Did Hermes Agent just kill OpenClaw? (full guide)
TL;DR
Hermes feels faster and more transparent than OpenClaw — Alex says Hermes was noticeably quicker on the same Opus 4.6 model and lets you watch every tool call, website visit, snapshot, and reasoning step in the CLI.
The standout feature is self-improvement through reusable skills — in his demo, Hermes turned a complex 9:00 a.m. Hacker News briefing workflow into a named skill, attached it to a cron job, and delivered a custom Telegram audio podcast with ranked stories.
Hermes is built for tinkerers and open models in a way OpenClaw isn’t — Alex highlights built-in ML and reinforcement learning tools plus strong support for running on open models like Qwen 3.5, while saying OpenClaw has publicly discouraged open-model use.
OpenClaw still has real structural advantages — he points to the larger team, Peter Steinberger’s backing, support from OpenAI/Aqua and Nvidia, faster feature shipping, native plugin support in tools like Cursor and Claude Code, and a bigger community.
His answer is not ‘replace OpenClaw’ but ‘run both together’ — Alex uses OpenClaw as the main orchestrator, lets it spin up Hermes via ACP for the tasks it’s better at, and often runs both side by side so one can handle frontend while the other handles backend.
A two-agent setup also acts like an insurance policy — when Hermes or OpenClaw breaks after an update or config issue, he uses the other agent to diagnose and repair it, which he says makes the workflow cleaner, safer, and less stressful.
The Breakdown
Hermes arrives as the hot new OpenClaw rival
Alex opens with the big question everyone is asking: is Hermes Agent actually better than OpenClaw? He frames Hermes as another autonomous AI employee that can control your computer from familiar surfaces like CLI, Telegram, Discord, and WhatsApp — then immediately hints that the hype is real because he’s been using it “hardcore” for weeks.
The flashy demo hides a serious point: Hermes is fast
Before getting technical, he shows off Hermes making ASCII-art video and even generating a polished comparison chart with built-in Excalidraw skills. But the real point is performance: on the same Opus 4.6 model, he says Hermes feels much lighter and faster than OpenClaw, especially after he ran into recent OpenClaw performance issues.
The killer workflow: a self-building Hacker News morning podcast
Alex gives Hermes a meaty prompt: every day at 9:00 a.m., pull the top five Hacker News stories, fetch the source articles, summarize and rank them for AI/startup relevance, then send a Telegram text summary plus an audio briefing of the top three. Watching Hermes work is the fun part — it visibly crawls sites, narrates its actions, then packages what it learned into a reusable “hacker news daily AI briefing” skill and wires it into a cron job before sending him what he calls his own personal morning podcast.
Why tinkerers are flocking to Hermes
He says Hermes has a distinctly hacker-friendly vibe: built-in machine learning tools, reinforcement learning hooks, and explicit support for open models. For Alex, who’s big on local AI and custom hardware, that matters a lot — he notes many people are already running Hermes fully on Qwen 3.5, while OpenClaw has said on record it doesn’t recommend open models.
OpenClaw is still the heavyweight with more backing
Alex is careful not to turn this into a clean knockout. He says OpenClaw still benefits from a much larger operation, with creator Peter Steinberger backed by OpenAI/Aqua resources and Nvidia contributing substantial development support, which translates into constant shipping and better stability in some cases. He also mentions OpenClaw’s bigger community and native plugin support across Cursor and Claude Code as meaningful advantages.
His actual recommendation: don’t switch, stack them
The punchline is that he doesn’t think Hermes kills OpenClaw at all — he thinks the best setup is both. He uses OpenClaw as the orchestrator for complex jobs and tells it to spin up Hermes via ACP when needed, while also running them side by side for parallel work so he’s not sitting there doomscrolling while one agent thinks.
The two-agent workflow gets very practical, very fast
He gives a concrete example from an app he’s building: “Henry,” powered by Opus, works on frontend code while Hermes, powered by ChatGPT, handles backend tasks. That split lets each model lean into its strengths, and if one agent breaks after an update or config issue, the other becomes the emergency mechanic — a nice reliability hack Alex clearly values.
The closing advice: spend 15–20 minutes a day playing with new tools
Alex ends with a broader point than Hermes vs. OpenClaw: if you care about getting ahead with AI, you need daily tinkering time. His current default model pick is still Claude because he calls it the most “agentic,” but he also recommends testing cheaper options like ChatGPT and local setups like Qwen 3.5 — because staying current with tools like Hermes is, in his view, how you create more economic value.