This post was originally published on LinkedIn.

1. Agent governance: could your agent send your APIs keys to anyone on the internet (like in the image below)? Do you even know what your AI SDR agent is saying to prospects? Did you give your agent both read and WRITE permissions?
2. Security: Claude Mythos will probably be able to hack your vibe coded CRM. Can someone prompt inject your agent to send your latest list of customers to them?
3. Permissioning: Who on your team has connected Claude to SFDC via MCP? What data does your agent have access to? Can anyone find sensitive data using AI search in your systems?
4. Guardrails: is your probabilistic Skill too vague and could cause damage? Who's monitoring all the new automations?
5. PII: how much PII are you leaking into models? You going to get a GDPR bill sometime soon?
6. Token costs: how are you tracking token costs? What limits do you have? Can one person create an accidental $100k bill at the end of the month?
7. Tech debt: how complicated are the things that you're building right now? Who is maintaining and monitoring them?
8. Data foundation: is your data a mess and therefore going to corrupt any agent on top of it?

All these things have been important but right now they're more important than ever as the cost of mistakes gets higher and higher.

We're still in the wild west innings with AI GTM, but people are going to get fired over mistakes here

And there's going to be whole new categories of jobs created around this stuff