Your AI agent just did something you didn't approve. Did you know?

Probably not. And that's exactly the problem.

AI agents are no longer a side experiment. They're inside real businesses right now — reading files, writing code, sending messages, making decisions. And in almost every case, the team that deployed them has zero visibility into what those agents are actually doing once they're running.

Think about that for a second.

You have handed an autonomous system access to your tools, your data, your infrastructure, and you have no audit trail, no way to pause it, no kill-switch. If it does something wrong, you won't know until after the damage is done.

I couldn't stop thinking about this. So I built TrustLoop.

What it does:

TrustLoop sits between your AI agent and the tools it uses. Every single action gets intercepted, logged, and checked against your governance rules, in real time. If an agent tries to do something it shouldn't, you block it instantly from a dashboard. One click. No code changes.

It works with Model Context Protocol (MCP), which is quickly becoming the standard for how AI agents use tools, and plugs into Claude, GPT, or any LLM-powered agent in minutes.

Where I am at:

• Shipped and live, not just a landing page

• Claude Desktop integration working end-to-end with real governance enforcement

• Real time dashboard showing every agent action, blocked or allowed

• Early Access open at trustloop.live

I am Soji Joseph, building this solo from London. No investors yet. No big launch. Just a problem I kept seeing everywhere and a product that I genuinely believe needs to exist.

This is my first time talking about it publicly. Would love brutal, honest feedback from people actually building with agents.

If you are shipping AI agents, what is your current answer when someone asks what did your agent actually do today?