Kore.ai Launches Autoloop

Kore.ai has launched Autoloop, the optimization engine of the Kore.ai Agent Platform, Artemis edition. with it, companies set the goals and Autoloop builds the agents, measures them against those goals, and keeps optimizing them automatically, from first draft through production.

"Every enterprise knows what it wants from its agents: finish the job, follow the rules, stay safe, and do it at a sensible cost," said Raj Koneru, founder and CEO of Kore.ai, in a statement. "With Autoloop, your agents keep improving against the goals you set. The companies that scale AI will be the ones using AI to build, govern, and optimize AI."

<>Autoloop starts from the following goals:

• Task completion: the agent finishes what the user came to do, handing off to a human only when it should.
• Accuracy and grounding: every answer is correct and backed by company data.
• Business-rule adherence: complies with policies, eligibility checks, and limits on every interaction.
• Guardrails and safety: no data exposure, off-policy actions, or unsafe responses.
• Consistency: consistent behavior across phrasings, languages, channels, and edge cases.
• End-user experience: fewer turns, less repetition, and lower latency across chat and voice.
• Cost efficiency: the same outcome with fewer, faster, and more affordable model calls and tokens.

Against these goals, Autoloop runs one continuous loop spanning build, evaluate, diagnose, optimize, and re-verify. Before launch, it builds the agent and its test coverage from companies' own operating procedures and iterates until the agent meets every goal. In production, real interactions start new optimization cycles. Every change is scored against all goals at once, so a gain on one, such as lower token spend, cannot quietly hinder another, such as task completion or safety.

Automatic optimization requires complete visibility into agent activity and control over how they change. Two Kore.ai innovations give Autoloop these capabilities:

• StateTrace delivers the visibility, evaluating agents against their full production execution context. It traces every handoff, delegation, state, tool call, and piece of context across the agent network, so Autoloop knows where and why a goal was missed. A patent-pending five-layer validation architecture makes most checks deterministic, which keeps continuous optimization affordable at enterprise scale.
• Agent Blueprint Language (ABL) delivers the control by compiling supervision, routing, handoffs, delegation, tools, business rules, and guardrails into an executable state machine. Every step in a trace maps back to the blueprint, so Autoloop changes exactly the part that caused the miss, and nothing else.

"You can't optimize what you can’t see or fix precisely what you can't express precisely," said Prasanna Arikala, chief technology officer and chief product officer of Kore.ai, in a statement. "StateTrace lets Autoloop see exactly what agents did, and ABL allows it to change anything that needs changing. These are the technologies that make automatic optimization a reality."