As KubeCon + CloudNativeCon North America 2026 approaches, the conversation in cloud-native circles is shifting from building AI models to running AI applications reliably in production. With CNCF debuting a new AI Inference + Agentic track, much of the attention has gone to GPU scheduling, model serving, and observability. Ajay Khanna, CMO at Yugabyte, says that the data layer underneath is just as critical. Next-generation applications, he says, won’t only serve people. They will also support fleets of autonomous AI agents that need real-time access to data, and that changes what database infrastructure has to do.
In this VMblog pre-show Q&A, Khanna explains how Yugabyte is responding with YugabyteDB AMP (Agentic Multitenant Postgres), a serverless, scale-to-zero Postgres tier on Kubernetes that gives every agent its own isolated database, and Meko, a context engine that gives agents governed shared memory and knowledge. He also covers platform engineering, multi-cloud strategy, FinOps, and ROI for the C-suite. Attendees can meet Yugabyte engineers at booth #153 and join the free Distributed SQL Summit on November 9 at the Gallivan Center in Salt Lake City, themed “Distributed PostgreSQL for the AI Era.”
Read on for Khanna’s full take on where cloud-native is heading in 2027 and beyond.
++
VMblog: What’s your elevator pitch for KubeCon + CloudNativeCon North America 2026? If attendees only remember one thing about your company after visiting your booth, what should it be?
Ajay Khanna: The next generation of cloud-native applications won’t only serve people. They will also support fleets of AI agents that work autonomously, access data in real time, and collaborate at scale. That shift changes what data infrastructure has to do.
At Yugabyte, we’re building on a decade of experience running mission-critical applications on distributed PostgreSQL. YugabyteDB AMP (Agentic Multitenant Postgres) is a serverless, scale-to-zero Postgres tier that runs on Kubernetes and lets every agent have its own isolated database, spun up in seconds. Meko, our context engine for the agentic enterprise, provides those agents with governed shared memory and knowledge, enabling them to learn from one another rather than repeat work.
The big takeaway: One stack, from distributed PostgreSQL to agent-operated database fleets to a shared context layer, built for agents that work and learn together.
VMblog: CNCF has introduced a brand-new “AI Inference + Agentic” track for 2026, covering GPU scheduling, model serving, and observability for production AI systems. Where does your company fit into that conversation, and what’s driving your investment there?
Khanna: The new track reflects an important shift: the conversation is moving from building AI models to running AI applications reliably in production. GPU scheduling, model serving and observability get most of the attention, but the data layer underneath is just as critical.
Agentic workloads are unpredictable. Agents generate sudden bursts of database activity, need several kinds of data, and operate independently across distributed environments. YugabyteDB AMP is built for that pattern: serverless, multitenant PostgreSQL that gives each agent its own isolated database while sharing the underlying infrastructure, with scale-to-zero so idle agents cost almost nothing. Teams can spin up hundreds of databases in seconds, set per-pool quotas, and scale AI workloads without adding cost or operational complexity.
VMblog: Beyond the booth presence, what’s your broader community engagement strategy for KubeCon NA 2026? How are you contributing to the cloud-native ecosystem’s growth?
Khanna: Our annual Distributed SQL Summit (DSS) is the centerpiece. It takes place November 9 at the Gallivan Center in Salt Lake City as an official KubeCon + CloudNativeCon co-located event, it is free to attend, and this year’s theme is “Distributed PostgreSQL for the AI Era.” Yugabyte co-founder and co-CEO Karthik Ranganathan will be joined by our solution engineering and developer advocacy teams.
For the community, we are also offering a free hands-on lab with certification starting Nov 09, 10:00 AM–12:00 PM MST: “From Zero to ADLC: Set Up Your Agentic Dev Lifecycle with Meko in One Session”
The day brings together developers, architects, customers and industry experts for keynotes, breakout sessions, fireside conversations with executives who have modernized their data layer, live demos and partner content. The emphasis is practical: what is working in production today, from scaling mission-critical applications to building agentic AI systems on PostgreSQL, rather than what is hyped.
We’ll also be at KubeCon itself, so attendees can meet our engineers at booth #153. Our goal across both is to share what we’ve learned and give developers tools they can use right away to build more resilient, scalable applications.
VMblog: Let’s get technical — what specific architectural challenges or operational bottlenecks does your solution address for cloud-native teams heading into 2027?
Khanna: Cloud-native teams are managing increasingly distributed, unpredictable workloads without wanting more infrastructure complexity. Traditional databases struggle to deliver the horizontal scalability, resilience and low-latency access these applications need. YugabyteDB addresses this with distributed PostgreSQL: familiar developer tools, plus ultra-resilience, high availability, and multi-region scalability.
