Solo.io announced today, November 10, 2026, that its enterprise releases of Agent Substrate and kagent are generally available. If you build AI agents, this one is worth ten minutes of your attention — not because of the announcement itself, but because of the problem it names: agents are becoming the next major workload for Kubernetes, and the way we run them today does not scale.

I haven't run Agent Substrate myself. Nobody outside Solo's launch customers has had the enterprise release in general availability before today. This article is built from Solo's announcement (GlobeNewswire, November 10, 2026, 10:00 ET) and checked the same day. Where a claim is Solo's, I'll say so — because a launch-day claim is a vendor claim, not an independent result.

The problem: one pod per agent doesn't scale

Here's the core argument, and it's a good one. An agent session carries its own state and context across multiple turns, and at tens of millions of sessions, giving each one a dedicated pod ties up resources while the agent waits — on a model, on a tool, on a human. Most of the time, an agent session is idle. Paying for a pod to sit idle is the kind of waste that only shows up when you go from demo to production.

Agent Substrate's answer is an execution model that suspends idle sessions and resumes them on demand, on the Kubernetes infrastructure enterprises already run. You don't rip out your cluster and buy a new platform; the substrate runs on top of what you have. That's the pitch, and it's a sensible one — Kubernetes is everywhere and well understood, which is exactly the point Google's Tim Hockin made in the announcement, calling the goal "to make the orchestration of [agentic workloads] a non-issue."

The actor: a runtime unit with isolation built in

Agent Substrate introduces the actor as the fundamental runtime unit for agents. Each actor bundles an agent session with two mandatory controls: a sandbox that isolates execution, and a gateway that governs who can reach it and what it can access. State, identity, and access controls stay with the actor even as the compute underneath changes.

That last detail matters more than it sounds. In most agent setups today, isolation and governance are things you bolt on — a wrapper here, a proxy there, a policy you hope every tool call respects. Here they're integral to every session by design. If an agent can only ever act through its sandbox and its gateway, a whole class of "the agent did something it shouldn't have" failures gets structurally harder. That's the promise, at least. Whether it holds up in a messy real deployment is the question launch customers are now answering in production.

kagent: the developer experience on top

Agent Substrate is the infrastructure layer. kagent is the part developers actually touch: the experience for building and managing agents, tools, and skills. Solo's enterprise distributions pair the two, with agentgateway as the actor's gateway, enforcing identity and policy controls on ingress and egress.

The enterprise feature list is exactly what you'd expect a bank or a regulated platform to ask for: granular access control, agent tracing, infrastructure monitoring, and evaluations built from real execution traces. Enterprise support and deployment options — including air-gapped installations — are part of the package, for environments where the agent runtime can't phone home.

Solo also published a sizing calculator on the kagent project site, distilled from its own performance and scale testing for GA, to help teams size environments and set service-level objectives. A vendor's own sizing tool is not an independent benchmark, but it's more useful than nothing: it's the first concrete hint of what "tens of millions of sessions" costs in real resources.

Who's already running it

Two launch customers are named, and they're the kind that make the enterprise framing credible. OutSystems runs a large multi-tenant platform and says a dedicated always-on pod per agent "simply isn't possible" at the number of agents it expects to support; it cites sandboxed runtimes that suspend when idle and resume on demand, plus the APIs, lifecycle management, and gateway controls from kagent and agentgateway. Swissquote runs a sovereign AI stack with full control over inference, infrastructure, and the AI lifecycle; its Codename Diego financial-crime investigation agents and MCP servers are already in production on kagent GA, in what it describes as a multi-jurisdictional regulated environment where "security, resilience and explainability are non-negotiable."

A financial-crime investigation agent running in production at a Swiss bank is a stronger proof point than a demo video. It doesn't prove the platform is right for you — but it does prove someone with real compliance constraints was willing to bet on it.

The open-source foundation

Agent Substrate is co-maintained by Solo and Google, and the open-source community's release of Agent Substrate 1.0 is the milestone behind this announcement. Solo says the project has gathered more than 6,000 GitHub stars and over 100 contributors from more than 40 organizations in six months. That's fast momentum for infrastructure, and it matters: the enterprise release is a supported distribution of an open project, not a closed product. If the open project stalls, the enterprise story weakens with it — but for now, the community numbers point the right way.

The catch, stated plainly

  • No pricing in the announcement. Solo says nothing about what the enterprise releases cost. A supported runtime with air-gapped deployment options is enterprise-priced by definition, and until the numbers are public, you can't compare it against running your own agent infrastructure on plain Kubernetes.
  • Every production claim is Solo's. OutSystems and Swissquote are named customers with direct quotes, which is better than anonymous logos — but the performance and scale figures behind the sizing calculator come from Solo's own testing, not an independent benchmark. Treat the calculator as a starting point, not a guarantee.
  • The suspend-and-resume model is new at this scale. Suspending idle sessions sounds simple; doing it without losing context, breaking a multi-turn conversation, or adding latency on resume is the hard part. The launch customers say it works for them. Your workload is not theirs.
  • This is infrastructure, not a shortcut. Agent Substrate doesn't make your agents smarter. It makes them cheaper to run and harder to misuse at scale. If your agent logic is the bottleneck, this release doesn't fix it.
  • No hands-on testing yet. I haven't deployed it, and neither has anyone outside Solo's launch program. The first independent deployments will tell us more than any announcement can.

The bottom line

Agent Substrate is trying to do for agentic workloads what Kubernetes did for traditional deployments: make the orchestration a non-issue, so teams can focus on the agents themselves. The actor model — a session with a sandbox and a gateway attached, always — is a genuinely good idea for production agents, and the fact that it's co-maintained with Google and already running financial-crime agents at a Swiss bank gives it more weight than the average launch-day announcement.

The honest test is a small one: if you're already running agents on Kubernetes and watching idle pods eat your budget, the open-source Agent Substrate 1.0 is worth a look this week. The enterprise release is for teams that need the support contract, the access controls, and the air-gapped option. Either way, the question to ask isn't whether agents need infrastructure — they do — but whether this is the layer that becomes standard, or just the first of several competing ones.

Are you running agents on Kubernetes yet — and if so, what's eating your budget: idle pods, context, or something else entirely?

Sources: Solo.io via GlobeNewswire, "Solo.io Brings Agent Substrate to Enterprise Production" (https://www.globenewswire.com/news-release/2026/11/10/3399805/0/en/solo-io-brings-agent-substrate-to-enterprise-production.html), 10 November 2026. All capability, customer and community claims are Solo.io's, stated as such. No hands-on testing was performed for this article. Nothing here is sponsored — no affiliate links on this page. This breaking article is not the daily buyer-intent post. Plain links for readers: solo.io, ate.dev, kagent.dev.