A+E Global Media boosts AIOps with deterministic AI

What once needed extensive scripting and fine-tuning has become more precise and easier to use after recent updates to Kubiya's agentic AIOps tools.

Agentic AIOps is still evolving, but for A+E Global Media's platform team, new deterministic AI workflows marked significant steps forward.

A+E Global Media, previously known as A+E Networks, is a joint venture of Disney Entertainment Television and Hearst that includes media brands such as A&E, Lifetime and The History Channel. Last year, the platform team at A+E began experimenting with Slack-based virtual agents built with tools from Kubiya to automate incident management communications.

This was a step back from the original goal of automating incident resolution and still involved upfront toil, according to Neeraj Mendiratta, vice president of DevSecOps platform architecture and engineering services at A+E.

"When we were doing incident management, we would say, 'For this incident, give me this information,' but I would get a lot more information than I needed," Mendiratta said. "Then we would write a Python script to say, 'Limit it to incidents that happened in the last week' or 'since this morning' or 'today.' Those kinds of tweaks had to be done with those agents to make it more [accurate]."

Still, with help from a dedicated Kubiya infrastructure engineer -- a service the vendor added earlier this year -- Kubiya AI agents reached production at A+E for failure analysis in its Azure DevOps environment. What was previously a 45-minute manual process for DevOps engineers to analyze GitHub, code pipeline and other log files to troubleshoot failures and set up a bridge in Slack is now performed automatically by Kubiya agents.

"[The system] figures out where the error is, and then it will post a summary on Slack to say, 'Hey, this build failed, the code was checked in by this person at this time, and here is the error that it encountered, that maps to this line in the code,'" Mendiratta said.

Last month, Kubiya launched updated AI agents that require less labor-intensive setup from end users or professional services engineers. Kubiya Composer, also launched in June, features a new visual editor for creating agentic AI workflows that replaces the scripting Mendiratta described.

Kubiya's making agents more deterministic to give you concise information, which is more actionable.
Neeraj Mendiratta,Vice president, A+E

With this release, Kubiya also added deterministic AI agents with a directed acyclic graph (DAG) and workflow engine that create structured execution plans and generate infrastructure code in response to natural-language prompts. Without this execution plan, inherently probabilistic AI agents could devise a plan that breaks the infrastructure or doesn't work. The user can review a proposed execution plan and code changes before deployment, and Kubiya provides built-in LLM-driven verification mechanisms both pre- and post-deployment.

"When something is failing, don't give me a lot of information about, 'Oh, I see these 10 builds failing, and this is what happened five days ago," Mendiratta said. "Tell me exactly what happened just now. That's where Kubiya's making agents more deterministic to give you concise information, which is more actionable."

Kubiya deterministic AI targets infrastructure ops

Kubiya isn't unique in embracing a DAG to create deterministic AI workflows. Google's Agent Development Kit supports deterministic workflows, as does Amazon Nova through an integration with LangGraph. But while most early agentic AI adopters, such as Intuit, will use a mix of deterministic and probabilistic workflows, AIOps and infrastructure automation demand emphasis on deterministic AI, according to Amit Govrin, founder and CEO at Kubiya.

"It depends on who it's intended for, who the end user is," Govrin said. "For anything marketing or sales related, AI should be free-flowing and dynamic, because in the worst-case scenario, you updated the wrong Salesforce record. But if we're talking about DevOps, platform engineering, security operations, and infrastructure operations, you can't go wrong there."

Thus, Kubiya's ops-centric approach differs from tools such as Amazon Nova, which are built for app developers, Govrin said.

"Nova is a code-first framework that's great for stitching together multistep, multmodal agents, if you're OK wiring up the logic, flows and guardrails yourself," he said. "Kubiya, on the other hand, abstracts that layer entirely.

"We use DAGs too, but they're generated dynamically based on goals and optimized for production-grade infrastructure ops," he said. "Our DAGs don't just represent execution order, they encode everything from identity to environment state and policy logic for teams that need reliable, auditable, and repeatable automation without hand-authoring every edge case."

Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.

Dig Deeper on Systems automation and orchestration