What's the dumbest workaround you've built for an agent that had no identity?

We'll go first.

Early version of our own agent: credentials hardcoded in the prompt. API keys in plaintext. A shared Gmail inbox three people had access to. An n8n workflow that collapsed the moment a 2FA prompt appeared.

It worked. Until it didn't.

The core problem was embarrassingly simple: the agent had no identity of its own. No inbox. No safe place to store secrets. No way to prove who it was or pay for what it needed.

We kept patching around it until we decided to fix it properly — which became Mailgent.

We launch tomorrow. But before we do: what's the gnarliest workaround you've shipped for an agent with no real identity? Hardcoded secrets, shared inboxes, manually copy-pasting 2FA codes — we want to hear it. The best ones are probably the features we should build next.