Sarah Pirie-Nally
AI Strategist · Keynote Speaker · Author
Everyone is talking about prompts. But the leaders who are quietly winning with AI are obsessing over something far less glamorous — and far more powerful. They're building the infrastructure that makes AI actually useful. That means understanding two things: RAG and MCP.
If those acronyms make you want to scroll past — stay with me. Because once you understand what they actually do, you'll never think about AI implementation the same way again.
The problem with "off the shelf" AI
Here's what most organisations do: they hand their team access to ChatGPT or Claude, call it an AI strategy, and wonder why nothing sticks six months later.
The reason is simple. General-purpose AI doesn't know your business. It doesn't know your policies, your customer history, your internal processes, or your industry nuances. It's brilliant at general knowledge. It's useless at your knowledge.
That's the gap RAG and MCP are built to close.
RAG: teaching AI what only you know
Retrieval-Augmented Generation — a method that lets AI search and pull from your own documents, databases, or knowledge before it responds.
Think of the base AI model as a brilliant new hire on Day 1 — extraordinarily capable, deeply knowledgeable about the world, but completely unaware of your organisation's specific context. RAG is the induction process. It grounds your AI in your reality: your past projects, your customer feedback library, your brand guidelines — so when they respond, they're drawing on your institutional knowledge, not just general training.
The Wonder Conductor analogy: Hold this: giving your AI a searchable brain built from everything your organisation has ever learned, instead of hallucinating a policy, a result, or a recommendation from thin air.
Where RAG transforms CX:
Imagine building AI that actually knows your product. Imagine it surfacing the right case study, the right contract clause, the right precedent — instantly. Onboarding systems that answer "how do we do this here?" with actual company knowledge rather than generic best practice.
This is the difference between an AI that sounds helpful and one that is helpful.
MCP: giving AI hands, not just a brain
Model Context Protocol — an open standard that lets AI securely connect to and take actions inside your existing tools and systems.
RAG gives AI knowledge. MCP gives it the ability to act.
Without MCP, your AI assistant is like a brilliant advisor who can only give advice through a chat window. With MCP, that advisor can log into your CRM, update a record, send a follow-up email, and file the outcome — all in a single conversation, with your authorisation.
Think of it this way: RAG is your AI's long-term memory. MCP is its ability to reach out and interact with the world. Together, they transform AI from a smart system into an intelligent agent that can actually do things.
What MCP enables in practice:
A customer service prompt that can trigger a recommendation — it executes it. A sales copilot that can look up a client record, draft a proposal, and create a follow-up task without leaving the conversation. An operations tool that can query live costs, surface anomalies, and trigger workflows.
"Most organisations are building AI on the surface layer. The ones building real competitive advantage are going deeper — into the infrastructure."
RAG vs MCP — what's the difference?
| RAG | MCP |
|---|---|
| Retrieves your data and documents | Connects AI to your systems |
| Grounds AI in your specific context | Enables agents to work autonomously |
| Passive: information retrieval model | Active: AI can use tools (CRM, calendar, etc.) |
| Reads, doesn't act | Acts, doesn't just advise |
These aren't competing technologies — they're complementary layers. RAG makes your AI informed. MCP makes it capable. The magic happens when you combine both.
The conscious CX lens
Here's where this gets important for now — think about AI strategy, technology infrastructure only matters if it serves people better.
RAG and MCP aren't just efficiency plays. When implemented with intention, they give the foundations of a truly human-customer experience — one where AI has the context to respond with genuine relevance, and the capability to actually resolve rather than redirect.
A customer who never has to repeat themselves. A team member who gets the right answer immediately. A leader who can use their AI systems because they're grounded in real organisational knowledge, not hallucinated guesses.
That's what this infrastructure enables. Not just faster AI. Better AI. Better — on the human side of every interaction.
What this means for your AI strategy
If you're building an AI rollout — or wondering why your current one isn't delivering — ask yourself:
Does our AI know what only we know? If not, you need RAG.
Can our AI take action in our systems? If not, you need MCP.
Are both in our roadmap? If not, you're building on sand.
The prompt layer matters. But the infrastructure layer is where the real transformation happens.
Ready to build AI that actually knows and serves your business? Wonder Conductor is a 12-week live cohort for leaders with an AI mandate and no time to waste.
Cohort Opens May 2026
Ready to Go From AI-Curious to AI-Confident?
Wonder Conductor is a 12-week live coaching program where you build your AI strategy, automate your systems, and lead with the confidence that comes from knowing exactly what to do next.
Sarah Pirie-Nally
AI Strategist · Keynote Speaker · Author · Founder, Wonder & Wander
Sarah helps leaders and organisations harness the power of AI without losing what makes them irreplaceable — their humanity. She has spoken on 6 continents, built the Wonder Conductor program, and runs fortnightly Practical AI masterclasses attended by 550+ leaders.
