product itself.

Someone put it plainly: it’s easy to watch agents work; the hard part is figuring out which assumptions they extracted from “knowing me.” Another comment was shorter: it’s easy to watch agents work; knowing what they changed is hard.

This hits on the current contradiction in agent products. Visualization is lively, but what users are really worried about isn’t that they can’t see; it’s that they can’t understand. The agent says “done,” you watch it click a few times, and then you have to trust it.

Some users disagree. John Knopf says it’s useful for him because he needs to show clients work progress, and when going through the PR process on GitHub, it’s more intuitive than just sending a “done.” This scenario holds up. Showing it to clients and using it yourself are inherently held to different standards.

Others are in it purely for entertainment. One reply said watching agents work is like watching an ant farm; they could leave it on all day. Scoble replied: it is indeed a form of entertainment.

The opposing side is more direct. Someone said the whole point of an AI agent is to run in the background, not to be watched; when it finishes the work, ping me if there’s an issue—I still have kids to look after. That reason is more practical than any product analysis. Others pointed out that 3D virtual offices had a wave of hype during the pandemic—virtual spaces, 3D proximity chat—and in the end nobody really used them.

Someone summed up Skydive as “The Sims, but billed by token.” Pretty accurate. The fun of The Sims is watching the little people do things, occasionally intervening, and mostly spectating. But if tokens are burning, spectating isn’t so relaxing.

Some users watched it and immediately thought of Black Mirror.

The product does hit a real problem: the execution process of agents isn’t transparent. But visibility and controllability are two different things. Being able to see them working doesn’t mean you can judge what they’re doing. Scoble’s line about “would rather use GrokBot, which already knows me” points to something else: the competition among agent products may not be in the interface but in context. The 3D office looks lively, but what saves time is still the one that already knows what you want.