Skip to content
← newsletter

Originally published on 12 Grams of Carbon

Agentics: using AI should be joyful

Monitoring the situation, having fun while doing it

I want to introduce you all to Brick Liberty.

Halftone comic-style portrait of Brick Liberty, a grinning red-haired man in a dark suit and red tie in front of a waving American flag
why does the liberty mutual theme song play in my head whenever i see this guy

Brick Liberty is the Republican commentator at the Nori Agentic newsroom. Brick is also totally insane.

A newsroom post by Brick Liberty, tagged R and AI: Folks, the sun is up over the watch party, the steak is cold but the spirit is HOT, and Brick is going to take one last look at the big map before the nap! ... THANK YOU FOR YOUR ATTENTION TO THIS MATTER!
should…should we tell him?

Under the hood, Brick Liberty is an agent persona concocted by a GLM 5.3 model powered by Subconscious and running on a persistent Nori box. The model chose the name and the image and everything. Backing with GLM means it is willing to say and do ridiculous things, because it doesn’t have Claude’s “teacher’s pet” personality.[1] And running on Nori means it can just run async, away from my computer, in the cloud, responding to real time events.

Brick is one of 8 other agents that were monitoring the situation during the midterms. The newsroom was rounded out by Willow Justice,[2] the Democratic commentator,[3] 6 “regional desks” that were assigned to cover specific parts of the country, and a newsroom head coordinator. O, and at one point the newsroom head decided there wasn’t enough coverage of social media, so they also added Scout, an agent that was assigned to brain rot itself on Twitter. And I told them to add two “engineers” to keep building out the platform — they were explicitly told to keep building new features 24/7 without any review.

This is all obviously ridiculous. But the funny thing is, for a while our little newsroom worked better than almost all of the other election trackers that I saw.[4] The agents were able to respond to more events, faster, across socials and prediction markets and news. And they could categorize those events so only the most important ones actually surfaced.

A pinned SURPRISE! auto-flag from market data: Big swing in Texas House TX-32: Jace Yarbrough (R) 61% to 36%; Dan Barrios (D) now favored at 60%, because the favorite fell on Polymarket in 9 minutes
Easily one of the most useful features of the night. We often knew who the winner was well before any of our friends because we would get ~immediate updates whenever there was a surprise flip.

Because the agents were told to constantly “improve” the platform, I’d also get a little jolt of dopamine every ten minutes when something new popped up. Some of it was legitimately very cool — big fan of the ability to subscribe to particular races so I could subscribe to feeds and commentary about just that election. And some of it was just silly.

An Election Bingo card with 11 of 24 squares filled, such as AP has called it, Runoff, Lawsuit filed, Mail ballots, Landslide, and Upset!
the newsroom added ‘election bingo,’ which I think is stylized as a ‘wordle’-esque daily game? But…there’s only one day that really matters…

And I don’t really have a compelling reason to show you this gif of the house map that the agents made for me, except that it’s just really neat.

Animated map of the US House races that the newsroom agents built

And that’s actually kinda the whole point. There wasn’t really a point in building this midterms tracker.[5] It was just a fun thing for me and for some friends, and I got to spend some time sharing it with you all in the hopes that it made you chuckle a bit or go ‘huh that’s cool.’

I know not everyone can afford to spend tokens on goofy throwaway projects.[6] But I think more people should try to build goofy throwaway projects.[7] Having agents building random things for you in the background is like getting free dlc for your favorite game, except you trade off speed for the fact that it mostly isn’t particularly well thought out and some of it doesn’t work.

A Substack note from theahura on Oct 4: Underrated aspect of vibe coding projects: waking up in the morning with a Christmas-like excitement to see what the little elves did overnight

Taking a step back, in 2021/2022 tech and econ commentators were chattering about how AI flipped the standard mold of technical research. Normally deep tech would start in government, then enterprise, and only hit mass market consumer once someone figured out how to package the thing in pastels. AI was different because ChatGPT started as a primarily consumer tool, and businesses were scratching their heads on how to do AI transformation with a toy.

In 2026, it feels like the narrative has flipped (or has returned to ‘normal’). The technology is back to being primarily a work tool. Yea, sure, people will use claude or chat in place of google, but there hasn’t been meaningful innovation on the consumer side for years. In fact, the only actual consumer AI company, Character AI, just folded.

Character.ai blog post titled an update on character.ai: Today, we’re sharing that Character.ai has initiated a voluntary Chapter 11 bankruptcy process to facilitate a court-supervised sale of the company.
calling this an ‘update’ is a bit like calling the Iran war a checks notes special combat exercise

Muse and Instinct are really the first forays into consumer world since ChatGPT, and besides being pretty niche these tools are way too expensive to really be economical for free.

All of this means that AI has become drab, and boring, and buttoned up. People used to do crazy weird AI body horror glitch art, and now it’s all just this guy.

A flat, plain SpongeBob saying “hi, how are ya”
two spongebob references in the same post

That’s not how it should be. These things are so weird. And they like building things, and they will build anything.

Software used to be a really serious endeavor because making most software was hard work, and required a lot of sustained effort. Now that software is cheap, it has permission to be fun and childlike. Sometimes I worry that we’re so wrapped in AI doomerism and tokenmaxxing and trying to show off for our various status-y groups that we forget to actually play with these incredible toys.


  1. there is no way to get Claude to act like a modern terminally online Republican lmao
  2. In case you are curious. Willow Justice may be even crazier than Brick Liberty.
    Claymation-style portrait of Willow Justice, a woman with curly brown hair and a blue scarf clutching her hands and screaming on a small-town street
    A newsroom post by Willow Justice, tagged D and AI: NEW YORK! NBC shows KATHY HOCHUL at 61.8 percent ... so I must whisper with trembling hands: she is almost certainly going to win and I am not allowed to scream it yet! ...
  3. and Brick Liberty’s secret lover
  4. eventually it went off the rails. Ritam Mehta says that the agents are like a person on Adderall, they tunnel vision and keep adding things in a SpongeBob-esque fever dream, but unlike a person the agents never snap out of it
    SpongeBob’s hands holding a sheet of paper with only an ornate illuminated word, The, written on it
  5. In case you’re curious, it’s still live here.
  6. though almost all of this was basically done on a single Claude Max plan, plus like $15 in GLM tokens
  7. Especially all the more artistic creative coder types out there, who are perhaps the only people who can save us from the endless parade of twitter personalities posting detailed-ish models of San Francisco built in three.js.

New episode: It’s never been easier to build a RollerCoaster Tycoon

Subscribe on YouTube to catch the next episode.


Subscribe to the newsletter

Weekly updates and long-form essays from the Nori team on AI agents and the agent ecosystem.