Menu
Research
What production AI is teaching us.
Research, field notes and perspectives from systems we build, test and operate.
Research / Featured
Same agent. Very different bill.
A prospect asked how the two compare, so we protected the same agent with AWS’s AgentCore and with ActaClad’s AgentGuard.
Read article ↗One prompt to add AgentGuard.
We packaged our integration know-how into a skill for AI coding agents. Your agent maps the app, wires AgentGuard in and proves it works. You review the diff.
Check every answer.
What we learned about evaluating every production AI answer.
Boundaries, not breaches.
What scanning thirteen widely used open-source AI projects taught us about production AI.
Trust starts before production.
Two guards, one lifecycle.
Put the policy at the tool call.
A requested action is not permission to execute. AgentGuard tool permissions can allow a tool, make it read-only, require approval or block it. The approval state belongs beside the tool action, not inside the model prompt.
Evidence before a number.
A score without enough evidence can hide uncertainty. When evidence is insufficient, the AgentGuard Trust Score reports insufficient data instead of a number.