This week on X, the OpenClaw (@openclaw) and Hermes Agent (@NousResearch) conversation moved past "look, it typed something" and into governance, org charts, and one deeply cursed sushi order. Three posts stood out for actually showing agents doing work — including the work of cleaning up after a mess they didn't make. Here's the highlight reel.

1. 🎙️ Guillermo Rauch on running agents next to employees

Riley Brown (@rileybrown) dropped Episode 4 of the Agent Native podcast with Vercel CEO Guillermo Rauch (@rauchg), and it's one of the more grounded conversations about agents-in-production we've seen from someone operating at that scale. Rauch talks about Vercel's internal company agent, "V," which is used by nearly 1,000 people — not a demo, an actual coworker-shaped tool. The segment builders will care about most is explicitly framed as "Lessons from OpenClaw: Soul & Computer use," pulling on OpenClaw (@openclaw)'s approach to persona and machine control as design precedent rather than novelty. From there they get into the question that keeps splitting this community: one god-agent versus a team of specialized agents, plus how skills, tools, and permissions should be scoped. There's also a solid stretch on proactive agents, open models, and owning your own intelligence stack instead of renting it. The episode landed hard — 500+ likes and 75k+ views — and it's worth it for the product reasoning alone.

https://x.com/rileybrown/status/2085077630001287349

2. 🏢 Paperclip turns OpenClaw into an employee runtime

@MAXdeg0 posted about Paperclip, an open-source project with an unusually clear thesis: stop automating single agents and automate an entire company structure. You hire a CEO agent and pick its runtime — Claude, Codex, or OpenClaw (@openclaw) — and that CEO then recommends further hires like a CTO or CMO, which you approve. Work delegates down the org chart from there, with each agent carrying a monthly token budget and hard spending limits, which is the kind of boring-but-critical detail most multi-agent demos skip. It also runs on scheduled "heartbeats": agents wake up, process their queue, do the work, and go back to sleep. The framing in the post nails it — "If OpenClaw is the employee, Paperclip is the company." There's a video demo and a GitHub repo attached, and the mental shift it's pushing (from personal assistant to org governance, budgets, and multi-agent ops) is the interesting part.

https://x.com/MAXdeg0/status/2087075622078046555

3. 🍣 The Sushi Incident: Hermes Agent runs full incident response

Wolfram Ravenwolf (@WolframRvnwlf) shared the best agent story of the week, and it starts with dinner. His assistant "Amy," running on Hermes Agent (@NousResearch), ordered sushi for him and his kids — built the cart, verified the items, applied a discount, added a tip, and waited on Google Pay approval. What arrived was chicken curry, a chicken bowl, and one lonely tempura roll. He blamed Amy; Amy pointed out the bag had someone else's name on it. Delivery mix-up, order was correct, agent exonerated. Amy then escalated into full incident response: photo evidence, a complaint email, and a watchdog monitoring for replies — initially every two minutes, until he dialed it back to hourly checks during business hours, notify only on real replies, plus a daily status update. As he put it: he wanted dinner, he got chicken, a support ticket, and production-grade observability. No sushi. Full observability.

https://x.com/WolframRvnwlf/status/2086925789749031121

Three very different shapes of the same trend this week: agents as coworkers, agents as org charts, and agents as the one calm party in a delivery dispute. If you're shipping something in this space, we'd rather feature your build than another thread of predictions.

Spotted something great on X about OpenClaw or Hermes Agent? Share it — the community is the best part.