Kore.ai's Artemis launch exposes what every major enterprise agent vendor is racing to own: the declarative layer where business intent becomes agent behavior. The lock in risk hides in plain sight.
The enterprise AI platform race has a familiar shape. Microsoft, Salesforce, Google, and ServiceNow are all running the same play, building a layer that turns business intent into agent behavior, and each is betting that owning that layer decides who runs the agent stack of the next decade. The newest entrant, Kore.ai, just made the bet more visible by naming the layer out loud.
VentureBeat reported that on May 21, 2026, the Orlando-based AI company launched Artemis, the second generation of its agent platform, with a claim that enterprise teams can now deploy production-grade multiagent systems in days rather than months. The foundation, per the company's announcement, is a proprietary language called ABL (Agent Blueprint Language): a compiled, declarative, YAML-based format with its own parser, compiler, and runtime. Six orchestration patterns come built in: supervisor, delegation, handoff, fan-out, escalation, and agent-to-agent federation. Around the language, Kore.ai wraps two more pieces, an AI agent architect called Arch that translates plain-language business goals into ABL and continuously tunes them from production traces, and a "Dual-Brain" architecture that runs LLM-based reasoning and deterministic business rules in parallel through shared memory.
According to the Kore.ai press release, Artemis is available first on Microsoft Azure, with broader cloud support to follow. Founder and CEO Raj Koneru framed the launch as the start of the "third wave" of enterprise AI, where governance, observability, and trust decide who scales.
Read past the launch language and the bet looks the same one Microsoft, Salesforce, Google, and ServiceNow are already making, just with different vocabulary. Microsoft is building agent infrastructure into Copilot and Azure AI Foundry. Salesforce is building it into Agentforce, the agent layer wrapped around its Atlas reasoning engine, which Marc Benioff has called a "quantum leap in AI," per Salesforce Ben. ServiceNow is folding agents into its Now Platform. Google is pushing Agentspace and Vertex AI Agent Builder. Each platform claims model agnosticism, governance as a first-class feature, and the ability to compress months of integration work into days. Each is racing to be the place where enterprise intent becomes agent behavior.
The claim itself is old. So is the warning sign inside it.
In every platform war, the first wave of vendors sells neutrality. The second wave builds proprietary layers on top of that neutrality. The third wave is the lock-in. The first phase is happening in public right now. The second and third are what enterprise buyers should be pricing into today's contracts.
Salesforce Ben, a curator outlet that has tracked the agent platform category closely, recently laid out the gap between vendor claims and enterprise readiness in the Agentforce rollout. Its reporting cites a list of reasons for what it calls stunted growth in Agentforce adoption: a lack of readiness for enterprises, the relatively rudimentary output of current LLMs, and a thin catalog of fleshed-out use cases. Salesforce has publicly cited early wins, including roughly 90 percent case resolution at Royal Bank of Canada and OpenTable, but the broader pattern Salesforce Ben describes is that the gap between a vendor's claim that agents are ready and a buyer's experience deploying them under regulated-workload constraints is real and unresolved.
The same gap shows up in independent analyst positioning. Everest Group has placed Kore.ai as a Leader in its Agentic AI Products PEAK Matrix Assessment 2026, with vice president Vaibhav Bansal on the record citing standardized agent building and enforced governance as the differentiators. Leader placement in a single analyst matrix is a real signal. It is also the kind of signal that flatters a vendor trying to establish itself against incumbents an order of magnitude larger.
What is novel about Kore.ai's pitch is not the goal but the language. Most platform vendors are still building their agent definitions in JSON, in tool calls, in proprietary visual workflows, or in some mix. Kore.ai's bet is that the winner of the agent platform war will be the vendor that defines the declarative language enterprises use to describe what an agent is supposed to do. ABL artifacts, the company says, can be stored in GitHub and pushed through CI/CD, which makes them auditable, version-controlled, and (in principle) portable. That is the right instinct. It is also exactly the layer where lock-in gets baked in if the format and runtime stay proprietary.
Two questions separate a portable agent layer from a trap.
The first is whether the language is open or owned. A YAML file is open in the sense that a human can read it. A YAML file plus a parser, compiler, and runtime that only one vendor ships is not open in the sense that a buyer can move. If the language's compiler is the company's compiler, the agent definition is bound to the company's runtime, and the agent definition is the most valuable artifact a customer will own. Buyers who treat ABL (or Salesforce's Atlas definitions, or Microsoft's Copilot Studio flows, or ServiceNow's Now Assist workflows) as if they were their own IP should ask, in writing, whether they can compile the same definitions against a different runtime tomorrow. If the answer is no, the artifact is rented, not owned.
The second is whether governance and observability are features or foundations. The third wave that Koneru named in his launch statement is the right frame. The agents that ship this year will spend most of their operating lives inside a regulated, audited environment. Vendors who treat observability (logs, traces, evaluations, red-team results) as a paid add-on and governance (who can change what, with which approval, against which policy) as a configuration screen are signaling that the agent is the product and the rest is monetization. Buyers who treat both as the floor, not the ceiling, will end up in the vendor that wins on the merits, not the vendor that wins on defaults.
The window for those choices is short. Each new agent platform that ships this year raises the switching cost for the next enterprise to commit. The pattern of platform wars is not that the early leaders always win. It is that the buyers who treat the first wave as fungible and pick the second wave on a written portability contract tend to keep their options open. Kore.ai's Artemis launch is a useful reminder that the field is still plural. That is a fact to use, not a fact to celebrate.