Notes on governed automation
How we think about putting AI to work over real business tools, where it helps, where it must stop and ask, and why the guardrails are the product.
-
Principles
Why we don't let the AI send anything
Most AI tools race to act. Taskpath's AI drafts and proposes, but a person approves anything that leaves the building. Here's why that's structural, not a setting, and how it still scales.
Read it · 4 min →
-
Governance
One AI wrote people off. The other stopped and asked.
The New York Times ran a staffing-budget task through an unguarded AI agent, and it quietly marked employees on leave as cuttable. We ran the same task through Taskpath. It stopped and asked a person. Here's exactly what each did.
Read it · 4 min →