Insights archive
Page 1 of 6 · 55 posts
Guardian Agents Are Coming: What Gartner's 2028 Prediction Means for Teams Shipping Today
· Agents, Governance · 4 min read
Gartner predicts that by 2028, 40% of CIOs will demand guardian agents to oversee their AI agents. Here is what reviewer, monitor, and protector practices any team can build today, without buying anything.
Kill Switches Aren't Optional Anymore: Designing Circuit Breakers for Autonomous Agents
· Agents, Governance, Engineering · 5 min read
A kill switch that has never been tested is not a control. Here is how to design one that actually stops a run, and stays stopped.
What 89 Percent Adoption and 52 Percent Eval Coverage Tells You About the Observability Gap
· Agents, Observability, Engineering · 5 min read
LangChain's latest survey finds 89% of teams have agent observability in place, but only 52% run offline evals and 37% run online ones. Seeing what an agent did is not the same as knowing it was right.
Open Agent Safety at Commodity Scale
· Agents, Governance, Engineering · 4 min read
Nvidia's new Open Agent Safety Platform pairs a hardware watcher with hard limits on what an agent can touch. Here is what the same defense looks like built from ordinary cloud parts.
Shadow AI Is the New Shadow IT, Except the Agent Has Your Credentials
· Agents, Governance · 4 min read
Shadow AI starts like shadow IT: unapproved tools holding company data. But an agent acts with your credentials, and its mistakes land under your name.
The Governance-Containment Gap: Why 58 Percent Monitor but Only 40 Percent Can Actually Stop an Agent
· Agents, Governance · 5 min read
58% of organizations now monitor their AI agents continuously, but only 40% can quickly stop one that misbehaves. A new survey puts numbers on the gap between watching and stopping.
The Missing Layer, Part Two: Why Observability Without Containment Is Just a Dashboard
· Agents, Observability, Governance · 4 min read
Watching an agent is not the same as stopping it. Observability tells you what happened; containment decides what can happen at all — and the two only work as a pair.
The Missing Layer: Why Enterprise AI Can't See Itself
· Engineering, Observability · 4 min read
Most enterprise agent pilots stall before production, and the reflex is to blame the model. That's the wrong diagnosis. What's missing is the layer that lets a system see itself: observability.
What Ships and What Doesn't: Why Our Deploy Machinery Never Reaches the Customer
· Engineering, Governance · 2 min read
Our build tooling can merge and deploy for us. It never ships. So in the product, every change still waits for a person: not because a setting says so, but because the thing that skips the wait isn't there.
Nested Conductor: Pros and Cons
· Agents, Engineering · 6 min read
A LangGraph Conductor running inside my existing launcher: what the graph takes over, what stays outside it, and what the lab runs showed. An in-house lab build, not production.