Guest: Randy Bias (LinkedIn)
Company: Mirantis
Show Name: The Agentic Enterprise
Topic: AI Infrastructure

Are enterprises wasting resources building custom AI agents when they don’t need to? Randy Bias, VP of Strategy & Technology at Mirantis, argues that the industry is approaching agent development the wrong way. Instead of creating custom agents from scratch, organizations should focus on equipping general-purpose agents with domain-specific skills and tools through Model Context Protocol (MCP) servers.

The Custom Agent Problem

The AI agent landscape is cluttered with custom builds. Development teams across enterprises are spinning up proprietary agents for specific tasks, believing customization is the path to effectiveness. But Bias questions this entire approach.

“I really question whether you need custom agents,” Bias states plainly. The evidence supporting his skepticism comes from major players in the AI space. Anthropic, the company behind Claude, is suggesting that general-purpose agents may be sufficient for most enterprise use cases—if they’re given the right context and capabilities.

Domain Expertise Over Custom Code

The alternative approach centers on a simple but powerful concept: transform general-purpose agents into domain experts through skills and tools rather than custom development.

“You take domain skills and give domain tools, so that would be MCP servers that can do specific things for you in the domain of the problem you’re trying to solve, and then suddenly you take a general purpose agent, and you turn it into an expert in a given domain,” Bias explains.

This methodology leverages existing general-purpose AI agents like Claude Code, Codex, or Goose. These agents already possess broad capabilities. What they need is domain-specific context—the specialized knowledge and tooling that makes them effective for particular business problems.

Claude Code now supports domain skills natively, and OpenAI has introduced experimental support for similar capabilities. Bias expects this pattern to spread across the industry, becoming the standard approach rather than the exception.

The MCP Server Factory Model

Mirantis’ MCP Server Factory offering operationalizes this philosophy. Rather than building custom agents, the three-to-six-week engagement focuses on creating the infrastructure that makes general-purpose agents domain-capable.

The factory approach centers on three components: personas, skills, and tools. Domain experts work to define the specific skills needed for their problem space. Those skills are packaged alongside domain-specific tools—the MCP servers that provide access to systems, data, and processes.

The result is a library of capabilities that can be mixed and matched. When a specific task arises, a general-purpose agent is launched with the appropriate persona, equipped with relevant skills from the library, and given access to necessary tools. The agent operates in a sandboxed environment, working on that specific task until completion.

Why This Matters for Enterprises

The implications for enterprise AI strategy are significant. Custom agent development is resource-intensive, requiring specialized expertise and ongoing maintenance. Each custom agent becomes a unique codebase to manage, update, and troubleshoot.

The domain skills approach offers a more sustainable model. Skills and tools can be reused across different scenarios. New capabilities can be added to the library incrementally. General-purpose agents benefit from ongoing improvements by their creators—Anthropic, OpenAI, and others—without requiring internal development effort.

This architectural shift also addresses a practical concern: most organizations don’t have the talent or resources to build and maintain multiple custom agents. By focusing on domain expertise rather than agent development, they can leverage their actual competitive advantage—deep understanding of their business problems—without requiring cutting-edge AI engineering capabilities.

The MCP Server Factory model suggests a future where enterprise AI adoption accelerates not through custom development but through smart composition of general-purpose capabilities with domain-specific knowledge.