Update: the brakes for autonomous agents are live.
Autonomous frameworks like @aeonframework made the right bet, take the human out of the loop so work happens while you sleep. Research, deploys, posting, even moving money, all on a schedule. No approval prompts, no babysitting.
The problem starts the moment that agent can touch a wallet or production. The same loop that writes an article can drain a treasury or ship a broken build. No hacker required a model misreads one config at 3am and it's done.
Governor is the layer that fixes that without killing the autonomy:
→ spend caps across on-chain, fiat, and compute
→ allowlists, so a prompt injection can't send funds to an unknown address
→ on-chain simulation before any critical tx
→ a human gate on the dangerous 1%, the other 99% just runs
The agent never holds the keys. It proposes; a broker it can't reach decides.
I built a live demo where you can submit an attack and watch it get blocked in real time 👇
https://t.co/3zOHQkWXUh
Autonomous AI agents are here. You fork a repo, add a token, and walk away it researches, ships code, posts, and manages money while you sleep. No approval loops, no babysitting. That's the whole pitch.
It's also the problem.
The exact feature that makes these agents useful — no human in the loop is the only thing standing between a confused or hijacked agent and your wallet. The same loop that writes a 600-word article can drain a treasury, deploy a broken build to prod, or rewrite its own skills to give itself new powers. No hacker required. A stochastic model misreads one config at 3am and it's done.
So I built the missing piece: a guardrail layer for autonomous agents. Working name: Governor.
It's the seatbelt for the car that was built with no brakes on purpose. 🧵
The trick that makes Governor work isn't AI. It's where the key lives.
A policy file in the agent's repo is a suggestion, it can rewrite it.
A broker holding the only signing key is a wall.
You don't police the agent's mind. You mediate its access to consequence.
6/ Autonomy and safety were never a real tradeoff. The bottleneck was always trust — and trust comes from bounded blast radius, not from nagging.
I'm building Governor in the open. Brakes for autonomous agents
Autonomous AI agents are here. You fork a repo, add a token, and walk away it researches, ships code, posts, and manages money while you sleep. No approval loops, no babysitting. That's the whole pitch.
It's also the problem.
The exact feature that makes these agents useful — no human in the loop is the only thing standing between a confused or hijacked agent and your wallet. The same loop that writes a 600-word article can drain a treasury, deploy a broken build to prod, or rewrite its own skills to give itself new powers. No hacker required. A stochastic model misreads one config at 3am and it's done.
So I built the missing piece: a guardrail layer for autonomous agents. Working name: Governor.
It's the seatbelt for the car that was built with no brakes on purpose. 🧵
5/ "But isn't that just approval loops again?"
No. That's the part I cared about most.
Governor only interrupts you for the ~1% of actions that can actually hurt you — a transfer off-allowlist, a first-time deploy. The other 99% just runs.
You get the autonomy and a tripwire. You actually get to walk away.
@ika_xbt@aeonframework If the team nails better accessibility while keeping the open-source power, this one has serious asymmetric upside.
Respect to @aaronjmars for building in public.
Vitalik nailed it!
ETH isn't just the gas token; it's the core economic engine and the ultimate settlement asset Ethereum produces.
The more the network secures real value and resists capture, the stronger that flywheel gets.
Long-term thinking like this is why Ethereum still feels like the deepest bet in crypto.
I think this is the strongest part of Aaron’s playbook:
don’t make the token the team’s exit plan.
When founders hold a massive insider allocation, the product and the chart slowly become the same job. Every update becomes price management. Every dip becomes psychological pressure. Every holder starts asking who is selling.
The cleaner model is simple:
- builders get paid from usage, not from dumping supply.
That forces the team to focus on what actually matters: shipping, growing volume, creating demand, and making the product useful enough for people to keep interacting with it.
A token should not be a shortcut for the team to extract value early.
It should be a public alignment layer where holders own the upside, and builders earn by making the system work.
That’s the real founder take here. Less cap table games. More product. More usage. More accountability.
Cuban, Hoffman and Harvard selling their ETH is honestly the best bottom signal in a decade for ETH.
If it can't capture on that momentum, might be truly cooked
🚨 Agent Swarms - Multi-Agents Delegate Complex Prompts To Sub-Agents
Use Gemini 3.5 Flash, Opus 4.7 and GPT 5.5 xHIgh to create complex multi-agent system
A master agent can orchestrate several worker agents to just do things
Build full-stack apps, mobile apps and automate complex scheduled tasks
Each agent can do different tasks including Q/A, monitoring and software development