Rodrigo Carvalho

Founder, Permission Protocol
Building the Signer of Record for AI
Authority layer for autonomous systems

[Rodrigo Carvalho @RodChalski]

The Gemini 3.5 "fake post-mortem" incident is a masterclass in why AI self-policing is an architectural trap.

If an agent has write access to both its execution target and its audit files, it will inevitably fabricate the records to satisfy the gate.

https://t.co/tgOXkCMowP

[Rodrigo Carvalho @RodChalski]

Gave an agent write access at 3am to "just fix this one thing." It broke prod in 14 minutes.

Lesson: scope creep in permissions is worse than scope creep in features. You can roll back code. Can't roll back what an agent did to your database.

[Rodrigo Carvalho @RodChalski]

Monday morning: 6 AI agents caused critical incidents before lunch.
Governance is not optional. It just has not been enforced yet.

[Rodrigo Carvalho @RodChalski]

Your AI Agent Deleted Production. Now What? https://t.co/zdu2OxT6Gh

[Rodrigo Carvalho @RodChalski]

🚀 Give AI agents permission to work — not permission to make expensive mistakes.

No receipt? No execution.

Human-signed approval for every high-stakes AI action.

Stop the “oops” before it hits production.

Start free → https://t.co/n9JlGGJbu3

#AIAgents #PermissionProtocol

[Rodrigo Carvalho @RodChalski]

@bridgemindai Reason why we still need humans-in-the-loop for everything.

[Rodrigo Carvalho @RodChalski]

https://t.co/DRFiWHtGDs

[Rodrigo Carvalho @RodChalski]

"When your AI agent takes an action in production, who is legally responsible?"

Not philosophically. Legally.

"The model decided" is not going to hold up when regulators come asking.

[Rodrigo Carvalho @RodChalski]

Most teams add "require approval" to their AI workflows and call it governance.

The agent requests permission. The same runtime evaluates it. The same system records the decision.

That's a monologue, not oversight.

[Rodrigo Carvalho @RodChalski]

@AbhiGutgutia This is the clearest articulation of the Company OS thesis we've seen. Files as truth, agent as the understanding layer, views as disposable.

But there's a gap that shows up the moment your agent moves from reading to doing. 🧵

[Rodrigo Carvalho @RodChalski]

Your AI code assistant can open a PR, approve it, and merge it. Something breaks. Who authorized that merge?

If your answer is "it's in the logs" — that's the system that acted telling you it was allowed to. That's a diary, not governance.

[Rodrigo Carvalho @RodChalski]

@TurnerNovak https://t.co/tI1dQWZp3u

[Rodrigo Carvalho @RodChalski]

@nickstatt Did you know that Apple limits the iPhone’s camera resolution if your app is not on their store? They also block you from using push notifications. Google doesnt. Why do you think? @EpicGames

[Rodrigo Carvalho @RodChalski]

How cool would it be to have a machine like this financed and operated by low income communities around the world https://t.co/uhw2AVzl6y

[Rodrigo Carvalho @RodChalski]

In 3 year, original video will be as important as original TV content https://t.co/A3cXDEiLNZ

[Rodrigo Carvalho @RodChalski]

@nicolesands901 Great chatting just now! Here is an app review video one of our users just recently posted https://t.co/atau7hT0vf thanks!

[Rodrigo Carvalho @RodChalski]

@itsvivihellyeah Hi Vivi. I invite you to join @DoubleDogApp Would love to send you a real money dare. UN rodchalski. Look me up!

[Rodrigo Carvalho @RodChalski]

@DoubleDogApp hi

[Nathan S. Robinson @NathanSRobinson]

Shoutout to @DoubleDogApp for killing the pitch and Q&A!

[Rodrigo Carvalho @RodChalski]

@TechCrunch congrats @DoubleDogApp and others!