Hermes Agent w/ ChatGPT 5.5 is literally magic
TL;DR
Hermes is pitched as a lighter, more reliable alternative to OpenClaw — Alex Finn says Hermes has stayed fast for months and “never broken” on updates, while OpenClaw gets heavier over time and has repeatedly failed after updates for him.
His default recommendation is ChatGPT 5.5 over Claude Opus for agent work — not because it’s dramatically smarter, but because OpenAI’s $20/month plan has high enough limits for Hermes, whereas Claude Opus often pushes users toward a $200/month tier.
Setup is intentionally simple: terminal install, pick a model, connect Telegram — Finn walks through copying one install command, choosing GPT 5.5 as the agent brain, and using Telegram’s /newbot flow so the agent can be controlled from iPhone, iPad, or desktop.
The killer use case is 'build while you’re away' — he demos sending Hermes a Telegram prompt to create a 3D rocket simulator in Three.js, then uses /steer to change the rocket to blue mid-task and later asks Hermes to send a screenshot of the finished app.
Hermes’ self-improving skills are the main differentiator — after asking it to make a 60-second YouTube Short with Remotion, Hermes researches the tool on its own, finishes after about 7.5 minutes without asking for help, and saves a reusable “Remotion YouTube short” skill automatically.
Finn’s advice is not to replace OpenClaw but to go multi-agent — he argues Hermes and OpenClaw have different roadmaps, can coordinate through ACP, and work best as a team rather than in a tribal “pick one” setup.
The Breakdown
Hermes as the “AI employee” that actually sticks
Alex Finn opens hard: Hermes might be the most powerful AI agent available right now, basically a 24/7 AI employee living on your computer. He frames the whole video around a practical question builders care about — what it is, how it differs from OpenClaw, how to set it up, and whether it’s worth switching.
Why he’s frustrated with OpenClaw — and why Hermes wins on feel
The big contrast is performance and reliability. Finn says OpenClaw slows down as memory and context pile up, and he’s so burned by broken updates that he’s stopped updating it entirely; Hermes, by contrast, feels “super lightweight,” has stayed fast for months, and has never broken on him. He also shouts out Hermes’ cheap $20/month portal option, easy multi-agent setup, and unusually active social team.
Fast setup, plus his very specific GPT 5.5 recommendation
The install pitch is simple: copy the command from the Hermes site, paste it into terminal, and you’re basically in. For the model, he recommends ChatGPT 5.5 over Claude Opus not because it’s clearly better for agent tasks, but because it’s roughly on par and way cheaper to run via OpenAI’s $20/month subscription instead of Claude’s $200/month tier. He also recommends Telegram as the control layer, since it works across phones, tablets, and computers and fits agent workflows well.
Prototype building from the gym, Whole Foods, or a Tesla
His favorite use case is sending app ideas to Hermes while he’s away from his desk. He demos a Telegram prompt asking Hermes to build a detailed 3D rocket launch simulator in Three.js with altitude, speed, fuel, and mission time, then uses the /steer command mid-flight to say “make the rocket blue.” The result is exactly the kind of thing he’s selling: leave the gym, come back to a working prototype.
Screenshots, remote progress checks, and cron jobs that keep working
From there he shows a very practical remote workflow: ask Hermes to send a screenshot of the rocket simulator running, so you can verify progress from anywhere. He says he loves doing this while his Tesla is in full self-driving mode, having Hermes build and report back while he’s on the move. Then he shifts to scheduled automation, showing how Hermes can run cron jobs like a daily 9:00 a.m. AI stock research report or even monitor an OpenClaw instance every two hours to make sure it hasn’t crashed.
The Remotion demo is the real “magic” moment
The strongest proof point is a task Hermes doesn’t already know: build a 60-second YouTube Short about why people should use Hermes, using Remotion. Finn emphasizes that there’s no built-in Remotion skill, so Hermes has to research it, figure it out, recover from getting stuck, and finish end-to-end in about 7.5 minutes without asking him a single question. Then it goes one step further and saves a reusable Remotion skill on its own, plus creates a thumbnail he never asked for.
How to find your own use cases — and why he says don’t choose sides
Finn ends with two discovery methods he repeats often: give Hermes a full background on your work and ask for three time-saving automations, or manually write down every repetitive task you do in a day and ask Hermes which ones it can automate. On the OpenClaw-vs-Hermes question, his answer is basically “wrong framing” — the future is multi-agent, both tools have different strengths and roadmaps, and ACP lets them talk to each other anyway.