The agent workaround everyone cheered for

I came across a post on my timeline about how a “non-developer” solved a problem that a SaaS product team couldn't accommodate in months. The ask was to give a new employee access to all existing resources in one click, instead of manually going through every resource and adding them.

The chain of communication for a request like that can be quite long on its own:

client → customer support → product → prioritization → engineering → prioritization → design review → implementation → testing (if it exists) → deployment

The client used an AI agent to automate the manual requests from their end through a CLI, claimed victory, and shared it with a larger audience. Everyone rallied behind the success, and this pattern could apply to most flows.

I didn't want to spoil the party, so I avoided commenting on the thread. Here's why it's concerning.

Why it's concerning

1. It exposed that the SaaS service has no rate limiting, which now makes it a vulnerable target for DDoS.
2. The browser is a sandbox that allows safe interaction with websites. It has safety rules developed over time, and one of them is session management with cookies.
3. Websites rely heavily on cookies for session and state management. Cookie expiration is used as the logout mechanism, and interestingly, it's often the browser that remembers the expiration, not the server.
4. The user had the agent start the session in the browser, then moved it to the CLI for automation. Technically, that's session hijacking.
5. The session state is sent in LLM interactions. At that point, the authorized session has already crossed the boundary of your machine.
6. Most products have no reliable way to explicitly revoke sessions. Logging out of the browser is pointless now unless the product is sophisticated enough, which I highly doubt; otherwise it would have detected the change of client.
7. The cat is out of the bag. Depending on the product, the authorized session can now be used to pretty much bring the client down.

Who should be liable: the client who used the product in a way it wasn't meant to be used, or the service that didn't add the controls?

I can definitely see products and services changing their terms of service to be explicit about how they're meant to be used.