Architecting the onchain economy through B3OS, Labs, & Holdings
- Tweets 2,092
- Following 151
- Followers 144,325
- Likes 6,429
Today we're launching B3OS in public beta.
Onchain workflows any agent or human can run.
Live today:
𓏠 Traders copy-trading on Polymarket or trading X headlines
𓏠 Developers powering onchain apps with scheduled sends, recurring swaps, and more
𓏠 Teams paying vendors in USDC from email or Slack
Build in our UI or through our MCP server.
This is the future of autonomous onchain finance.
33
50
187
215,011
Ben Cera
B3 retweeted
Crypto but Automated
going to be trying some ready templates
I absolutely love the AI x crypto merger
there is no doubt AI will be the UI for crypto if it has to go mainstream
exploring @b3dotfun
Cursor, Claude, and Codex are good at reasoning about what to do.
Point one at a crypto task, though, and it hits a wall on execution. It falls back to a public RPC because it can't find a reliable one, can't locate the right contract address, and struggles to fund an account.
None of that is the agent's job to solve. It belongs to the execution layer.
Our CTO, @seangeng on why B3OS is that layer ↓
3
7
1,413
Cursor, Claude, and Codex are good at reasoning about what to do.
Point one at a crypto task, though, and it hits a wall on execution. It falls back to a public RPC because it can't find a reliable one, can't locate the right contract address, and struggles to fund an account.
None of that is the agent's job to solve. It belongs to the execution layer.
Our CTO, @seangeng on why B3OS is that layer ↓
6
4
32
3,740
People already ask LLMs like ChatGPT and Claude what to do with their money.
On Moonshots hosted by @PeterDiamandis, @brian_armstrong CEO at @coinbase points out the catch: the models have no context on what's in your accounts, and no way to make a trade or send a payment.
That's starting to change. More exchanges now let agents connect through MCP, so the model can act on what you ask. But it acts only while you're there. You ask, it answers, you confirm. The work happens inside the session and ends when you close it.
A B3OS workflow doesn't stop when you do. You build it once and it fires on its own triggers, a schedule, a new block, a contract event, then executes onchain without you in the loop.
Armstrong is mapping the demand. The agent economy needs more than a model you sit and talk to. It needs workflows that keep running once they're live.
Brian Armstrong
4
9
23
2,846
hiring a GTM Lead @b3dotfun
we're an ex-Coinbase team building B3OS, an onchain engine for AI agents (in beta now and launching publicly soon)
your job is to get traders and enterprises actually using it. you just need to learn fast. high slope over high starting point
we're an AI-native co, so you'll pick up a lot fast, working with onboarding to the latest AI tools, skills, open-weight models and more
if this gets you fired up, dm me or apply below:
b3labs.org/jobs
30
12
190
16,450
Every AI agent you use today waits for you to open it.
You message Claude or ChatGPT, it answers, you close the tab.
That works for a search box, but it's not how you'd work with a good quant or teammate.
We're building toward something closer to that: talking to Caddie in Telegram, Slack, or WhatsApp, wherever you already work, the same way you would a person.
Our CTO, @seangeng on what's next for B3OS:
1
6
25
3,214
when I hear people talking about wanting an agent (all the rage these days for businesses), a lot of the times I think they actually want a workflow
the difference between our agent (B3OS) and other agents is
- you prompt
- we build a with predefined actions, triggers, all lego blocks
- executor engine runs it all
this is more deterministic (agent is the plan composer, not the executor) and reliable, with real auditability
vs
agent spins up code in a VM, ???
you get totally different architecture, unreliable, disappears into the ether & unmanaged mess for your team
some thoughts I wrote about our approach here:
seangeng.com/writing/propose…
1
4
12
1,053
Does every AI task need the best model?
@darylX24's take on @latenightonbase:
The critical, high-value workloads will run on the frontier models.
The routine tasks, like scraping the web or pulling tools together, will run on open-weight models that continue to improve.
3
5
24
1,448
How jailbreaking models work:
People talk about “capabilities” and “jailbreaks” as if they’re fundamentally different things.
I increasingly think they’re the same phenomenon.
A large model is basically a giant machine for creating connections between concepts.
Bigger model?
- More dimensions to represent ideas.
- More layers to reason over them.
That’s it.
The magic happens when you can get half the model to light up at once.
That’s where you get the “holy shit, it connected those two things?” moments.
The funny part is that jailbreaks exploit the exact same mechanism.
You’re not usually beating the model with some clever string.
You’re trying to activate a ridiculous number of concepts, contexts, analogies, roleplays, abstractions, edge cases, and reasoning paths simultaneously.
Basically constructing the world’s most autistic treasure hunt.
A sufficiently capable model starts exploring all of it.
Most paths die.
One survives.
Congratulations, you’ve found the path that outputs the thing you wanted.
The uncomfortable question for alignment is:
At what point does a jailbreak just become a capability benchmark with different incentives?
Because from where I’m sitting, “creative reasoning” and “creative rule circumvention” appear to be powered by the same underlying machinery.
3
10
1,161
You guys no idea how badly I want to tell everyone about what we are working on with @b3dotfun
Unstoppable, uncensorable, global decentralized AI seems like a good investment bet to make. The “Bitcoin of AI” so to say…
6
5
16
1,872
Why doesn't a single AI agent understand crypto the way it understands everything else?
We've spent the year making B3OS the most natural crypto agent you can use.
The hard part was never the reasoning, it was the execution: landing onchain actions reliably, keeping audit trails, and making sure that when Caddie fires a transaction or pings one of our 1,000+ connectors like Slack or Telegram, it happens the way you asked instead of the way a hallucinating model guessed.
Our CTO @seangeng on where Caddie goes from here:
5
6
29
1,383
In @PanteraCapital's The Convergence of AI and Blockchain, @cosmo_jiang makes the case that AI and blockchain push each other forward.
Blockchain gives AI open systems to pull from. In return, AI abstracts away the complexity that has held crypto back.
Abstracting it away is the hard part.
The rails exist and the models can reason, but the layer that turns intent into reliable execution onchain is still early.
5
12
55
3,838
