The 4 Pillars: Persona, Skills, RAG, MCP

“Should I put this in RAG, a Skill, or the Persona?”

Every engineer building agents hits this wall. You have domain knowledge—a PDF, a database, a rule—and you don’t know where it belongs.

Get it wrong, and you get Context Overflow (expensive, slow agents) or Context Amnesia (hallucinations).

The Problem

Most developers treat the LLM context window like a junk drawer. They stuff strict rules, messy docs, and JSON schemas into one massive system prompt.

This is the “Swiss Army Knife” trap. It works for a demo, but in production, it fails because: -> Cognitive Load.

Just as humans struggle to multitask, LLMs degrade when instructions conflict. We need an architecture that separates concerns.

The Concept

There are four distinct pillars of agent context. Each solves a specific problem.

Validated by Google’s Framework

This structure mirrors the cognitive architecture defined in Google’s Context Engineering guide:

Pillar 1: Persona 🎭

Purpose: Define WHO the agent IS. When: Always present (System Prompt).

Recent research on Role Prompting shows that assigning a specific persona (e.g., “You are a Senior Security Engineer”) significantly improves reasoning capabilities, sometimes by over 20%.

The Mistake: Using Persona for mechanics.

• ❌ “You are an agent that outputs JSON with keys x, y, z…”
• ✅ “You are a pragmatist who values working code over theoretical purity.”

Governance Rule: The Persona defines the values the agent uses to make trade-offs.

Pillar 2: Skills 📚

Purpose: Teach HOW to do things. When: Loaded on demand (Tool Definitions).

Skills are procedural knowledge. If Persona is the “character,” Skills are the “script.” In modern terms, these are Tools or Functions that the model can call.

The Mistake: Hardcoding steps in the System Prompt. The Fix: Encapsulate logic in a tool. instead of telling the detailed steps of “How to valid email”, just give the agent a validate_email() tool.

Why? It moves complexity from probabilistic tokens (the LLM guessing) to deterministic code (the function executing).

Pillar 3: RAG 📖

Purpose: Access WHAT to know—facts and documents. When: Retrieved at query time.

Patrick Lewis et al. introduced RAG in 2020 to solve the “knowledge cutoff” problem.

The Enterprise Litmus Test for RAG: If the information changes faster than your deployment cycle, it implies RAG.

• Company Policies? RAG.
• Yesterday’s Sales Data? RAG.
• Java Syntax? Training Data (Model).

Pillar 4: MCP 🔌

Purpose: Connect to external ACTIONS. When: Invoked to change the world.

The Model Context Protocol (MCP) is the new standard for connecting AI models to data sources. It’s the “USB-C” for agents.

Why it matters: Before MCP, every agent needed custom glue code to talk to GitHub, Slack, or Postgres. With MCP, you write the connector once, and any agent can use it.

The Decision Framework

How do you decide? Use the Time-Horizon Heuristic:

Industry Applications

The 4 Pillars apply universally. Here’s how they map across domains:

Quick Examples

🏦 Banking: A loan officer agent uses Persona for risk tolerance, Skills for underwriting steps, RAG for current rate policies, and MCP to pull credit reports.

🛒 Retail: A shopping assistant uses Persona for brand voice, Skills for return handling, RAG for product details, and MCP to check inventory in real-time.

🎓 Education: A tutoring agent uses Persona for teaching style, Skills for pedagogical methods, RAG for course content, and MCP to update gradebook.

Key Takeaways

• ✅ Don’t clutter context: Use the right pillar to keep the “reasoning brain” clear.
• ✅ Persona is for values: Use it to guide decisions, not just format output.
• ✅ Skills are deterministic: Move complex logic out of prompts and into code.
• ✅ Standardize with MCP: Don’t build custom integrations if an open standard exists.

What’s Next

• 📖 Previous article: The Orchestra: Why Multi-Agent AI Works
• 📖 Next article: Skills: Progressive Context Disclosure — Escape the “Prompt Blob Monster” with on-demand procedural knowledge.
• 💬 Discuss: Which pillar is the biggest bottleneck in your current agents?

References

1. Google Cloud Research — Context Engineering: Sessions & Memory (2025). Defines the distinction between Procedural Memory (Skills) and Semantic Memory (RAG) in agentic architectures.
2. Anthropic — Prompt Engineering Guidelines. Source for Role Prompting effectiveness.
3. Lewis et al. — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (NeurIPS 2020).