Agentic AI adds two more challenges. The first is database sprawl. Multi-agent applications often require developers to stitch together a memory provider, an observability tool, a RAG stack, artifact storage, and a vector database. Meko context engine folds all this into a single unified data layer that provides shared context where agents do not repeatedly retrieve the same information or rebuild context from scratch, burning tokens unnecessarily. Meko is a unified context engine, with shared memory, knowledge, conversations, and decision traces behind one MCP endpoint, so multi-agent systems stop paying for the same work twice.
Second is seamless scaling to manage 1000s of database fleets. YugabyteDB AMP provides serverless multitenancy: each agent gets an isolated PostgreSQL database, on a serverless compute tier that runs on Kubernetes, without the cost of dedicated infrastructure.
VMblog: The industry narrative has shifted from “Kubernetes for web scale” to “Kubernetes for AI scale.” How has your technology evolved to meet that shift, particularly as inference is projected to overtake training as the dominant AI compute workload?
Khanna: AI changes both the scale and the shape of application workloads. Agents produce unpredictable bursts of activity and increasingly need their own isolated data environments, and as fleets of agents grow, managing them efficiently becomes the challenge.
YugabyteDB AMP, introduced in YugabyteDB 2026.1, is our answer. It is serverless, scale-to-zero, multitenant PostgreSQL whose compute tier runs on Kubernetes, with an API and MCP-based controller that creates and manages the fleet. Each agent gets its own isolated database on shared infrastructure, and agents or humans can branch a database with zero-copy thin cloning, treating data like code. Underneath, YugabyteDB’s distributed architecture provides resilience. Together, they let organizations scale AI applications without sacrificing reliability or adding unnecessary cost.
VMblog: Where does your solution fit within the modern cloud-native stack? How do you integrate with and complement projects like vLLM, KServe, Ray, and OpenTelemetry that are anchoring the new AI Inference + Agentic track?
Khanna: YugabyteDB sits at the data layer, providing distributed PostgreSQL for transactional workloads and applications that need scalable, resilient data infrastructure. YugabyteDB AMP extends that layer with isolated, serverless Postgres databases for agents.
Meko, the context engine, sits above it as the context layer of the agentic stack. vLLM and KServe handle inference and model serving, Ray handles distributed workload orchestration, and OpenTelemetry handles observability. Meko handles what those don’t: conversations, memory, knowledge, decision traces and artifacts, exposed through one MCP endpoint that different agents and tools can share. Observability within Meko shows what an agent did; Meko adds what it learned and why it decided, so other agents can build on it.
These pieces address complementary parts of the stack, giving developers both the infrastructure to run AI reliably and the shared context agents need to collaborate.
VMblog: Agentic AI workflows are emerging as a new class of Kubernetes workload, with autonomy and orchestration requirements that differ from traditional services. How is your platform adapting to support these agentic systems?
Khanna: Yugabyte is working on the problem of context in multi-agent workflows. If agents work in silos, they repeat research, retrieve the same information, and rebuild context that another agent already built. Giving them more context isn’t the fix either: too little context produces poor results, and too much overwhelms the model and dilutes quality.
Meko is built to solve that context paradox. It structures context automatically, tracks where every piece came from, and extracts only the relevant subset for the model, so quality holds up as context grows. It also puts humans in charge of what the organization learns. Memory is private by default and shared by choice: an agent discovers something, a human validates it, and Meko promotes it to shared knowledge with full provenance, so every future agent can use it. Decision traces show developers how agents reached their conclusions.
That is the step from running many independent agents in parallel to AI systems that collaborate under governance.
VMblog: Platform engineering continues reshaping how organizations approach cloud-native infrastructure. What’s your perspective on this shift, and how does your technology enable effective platform teams and internal developer platforms?
Khanna: Platform engineering is increasingly about giving developers self-service access to infrastructure while helping platform teams maintain reliability, governance, and consistency. Databases are a critical part of that equation, particularly as organizations scale applications across environments.
YugabyteDB lets teams standardize on PostgreSQL-compatible distributed infrastructure that scales across regions and clouds, so developers get a familiar database and platform teams get the availability and scalability they need. YugabyteDB AMP adds a fleet-management layer for self-service. Platform teams create shared pools with quotas and policies for the databases a team or project uses, and developers or agents provision databases from those pools through an API, UI or MCP. Zero-copy branching lets teams spin up dev and test copies of data quickly.
As AI applications add new layers of complexity, simplifying the underlying data infrastructure matters even more.
VMblog: Multi-cloud and hybrid strategies are now table stakes, and GPU capacity is often the scarcest resource across those environments. How does your solution help organizations navigate distributed, heterogeneous infrastructure — including accelerator availability?
