I started working on Preflight AI because AI is no longer just answering questions. AI agents are starting to take real actions inside systems, and that made me think about a simple question: who checks an AI action before it actually happens?

That is what I want Preflight AI to do. It sits between an AI agent and the action it wants to take. It checks what could change, who might be affected, and whether the action looks risky. If the risk is high, a person should have to approve it first.

I already have an early working prototype. It can simulate an action, evaluate the risk, ask for approval when needed, execute an approved action, and keep a record of what happened.

I built the prototype using AI-assisted development tools. I am not a traditional software engineer, and I think it is important to be clear about that. I am the person behind the idea and the direction of the project, and I have taken it far enough to see that there is something worth testing seriously.

What I want to do next is find out how far this idea can go with the right technical help. I would like to work with experienced engineering and security people, improve the way Preflight evaluates risk, connect it to more realistic environments, and test whether it can actually stop harmful or unintended AI-agent actions before they happen.

For me, the idea comes down to something simple. If we are going to give AI agents more power to act for us, I think people should still have a way to check important actions before they become real.