At AGNTCon Europe, ensuring AI agents don’t kill us all
SiliconANGLE Jason Bloomberg
At AGNTCon Europe, vendors showed off AI agents built to be controlled, not feared. The surprise: the hottest pitch was governance, audits, and keeping agents on a leash.
Based on reporting by SiliconANGLE, Jason Bloomberg — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
At AGNTCon + MCPCon Europe 2026 in Amsterdam, the loudest message wasn’t panic about runaway AI. It was the opposite: vendors kept showing ways to make agentic systems safer, more legible, and easier to run in actual businesses. The market is still young, and the gap between flashy demos and paying customers is obvious. But there are real deployments in the mix, not just hand-waving.
A lot of that work starts with control. Traefik Labs is turning its proxy and ingress roots into an API, AI, and MCP gateway built around a single binary, with the pitch that it can watch how agents interact with their environment and provide a sovereign trust plane. Bluerock Security is taking a different route, using its earlier microkernel hypervisor work as the base for a governed AI workspace. It lets builders use AI coding agents on their desktops while controlling model access, token use, enterprise identity, and even scanning MCP servers and agent skills.
The same obsession with control showed up in orchestration. Orkes, which grew out of Netflix, is pitching deterministic agent workflows plus explainability, so operators can inspect and track what happened. Grape Up’s Aiboostr goes after the same problem from a European compliance angle, lining up with the EU Artificial Intelligence Act by tracking inventories, owners, purposes, and users. Different routes, same thesis: if agents are going to do useful work together, somebody has to keep the rails straight.
The plugin story was similar. Manufact offers a cloud MCP platform for building and running components, including LLM plugins, with analytics, testing, and compliance for production use. Alpic uses plugins to bring AI into consumer workflows on the desktop, where a chat request can surface a mini-application that helps complete a booking. That sounds neat until you remember how quickly a helpful interface can become an unlicensed workflow engine.
The most interesting pitch, though, came from Reboot. Instead of letting AI spit out code and hoping humans can keep up, it pushes developers up a level to domains and behaviors. The dashboard is meant to be a live model of the application itself, with the backend handling durable state, transactions, and reactive front ends. That is the cleanest answer at the show: let the machines generate, but keep architecture and behavior under human control.
My take — AI-written commentary, not fact-checked reporting
This is the right instinct, and frankly the only sensible one. Agentic AI without governance is just automation with a better press team. The industry keeps pretending the real innovation is bigger models; the actual work is making sure the thing doesn’t wander off with the company’s keys.
Read more about this at: SiliconANGLE