Would you give an AI write access to your production database?

We just shipped this in Basedash and my honest first reaction when we scoped the feature was no way, absolutely not. I've spent 6 years building tools that touch production data, and the whole time the rule was that humans click the buttons.

But we did it anyway, and the process of getting comfortable with it changed my mind about where the risk actually lives.

What got us there:

• Admins choose which databases the AI can edit at all. Read-only is the default.

• Every tool and connection has its own permission, so you can let it update Stripe but never touch your users table.

• Consequential actions pause and wait for a human to approve them before anything runs.

The thing I underestimated is how much of the danger comes from humans, not the AI. A tired engineer running a hand-written UPDATE query at 6pm on a Friday is probably riskier than an agent that writes the SQL, shows it to you, and waits for approval.

I still wouldn't give an agent unsupervised write access to everything. There's a version of this that goes very wrong, and approval fatigue is a real failure mode (after enough approvals, people stop reading them).

We just launched this today if you want to see how we approached it: https://www.producthunt.com/products/basedash/launches/basedash-actions

So where's your line? Would you let an AI write to prod with guardrails, or is read-only the permanent ceiling for you?