Guest: Randy Bias (LinkedIn)
Company: Mirantis
Show Name: 2026 Predictions
Topic: AI Infrastructure

While the industry has obsessed over AI-assisted coding, a critical gap remains: AI agents in operations. Randy Bias, VP of Strategy & Technology at Mirantis, believes 2026 will be the year this changes—when the Model Context Protocol (MCP) dominates, general-purpose agents replace custom solutions, and enterprises finally rethink their workflows instead of just optimizing old ones.

The Shift from AI-Assisted to AI-Driven Operations

Mirantis has been on a clear trajectory: from virtualization infrastructure to container infrastructure, and now to AI infrastructure. But as Randy explains, the focus on AI operations hasn’t matched the attention paid to software development. “There’s very little work being done. I mean, it’s not none—there are people absolutely working on it—but it’s not at the level it needs to be. It’s nowhere near the level of attention being paid to software development,” he says.

This gap is becoming urgent as Mirantis works on large AI infrastructure deals. The question they’re asking themselves is: “How do we automate everything, top to bottom? How can the AI factory manage itself for the most part?” The answer involves applying AI agents at every layer—networking, storage, bare metal, virtualization, and cluster management.

MCP Will Dominate

Randy is convinced that 2026 will be the year MCP breaks out. Last year laid the foundation, but this year it becomes the de facto standard. “I’m convinced that this is the year it becomes a done deal. I don’t think there will really be any other protocols for most use cases for agents; they’ll just keep extending MCP,” Randy predicts.

The protocol’s recent addition of “tasks”—capabilities that allow long-running operations and asynchronous check-ins—enables entirely new workflows. Combined with the ability to run custom code, MCP servers become dynamic platforms that agents can extend on their own.

General-Purpose Agents Over Custom Solutions

The era of building custom agents for every use case is ending. Randy sees general-purpose agents like Claude, Codex, Gemini, and Goose being adapted for domain-specific purposes by combining skills and tools in the form of MCP servers. “It seems like these general purpose agents can basically be put to domain specific purposes by combining skills and tools in the form of MCP servers to basically attack domain specific problems,” he explains.

This approach allows organizations to ride the wave of innovation happening in those general-purpose platforms, with 80-90% of most agentic use cases solvable this way. More importantly, agents will begin writing their own code and pushing it to MCP servers for execution—a capability Randy demonstrated with his proof of concept called “nightcryer,” a triage agent that is automatically dispatched when production problems occur.

Enterprise Adoption: The Real Bottleneck

The technology is ready, but enterprises face the typical adoption curve challenges. Randy recently experienced this firsthand when answering a cybersecurity questionnaire from a major financial services company about an MCP server Mirantis provides.

“The cybersecurity team in question really understood MCP. They understood the issues. They understood AI—it was very, very good questioning—but you could tell the team was thinking, ‘Okay, this is a whole new attack surface. How do we take care of it?’”

Beyond security concerns, there’s a learning curve. Just as enterprises initially tried to run inappropriate workloads on AWS, they’re now learning which problems AI agents actually solve. “AI agents aren’t the best solution for every problem. They’re appropriate for what they’re good at, which is fuzzy logic,” Randy notes. Success requires looking at AI agents through a CEO lens: what is the business impact of applying agents to this specific use case?

The Workflow Revolution

Randy’s most critical advice for enterprise leaders: get your teams using the technology now. That same financial services company has a top-down initiative requiring every group to work with AI agents. But it’s not about adding agents to existing processes—it’s about creating entirely new workflows.

“It’s not about taking your old workflows and just jamming an agent in there. It’s about the new kinds of workflows that are possible because you’ve got AI agents in the process,” Randy emphasizes.

He points to workflow transformations happening across development, operations, marketing, and customer support as evidence that the real value comes from rethinking how work gets done, not just optimizing existing approaches.

Mirantis’ Two-Pronged Strategy

For Mirantis, 2026 priorities are clear. First, deliver high-quality AI infrastructure, especially for sovereign AI clouds, where capacity demand continues to exceed supply. Second, focus on the practical application of agents to real-world business problems—knocking down low-hanging fruit to build proof points and create the cultural DNA needed to cross the AI chasm.

As Randy concludes, the technology is here. The protocol is maturing. The question is whether enterprises will move fast enough to capture the opportunity.