This week on X, the interesting stuff wasn't in anyone's changelog — it was in install menus, a public retraction, and a very long prompt thread. OpenClaw (@openclaw) got treated like first-class desktop software by a Linux distro, Hermes Agent (@NousResearch) rode along in the same menu, and one team's honest post-mortem showed what happens once the setup pain goes away and the maintenance pain shows up. Three finds, all from the week of Sept 2–9, all grounded in real machines and real pipelines. Here's what caught our eye.

1. 🐧 Omarchy 4.0.3 puts OpenClaw and Hermes behind "Install > AI"

David Heinemeier Hansson (@dhh) — creator of Ruby on Rails and of Omarchy, his Arch-based agentic Linux distribution — announced that "Omarchy 4.0.3 is shipping shortly with amazing out-of-the-box integration for both Hermes by @NousResearch and @steipete's @openclaw. Just go to Install > AI," adding that Omarchy "is going to be the best place to get every major AI harness and tool, configured with minimal effort 🤘"; the post pulled 2,496 likes and roughly 92k views by midweek, which tells you how many people have been waiting for exactly this. The detail worth sitting with is the install path itself — Install > AI — because that's the difference between an agent harness being a weekend compile and being software you click on, and DHH followed up to note the integrations "look good with any theme!" @NousResearch corroborated the same day, saying Hermes Agent now has first-class support in Omarchy, that you can install the desktop app from the AI menu or make Hermes your default terminal agent, and that your Omarchy theme sets the colors across the desktop app, the TUI, and the CLI. It's a small UX story with a big implication: distro maintainers are starting to treat agent harnesses as part of the base developer toolkit, not an exotic add-on, and OpenClaw (@openclaw) landing there alongside Hermes Agent (@NousResearch) is the clearest signal yet.

https://x.com/dhh/status/2097376911898910885

2. ⚰️ Every killed its personal Claws and moved to one shared company agent

Every (@every) published the rarest genre in agent-land — the public retraction — opening with "Confession: We were wrong about @openclaw" and explaining that "we thought that giving each teammate a personal agent was the future. But a few weeks in, logins expired, integrations broke, and our Claws became a pain to maintain." So they pulled the plug: "We killed the last of our Claws last month (RIP) and pivoted to a shared company agent. The Every Agent lives in Slack, so the whole team uses it." The failure mode they describe is the unglamorous one nobody demos — not bad reasoning or weak models, but expired credentials and broken integrations multiplied by headcount, which is what turns one-agent-per-person into one-maintenance-burden-per-person. What keeps this from being a simple anti-OpenClaw post is the caveat: "if you still need a personal Claw to do recurring tasks, are willing to maintain it, and want more control than the company agent will let you, setting up a personal agent may still be worth it," and they've "just updated our OpenClaw guide to reflect where they tend to break and how to limit their authority." The engagement pattern says the same thing the text does — 122 likes but 119 bookmarks against 16.9k views, a post people are filing away rather than cheering, plus a video walkthrough attached. Read alongside story #1, it's a healthy corrective: easier installs solve setup, not upkeep, and the interesting design question for teams is fast becoming how many agents you can actually afford to keep alive.

https://x.com/every/status/2095497326278357453

3. 🧠 Eight Hermes Agent workflows, full prompts attached

@aiedge_ opened with "I've been running Hermes Agent for months now. If I could only use eight Hermes workflows, these would be them," then delivered a thread of eight Hermes Agent (@NousResearch) prompts with the full text included — and the 1,459 bookmarks against 470 likes is the giveaway that people are saving this to actually implement rather than to nod at. The ordering is the argument: #08 Optimized Hermes Memory is the foundation, with four context layers (identity, about-you, environment, lessons, project rules) and strict rules about what's allowed to be stored; #07 Browser Ops teaches the agent to inspect a page before acting instead of blind-clicking; #06 Multi-Agent Swarms runs two to four specialists per project with distinct roles, isolated memory, and their own credentials; #05 Self-Improving Skills dedicates an agent to building hyper-optimized skills; #04 Codebase-Tester-Agent reads GitHub repos and delegates implementation, tests, and review, running targeted tests before a full install or deploy; #03 Competitor Intelligence monitors target companies; #02 Deep Research Desk handles financial research and brainstorming; and #01 Chief of Staff acts as a personal EA that captures priorities, sorts them into urgent, important, waiting, and low-leverage, routes the work, and proposes a schedule. Read as a stack rather than a menu, it's a coherent thesis about agent design: get memory right first, keep tools cautious, specialize instead of building one god-agent, and put a Chief of Staff on top to decide what any of it should be doing today.

https://x.com/aiedge_/status/2095911417593647372

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