Agentic Engineering & ToolingConference50min
Building, Shipping and Governing AI Agents on Kubernetes
A live demo of building and deploying a Kubernetes-based engineering agent with KAgent, AgentGateway, and AgentRegistry. It shows how platform configuration can reduce tool-call overhead, limit context exposure, and enforce least-privilege access without changing agent code.
talk.summaryAiDisclaimer
Tom O'RourkeSolo.io
Ep HeijtingSolo.io
talkDetail.whenAndWhere
Thursday, October 8, 13:50-14:40
Room 3
An engineering agent that checks GitHub pull requests sounds simple: find the open PRs, check their status and report which ones are ready to merge. But when I connected one to GitHub's MCP server, it was provided 44 tools, including seventeen that could modify the repository. Checking which open pull requests were ready to merge took nineteen tool calls and moved around 75,000 bytes of raw API responses to produce a nine-line answer.
In this live demo, we'll build and deploy an agent on Kubernetes using three open source projects (KAgent, AgentGateway, AgentRegistry), then show how to make it safer and more efficient through platform configuration rather than changes to the agent code.
What you will learn:
In this live demo, we'll build and deploy an agent on Kubernetes using three open source projects (KAgent, AgentGateway, AgentRegistry), then show how to make it safer and more efficient through platform configuration rather than changes to the agent code.
What you will learn:
- Build and ship: Use agentregistry to scaffold an agent from approved components and deploy it with kagent as a managed Kubernetes workload.
- Cut the overhead: Use agentgateway to reduce nineteen tool calls to two and keep intermediate API responses out of the model's context.
- Control access: Keep credentials at the gateway and apply permissions per agent, so a read-only agent is only given the tools it actually needs.
Tom O'Rourke
I joined Solo in May 2026 as EMEA Field CTO, working on securing and scaling agentic AI platforms for enterprises.
I co-founded DynamicOps in 2008 which spun out of Credit Suisse's R&D division, where I was VP and led the original product and technical vision, and it had the bank's investment behind it. We built cloud automation for datacentre self-service, went from an internal tool to a platform other financial institutions ran, grew to 120 people, and VMware acquired us in 2012.
The product became vRealize Automation. It is VCF Automation now, still a flagship in Broadcom's portfolio fourteen years later. I stayed twelve years. From 2019 to 2022 I was Global Director of Engineering for VMware Development Platform, the mandated Kubernetes platform for every VMware SaaS product. I built that organisation from zero to 70+ people across nine countries, covering engineering, SRE, architecture, product and programme management, running nine production regions plus US GovCloud against a 99.99% uptime commitment.
A lot of that time went on compliance. We codified more than a hundred controls directly into Kubernetes, so the platform enforced them continuously rather than being checked after the fact. That covered FedRAMP High, IL5 and IL6, PCI, HIPAA, ISO 27001, GDPR, ISAP and IRAP, and it is what let VMware sell into US government. The same team built the Kafka platform, Strimzi on Kubernetes, which became the mandated Kafka service across the company and carried over 80% of its consumption at more than 8 GB/s.
For my last two years I moved back into an engineering role. As a Staff 2 engineer at Broadcom I pitched and won approval for moving twenty years of VMware appliances, vCenter, NSX and Automation, onto one Kubernetes-native stack, then built the workflow engine and the task-management operator that deploy them. I left in March 2026.
I write Go, Java, Python and TypeScript, and I still prefer to be hands on. Mostly Kubernetes internals, operators and controllers, platform engineering, observability, SRE and compliance. I was running Istio in production well before I joined Solo.
I co-founded DynamicOps in 2008 which spun out of Credit Suisse's R&D division, where I was VP and led the original product and technical vision, and it had the bank's investment behind it. We built cloud automation for datacentre self-service, went from an internal tool to a platform other financial institutions ran, grew to 120 people, and VMware acquired us in 2012.
The product became vRealize Automation. It is VCF Automation now, still a flagship in Broadcom's portfolio fourteen years later. I stayed twelve years. From 2019 to 2022 I was Global Director of Engineering for VMware Development Platform, the mandated Kubernetes platform for every VMware SaaS product. I built that organisation from zero to 70+ people across nine countries, covering engineering, SRE, architecture, product and programme management, running nine production regions plus US GovCloud against a 99.99% uptime commitment.
A lot of that time went on compliance. We codified more than a hundred controls directly into Kubernetes, so the platform enforced them continuously rather than being checked after the fact. That covered FedRAMP High, IL5 and IL6, PCI, HIPAA, ISO 27001, GDPR, ISAP and IRAP, and it is what let VMware sell into US government. The same team built the Kafka platform, Strimzi on Kubernetes, which became the mandated Kafka service across the company and carried over 80% of its consumption at more than 8 GB/s.
For my last two years I moved back into an engineering role. As a Staff 2 engineer at Broadcom I pitched and won approval for moving twenty years of VMware appliances, vCenter, NSX and Automation, onto one Kubernetes-native stack, then built the workflow engine and the task-management operator that deploy them. I left in March 2026.
I write Go, Java, Python and TypeScript, and I still prefer to be hands on. Mostly Kubernetes internals, operators and controllers, platform engineering, observability, SRE and compliance. I was running Istio in production well before I joined Solo.
Ep Heijting
Ep Heijting is an experienced technology leader focused on cloud-native and AI infrastructure. He has spent his career helping enterprises adopt emerging technologies across API management, Kubernetes, service mesh and application networking, with previous roles at Apigee/Google Cloud, Kong and DataStax. Today at Solo.io his focus is on the next generation of enterprise infrastructure for AI agents, including AI gateways, MCP, agent connectivity, security and governance.