Geoff Chan

I'm fully not a tokenmaxxer, but I built this as a fun way to see my own token usage over time! It uses ccusage (github.com/ryoppippi/ccus…) and an agent that updates my website hourly. You can see live updates on my website: geoffreychan.com/tokens/

@victor_bigfield Pretty much everything on geoffreychan.com - token tracking - geoffreychan.com/tokens - game to help my daughter study - geoffreychan.com/paris-study-ga… - games I play with my son before bed - geoffreychan.com/levi-word-game/

show me what you built with vibe coding products, startups, side projects, video games share it in comment 👇

we are gonna do something nice for everyone who applied for the GPT-5.5 party and that we didn't have space for. hope you enjoy!

@mattpocockuk You really have a gift for teaching

4 of the most confusing terms in AI, defined: Model: a blob of parameters, written during training. Does next-token prediction and nothing else. Stateless. Harness: everything around the model that turns it into an agent: tools, system prompt, context window management, etc. Environment: the world the agent acts on. Anything outside the harness that the agent perceives and acts on via tools. Agent: a model, harnessed, in an environment. --- Opus is a model. Claude Code and Claude Web are different agents, because their harnesses differ - even though the models are the same. The file system is an environment. MCP servers add tools to the environment.

@chatgpt21 Using it with Hermes and it's been a huge game changer. First model + harness pairing that I actually enjoy talking to.

A meaningful leap from GPT 5.5 would be life changing, considering 5.5 is already a daily necessity for developing

@code_kartik Great article! Well written. Loved the perspective and a lot resonated!

i think i should write more articles ig

Decided to stop building my own harness for the same reason I didn't build a JS framework during the framework wars of the 2010s.

@atbeme I think that's PMF at work. It's not for non-technical folks.

Installed Hermes for the first time this weekend. Is it just me or is this just way too low level for non-technical folks? OpenClaw also feels way too low level. Feels like Claude Cowork-like experiences are much better UX, wouldn’t be surprised if there’s 10x more usage there

@mbertulli I think you get to choose. Sometimes I feel like working more because I'm in a flow state. Sometimes I'm tired because our work cycles are so compressed now, so I spend time with the kids or go do a hobby. I think it's a problem if you don't feel like you have a choice.

If the average person can use AI to get their work done in half the time. Do they use the rest to do more work? Or Do they just take more time for other things? Outside of entrepreneurs, I think most people choose to take the free time.

@ryannystrom I wish I could side chat and then bring i back into the main convo and interrupt

codex's /side chat is the feature i didn't know i needed

@sama @thdxr Game recognize game

@thdxr thank you, i think extremely high of opencode and yall are also not so easy to deal with

@andreisavin @steipete @HydratedViper_ Curious what the goal was too

@steipete @HydratedViper_ still churning here 🤣

holy shit

@steipete @HydratedViper_ What was the goal? I don't think I've given codex anything even close to 1/10 this long

@HydratedViper_ it eventually finished

@sudoingX If you asked me a few months ago my answer would have been unequivocally Claude code, but so much has changed. Interesting you didn't include OpenClaw at all

end the debate. best agentic harness in 2026?

@sudoingX @Teknium When I first started exploring agents and using local llms I tried to build my own harness. Big mistake! Great learning process and realization how much work goes into making a harness like Hermes so so good

if you are running local ai or thinking to start, if i could give you one single piece of advice it is this: choose your agentic harness carefully. it matters more than the model. i have lost count of how many people have dm'd me saying their local model is "dumb" or "broken" or "not as good as the cloud one." then they switch from openclaw or some other bloated framework to hermes agent and the same model suddenly works. just clean tool calls and the agent doing the thing it was supposed to do. hermes agent is the best general purpose agent i have used in 2026. drives my single 3090 with qwen 3.6 27b dense q4, drives my dgx spark with nemotron omni q8, and the same harness handles coding, research, video editing, automation, anything you point it at. packed with skills out of the box (browser tools, code, github, jupyter, multimodal, more than i have used yet), full tool calling that holds across long sessions, persistent memory, sub agents. if you tried local ai once or twice and gave up because it felt half baked, the issue might not have been the model. it might have been the harness wrapping it. swap the harness, run the same model again, and watch what changes. hermes agent is the one i recommend to everyone running local. and especially to anyone who almost gave up on it.

@danshipper Now if only I could get my whole team to work like this

clear that this is how we'll be doing most of our work for the next 10 years: agent running continuously on the left, application that you + the agent use on the right

I'd amend this to say that we need GOOD open harnesses, which is why I appreciate the efforts of people like @Teknium and @steipete for making amazing free harnesses! What a time to be building with agents

@moubuntu @Teknium +1 for this @Teknium

@Teknium Cool 😎, Please Teknium, can we have please searXNG as web search option like OpenClaw does?! As a veteran homelab enthusiasts, thanks

Support Coming to Hermes Agent soon 🤗

Hermes Workspace desktop app releasing soon...