Khanna: YugabyteDB was designed for distributed deployments across regions, clouds and data centers. Organizations keep their data resilient across environments and place it close to applications and users, while retaining PostgreSQL compatibility. Its storage foundation can span multiple zones, regions or clouds, and customers can run the AMP fleet in their own environment through bring-your-own-cloud deployments.
GPU scheduling happens elsewhere in the stack. Our role is to make sure the data layer keeps up as compute demands and deployment strategies change, so an application can move between environments without its database becoming the constraint.
VMblog: As we head toward 2027, what’s your unique value proposition? With dozens of vendors claiming similar AI-infrastructure capabilities, why should CTOs choose your solution?
Khanna: Most AI-infrastructure vendors solve one slice of the problem. We offer one stack on one engine: YugabyteDB for distributed PostgreSQL, AMP (Agentic Multitenant Postgres) for agent-operated fleets of isolated databases, and Meko as the context engine on top. That means a CTO doesn’t have to assemble and secure five or more separate systems for SQL, memory, knowledge, traces, artifacts, and data.
It rests on a decade of building distributed data infrastructure for some of the world’s most demanding workloads, and it keeps what teams already rely on: PostgreSQL compatibility and its developer ecosystem. Meko is MCP-native and works with the agents teams already use, so context isn’t locked inside one application.
Enterprises can modernize existing applications today and prepare for agentic AI, without trading away performance, reliability, or operational simplicity.
VMblog: Which conference tracks, keynote themes, or technical topics are you most excited about at KubeCon NA 2026? Where do you see alignment with your strategic vision?
Khanna: The new AI Inference + Agentic track stands out because it shows how quickly the industry is moving to production AI systems. I am especially interested in the infrastructure questions that appear when organizations move from individual assistants to applications built on many collaborating agents: how each agent gets its own data, how shared context is governed, and how teams keep costs predictable.
That is the same ground we’ll cover at the Distributed SQL Summit on November 9, themed “Distributed PostgreSQL for the AI Era.” Attendees can start the KubeCon conversation there with hands-on, practical sessions on running distributed PostgreSQL and agentic infrastructure in production.
VMblog: Give us your executive pitch: how do you articulate ROI and business value to C-suite leaders who need to justify cloud-native and AI infrastructure investments?
Khanna: For executives, the ROI of cloud-native and AI infrastructure comes down to lower cost, higher efficiency and room to grow. Legacy databases carry hidden costs in downtime, complex maintenance and expensive licensing.
Yugabyte improves the economics of the data layer by minimizing downtime, reducing operational overhead and scaling efficiently. A recent ROI study found that a model $2.5B enterprise running a Tier 0 application on YugabyteDB can realize a 3-year net benefit of about $15M and 181% 3-year ROI. For AI, the economics are still favorable. YugabyteDB AMP uses usage-based pricing, so entry cost is near zero and you pay for what runs, and scale-to-zero means idle agent databases aren’t a standing expense. Meko reduces the tokens spent when agents share the context. As AI adoption grows, the goal is to control cost and support new workloads without a costly rearchitecture.
VMblog: FinOps and cost optimization have become even more critical as GPU spend joins cloud spend under scrutiny. How does your solution help organizations maximize efficiency and control costs across both traditional and AI workloads?
Khanna: Cost efficiency depends on how well infrastructure is used and how effectively applications use it. YugabyteDB scales distributed workloads horizontally. YugabyteDB AMP adds serverless multitenancy with scale-to-zero and per-pool quotas, so many independent workloads, such as one database per agent, shared infrastructure and idle databases stop costing money.
For multi-agent AI, Meko tackles a different source of waste: duplicated work. When agents repeatedly retrieve the same information and rebuild context, token consumption grows. Meko sends the model only the relevant subset of context and lets agents reuse what the system has already learned. In our internal testing, this substantially reduced the LLM input tokens.
Neither approach removes the need to manage AI compute costs, but both address inefficiencies that become more significant as applications scale.
VMblog: Looking beyond KubeCon NA 2026, what emerging trends or technologies should cloud-native practitioners be preparing for? What’s on your company’s innovation roadmap that signals where the industry is heading?
Khanna: Applications are shifting from serving mainly people to serving people and AI agents. As multi-agent systems become common, developers will need to think beyond model performance to memory, shared knowledge, coordination, governance and visibility into agent decisions. Context is becoming an enterprise asset to be engineered and governed, not something each application accumulates on its own.
Our investments in YugabyteDB AMP and Meko reflect that direction. AMP gives agents isolated, serverless Postgres with a no-rewrite path to dedicated YugabyteDB as they grow. Meko is moving to open source, so developers can run it on their own laptops, with Meko also available as a managed service and on private clouds.
We believe the next phase of cloud-native innovation will be defined not just by running more AI workloads, but by making increasingly complex AI systems more efficient, reliable, and practical to operate.
