Guest: Randy Bias (LinkedIn)
Company: Mirantis
Show Name: An Eye on AI
Topic: AI Infrastructure

Everyone assumes enterprise AI has reached the running phase. The reality? Most organizations are still crawling—struggling with hallucinations, PII leakage, and evaluation frameworks while their competitors race ahead. Randy Bias, VP of Strategy & Technology at Mirantis, delivers an essential reality check: AI transformation mirrors cloud adoption, complete with trial-and-error learning curves and the need for community-driven best practices.

The Cloud Adoption Playbook Repeats Itself

Bias draws direct parallels to the early cloud era when organizations debated all-public versus all-private strategies. The answer then, as now, came down to using the right tool for the job. Banks moved email to Microsoft but kept financial transactions on-premises. The pattern holds for AI—ChatGPT’s consumer success obscures how early enterprise deployments truly are.

“People don’t see, don’t really understand that it is early, early, early days,” Bias explains. While B2C AI grabbed headlines, B2B value extraction remains nascent. The real enterprise opportunity isn’t helping developers write code faster—it’s applying AI to differentiated data assets and core business processes.

Financial services firms, healthcare organizations, and construction companies each have unique identities and competitive advantages. If they’re not applying AI to those specific domains, they’re not getting ahead of competitors. Yet most enterprises are still experimenting with basic RAG implementations and chatbots, barely nudging into the “walk” phase of maturity.

The Gaps No One Wants to Discuss

Bias doesn’t sugarcoat the challenges. How do enterprises guarantee against hallucinations? When should they deploy on-premises open-source LLMs versus cloud-based models? How do they measure agentic workflow performance when traditional A/B testing frameworks fall short?

The hardest problems involve data hygiene, labeling strategies, and architectural decisions around fine-tuning versus RAG deployment. Should specialized use cases leverage small vision models for CCTV analysis instead of large language models? These questions have no templated answers—only trial-and-error experimentation.

“People always want to find some shortcut, but the reality is that it’s always been trial and error,” Bias notes. The value comes from community participation where enterprises share what’s working and what isn’t. The newly launched Agentic AI Foundation creates exactly this collaborative environment where financial services, healthcare, and other regulated industries can trade learnings while cooperating on open-source ecosystems.

Mirantis Evolution: From OpenStack to Agent Infrastructure

Mirantis built its reputation on open-source infrastructure, progressing from OpenStack to Kubernetes. Now the company is moving up the stack toward agent infrastructure—but with characteristic caution. While competitors rush products to market, Mirantis is leading with services, repeating the successful pattern from their OpenStack and Kubernetes journeys.

The company is deploying Model Context Protocol (MCP) capability across its entire product portfolio. Their focus targets operational leverage, not software development—an underserved market where agents can transform how operators handle firefighting, root cause analysis, and production troubleshooting.

“Operators are overworked, spending too much time firefighting and collecting data,” Bias observes. Agents can handle correlation and analysis work, freeing operators to focus on business process expertise while orchestrating automated systems. This operational agent deployment remains nascent compared to development use cases.

Mirantis announced MCP AdaptiveOps services built on over a year of internal learning and customer deployments. The company developed an agentic maturity model and open-source blueprints that help enterprises deploy agents and MCP servers securely in production with high confidence levels. Additional announcements around these services will drop over coming months.

Why This Matters Now

Enterprises face a fundamental gap: understanding where AI delivers competitive advantage versus where it creates operational overhead. They’re still identifying workloads, deployment patterns, and security frameworks. Without shared best practices and community learning, each organization reinvents solutions independently.

The organizations that succeed will combine internal experimentation with external collaboration. They’ll participate in foundations, contribute to open-source ecosystems, and share learnings while competing on execution. Mirantis’ services-first approach recognizes this reality—enterprises need proven blueprints and expert guidance, not premature productization.

As Bias concludes, infrastructure for agents represents the next evolution point. But rushing products to market ignores how much organizations still need to learn about evaluation, data governance, and architectural patterns. The companies building alongside their customers, gathering real-world deployment experience, will define what agent infrastructure actually requires.