Intent, Skills, and MCP: A Practical Architecture for the Next Generation of Software Engineering

Intent, Skills, and MCP: A Practical Architecture for the Next Generation of Software Engineering

Introduction

In recent months, a new conversation has started to emerge in the AI engineering community: how should we structure systems built around AI agents?

A recent article from The New Stack explored this question through the lens of two major components now appearing in modern AI systems: skills and the Model Context Protocol (MCP). These technologies help AI systems reason about tasks and interact with real-world tools.

But when we look at how software systems are actually designed and built in the enterprise, something becomes clear: there is another layer that sits above both of these.

That layer is intent.

Understanding how intent, skills, and MCP fit together may help clarify how AI-assisted engineering is evolving.

What Are Skills and MCP?

In the emerging agent architecture model, two technical concepts are receiving a lot of attention.

Skills

A skill describes how an AI system should approach solving a particular type of problem.

Skills typically include:

• structured reasoning steps

• reusable instructions

• domain-specific workflows

• engineering patterns

In practice, skills often appear as structured prompt files or reusable instructions that guide the AI through tasks such as:

• analyzing a codebase

• designing an API

• generating tests

• performing refactoring

Skills provide procedural knowledge — they help the system understand how to approach a problem.

Model Context Protocol (MCP)

The Model Context Protocol (MCP) is a standard designed to allow AI systems to interact with external tools and data sources.

Through MCP, an AI agent can:

• read files

• query databases

• call APIs

• execute commands

• access external systems

In other words, MCP allows AI systems to move beyond pure reasoning and interact with the real environment.

If skills represent how the system thinks, MCP represents what the system can access or do.

The Architecture That Is Emerging

When these components are combined, a common architecture begins to appear.

At a high level, many AI agent systems are structured like this:

Agent

   ↓

Skills

   ↓

MCP Tools

   ↓

External Systems

In this model:

• Agents coordinate reasoning and decision-making

• Skills guide the agent through structured problem solving

• MCP tools provide access to real-world systems and data

This architecture helps organize the interaction between reasoning and execution.

However, when we look at how software systems are actually designed in practice, there is an important step that happens before any of this begins.

The Missing Layer: Intent

Before engineers write code, choose tools, or implement architecture, they usually answer a more fundamental question:

What are we trying to build?

This question defines the intent of the system.

Intent can include:

• the purpose of the system

• architectural constraints

• domain rules

• performance or reliability goals

• integration boundaries

For example, an intent might look like this:

Build a fault-tolerant integration service between two enterprise systems using an event-driven architecture, including retry handling and dead-letter recovery.

Once intent is clear, engineers can then apply:

• design patterns

• implementation steps

• system integrations

This mirrors the emerging AI architecture layers quite naturally.

A Practical Way to Think About the Stack

If we combine traditional architectural thinking with the emerging agent architecture model, we arrive at a simple layered view:

Intent

   ↓

Skills

   ↓

MCP Tools

   ↓

Systems

Each layer serves a different purpose.

Intent

Defines the goal and architectural direction.

Skills

Provide structured reasoning or workflows for solving problems.

MCP Tools

Allow the system to interact with real environments, APIs, and data.

Systems

The actual software or infrastructure being built or operated.

This layered approach helps clarify how AI-assisted development can remain aligned with traditional engineering practices.

Why This Matters for Enterprise Engineering

Enterprise software development has always relied heavily on architectural intent.

Architects and senior engineers typically begin by defining:

• system boundaries

• architectural patterns

• reliability requirements

• integration strategies

AI tools can accelerate development dramatically, but without clear intent, they may produce solutions that are technically correct yet misaligned with the broader system design.

By defining intent explicitly, teams can help ensure that AI-assisted workflows remain aligned with:

• enterprise architecture

• domain requirements

• operational constraints

In this sense, AI systems become assistants to architectural thinking, rather than replacements for it.

Key Takeaways

• Modern AI agent systems are increasingly built around skills and tool access frameworks such as MCP.

• Skills provide structured reasoning workflows for solving tasks.

• MCP allows AI systems to interact with external tools, APIs, and data sources.

• However, traditional software engineering begins with intent, which defines the architectural goal of a system.

• A useful way to think about emerging AI-assisted engineering is:

Intent → Skills → MCP Tools → Systems

As AI tools continue to evolve, maintaining clarity around intent and architecture may become one of the most important ways engineers guide these systems toward reliable and maintainable outcomes.