Blog

Notes on permissioned AI agents for internal operations: guardrails, approvals, audit, and how GlowForm compares to the tools people ask about.

1. Connecting an AI agent to your database without regretting it Read-only credentials proven at the engine, scoped rows, bounded queries and no free-form SQL by default: the checklist. September 21, 2026 · 5 min read
2. Why operations files a ticket for something it could do itself The ticket is a permission request in disguise. What it costs, and what replacing it with scoped, approvable access looks like. September 21, 2026 · 4 min read
3. Chat-first internal tools: why the interface is a conversation The long tail of operational questions never gets a screen. A conversation, with sources and permissions, is the screen it never got. September 21, 2026 · 5 min read
4. What an audit log for AI agents has to contain Proposal, decision, approval, action, receipt — and why the log has to be append-only and written with the change, not after it. September 21, 2026 · 5 min read
5. Least privilege for AI agents, in practice Service identities, row-level scope, environment splits and daily caps: how to give an agent exactly the access the person asking has, and no more. September 21, 2026 · 6 min read
6. Human-in-the-loop, done properly: approvals for AI agents Where the approval sits, what the approver sees, and why the approval has to be a permission in its own right. September 21, 2026 · 6 min read
7. GlowForm vs Retool: dashboards you build versus questions people just ask Retool is a mature internal-tool builder. GlowForm skips the building and lets people ask, within permissions. When each one fits. September 21, 2026 · 5 min read
8. GlowForm vs Lovable: building an app versus governing what an agent may do Lovable builds new software from a prompt. GlowForm lets people act on the software they already run, within permissions. They solve different problems — here is how to tell which one you have. September 21, 2026 · 5 min read
9. Agent guardrails that actually hold Prompt rules are suggestions. Guardrails that hold are enforced by code, scoped to a person, and leave a record — here is what that looks like. September 21, 2026 · 6 min read
10. Agent permissions: what an AI agent should be allowed to do, and how to say so A practical model for permissioning AI agents inside a company: read, propose, act, and the approval that sits between them. September 21, 2026 · 7 min read