团队如何落地 AI：从 MCP 工作流到 guardrails 的鸿沟

Hi, I’m a part of an organisation where we are very supportive to AI.

We are have topical roles from a SAAS provider, that means we go from the guys who is a nerd to the most eccentric one that doesn’t know anything about computer, but knows every single details from the business.

The talk now is “AI come for a change”, “you shouldn’t be doing this repetitive task without using AI to do it for you”

And yes, that is what happens when you talk with devs, they see a Jira ticket, they connect MCP, git, workflows and everything else. Work done.

Then you talk to others who doesn’t know AI is a thing (some employees seem to not even read their email to the point they didn’t know they had access to a paid subscription).

Now we are discussing guardrails. What shared skills, configurations, force an agent to do this and that. While others can only use AI as a chat box.

I personally see all that talk about guardrails almost an impossible thing given how easy is to ask Claude to go and change its settings.json or share my database password and ask it to go and query the database for me. Which means whatever we talk, is easily bypassed or reconfigured locally.

What has been your guys experience? How are you guys handling teams? Your privileged access vs the AI access? Your guardrails vs “do whatever while I go sleeping”? How do you bridge the gap from 0 to hero in the meetings where you have the most AI believer and the paranoid?