Two things have to be true before anyone is comfortable with agents in production: a human stays accountable for what gets built, and somebody knows what the agent touched to build it. CloudGeometry and Tigera have been working on each end of that problem. This session brings both together, live.
Hosted by CloudGeometry and Tigera · a roundtable conversation on AI agents and observability, with live demos to back the methodology.
Registration
Live roundtable + demos · Thursday, September 10 · 1:00 PM ET / 10:00 AM PT / 6:00 PM UK. Registrants get the calendar invite first, plus the recording.
Loading registration form…
We'll only use your details to send you the invite, the recording, and related session materials.
Trusted by engineering teams at
Approval without enforcement is a signature on something you can't see. Enforcement without accountability is a wall your team routes around by Friday.
Most teams have half the answer. Lifecycle governance tells you a human approved the change. Runtime enforcement tells you what the agent could reach. Neither is enough on its own, and the gap between them is where incidents, sprawl, and awkward audit conversations live.
One layer on its own
Both layers together
CloudGeometry runs the lifecycle. Tigera runs the guardrails. In the session, we walk through how a single agent action runs through both layers, then demo each one live.
Accountability for what gets built
Agents work on requirements, implementation, tests, and documentation, grounded in a live model of your real system, not a prompt full of guesses.
Enforcement for what agents can reach
Lynx sits in the path of every agent call and decides what that agent is allowed to reach, while it runs.
The demos tie the two together: watch one agent-built change pass its lifecycle gates while its runtime calls are allowed, denied, and recorded in real time.
Where senior-engineer sign-off gates belong when agents write the requirements, the code, the tests, and the docs.
Why agents grounded in a live model of your system produce reviewable change, and prompt-context agents produce surprises.
What per-agent identity looks like in practice, and why shared service accounts make agent activity impossible to attribute.
How credentials that expire in minutes change the blast radius of a misbehaving or compromised agent.
Why a record of denied calls is as valuable as a record of executed ones, for debugging and for evidence.
How to surface the agents already running in your environment that nobody registered, approved, or remembers deploying.
What happens when agent A calls agent B with agent C's permissions, and how identity plus enforcement keeps the chain visible.
The evidence your auditor is going to ask for about AI agents, and what it takes to have it on hand instead of reconstructing it.
Both teams on what they've learned putting agents into production: where sign-off gates hold, where guardrails get routed around, and what broke before the model worked. Candid, practitioner to practitioner, not scripted.
Two short demos in support of the methodology, not instead of it. One agent-built change passes its lifecycle gates in AI-MSL, while Lynx allows, denies, and records the same agent's runtime calls.
Bring your questions about shadow agents, delegation chains, and what your auditor is going to ask you next year. The session ends when the useful questions do.
Live, with open Q&A. Can't make it? Register anyway. Every registrant gets the recording and session materials.

CloudGeometry
What it takes to run agents on real production systems: requirements, code, tests, and docs, with a senior engineer signing off at every gate and a named person accountable for the outcome.

Tigera · Lynx
An identity per agent, credentials that expire in minutes, and enforcement in the path of every call. From the team behind Project Calico.
Announcing shortly · register for updatesBefore you say yes at scale, you need to know what accountability and observability look like in practice.
A gate model you can defend, to the team and to the board.
Broad delegated access, no fixed identity, and activity your current tooling wasn't built to attribute.
An identity and enforcement pattern for every agent call.
When something breaks at 2am, the question is what ran, what was blocked, and who approved what.
The record to answer it: ran, denied, approved, and by whom.
Eight questions on how AI agents behave inside a production environment. They're the questions your security team, your platform team, and eventually your auditor will ask, so it's better to miss them here.
Two minutes, eight questions, instant score. No grading on a curve. These are the standards agent-run change gets held to once security and audit get involved.
The teams that win with agents won't be the ones running the most of them. They'll be the ones that can say, for every change: who approved it, and what it touched.
Join CloudGeometry and Tigera for a working conversation on accountability and observability for AI agents, with live demos of both layers.
Looking for a coding-assistant comparison or a prompt-engineering tutorial? This isn't that session. It's about the operating and security model around the agents.
About CloudGeometry. CloudGeometry is a strategic engineering partner helping enterprises adopt AI-powered software delivery with governance built in. AI-MSL, its platform and managed service, maintains, modernizes, and extends production software systems, with human sign-off at every lifecycle gate, full traceability of every change, and all code and assets remaining under client ownership and control.
About Tigera. Tigera, the team behind Project Calico, builds runtime security and observability for cloud-native workloads. Lynx applies that discipline to AI agents: an identity per agent, credentials that expire in minutes, and policy enforced in the path of every call, with a record of every call that ran and every call that didn't.
AI Agents: Accountability + Observability · live Sep 10, 1 PM ET · CloudGeometry × Tigera
Save my seat