For decades, software has been defined by its interface. Buttons, dashboards, menus — all designed to help us navigate systems and accomplish tasks. But that paradigm is beginning to dissolve. We’re not just adding AI to apps anymore — AI is becoming the app.
After recently building task-oriented agents in Microsoft Copilot Studio using the Dataverse MCP connector, it became clear that the long-promised shift — from users navigating software to software responding to users — is no longer theoretical. It’s here now.
We are rapidly approaching a time when we won’t need to open traditional business applications like Dynamics for everyday operations. And this changes everything.
⸻
From Clicks to Conversations: A New Interaction Model
Using business software today still means navigating a user interface: log in, find the right screen, enter data, hit save. But AI agents promise a different future. One where users simply express intent, and the agent handles the rest.
When I built a Copilot Studio agent connected to our Dynamics environment, I was able to:
- Add and update records across multiple entities
- Retrieve authoritative content from our knowledgebase
- Trigger tasks without ever touching the UI
The result? I could interact with Dynamics as a set of capabilities, not as an interface. This wasn’t just a time-saver. It felt like a shift in what software fundamentally is.
⸻
What Is Model Context Protocol (MCP)?
For those unfamiliar, Model Context Protocol (MCP) is an opensource framework that allows agents (like Copilot) to interact with enterprise data models in a consistent, secure, and semantically rich way. Rather than relying on brittle, app-specific logic, MCP defines how an AI can:
- Understand the structure of your data (e.g. tables, fields, relationships)
- Navigate business rules and workflows
- Interpret actions in context — e.g. “create a new customer record with this info” — and translate them into backend operations
Think of it as a bridge between natural language and structured systems. It’s not a single connector — it’s the context layer that lets agents act intelligently within enterprise environments like the Microsoft Dataverse.
⸻
From Applications to Orchestrations
As MCP matures and AI agents become more capable, we start to see a deeper shift:
- Users don’t open apps; they issue requests.
- Workflows aren’t built in interfaces; they’re executed via intent.
- The agent becomes the UI — orchestrating across systems on your behalf.
This effectively unbundles the app model. Rather than siloed systems with their own front ends, businesses will increasingly expose their services and data via clean, agent-friendly APIs and schemas. The best-designed product will no longer be the one with the most intuitive UI — but the one that’s easiest for an AI to operate reliably.
⸻
What Changes for Teams and Systems
From my experience and what’s now clearly emerging, this shift has several meaningful implications:
1. Less User Training, More Empowerment
Agents reduce the need for staff to learn the intricacies of business systems. New users can complete tasks with prompts instead of process manuals.
2. Consistency Across Roles
The same agent can serve different teams — sales, finance, support — with a consistent interaction model, tailored only by permissions and data visibility.
3. UX Investment Shifts to API Maturity
Great front-end design still matters, but the investment priority moves to robust, predictable APIs and semantic models that MCP can reason over.
4. Faster Iteration
Agents can be updated and re-trained faster than UIs can be redesigned. That creates a more agile, responsive relationship between business needs and technology delivery.
⸻
My MCP experience wasn't perfect
In my build, there were limitations — particularly around multi-step logic, contextual chaining, and fallback handling. But none of these were blockers. The experience was robust enough to deliver real value, and the rate of improvement from Microsoft suggests we’re just scratching the surface.
MCP doesn’t just make AI integration easier. It lays the foundation for intent-driven systems — where business logic, data integrity, and user safety are built into how the AI understands and performs tasks.
⸻
Final Thought
What struck me after building with Copilot Studio and MCP wasn’t just how much time I saved — it was the realisation that the long-held vision of AI as the interface layer is finally being realised in tools we can use right now.
We’ve been talking about AI agents and natural language interfaces for years. But for the first time, it’s not speculative. I didn’t need to imagine a world where software responds to intent instead of clicks — I built a working example of it.
This isn’t a future-tense conversation anymore. It’s happening.
And as Model Context Protocol matures, it’s easy to see how business applications will increasingly fade into the background. The UI becomes optional. The AI becomes your operating layer. And the job of building software becomes more about enabling intelligent orchestration than designing visual pathways.
AI is no longer just a feature in our tools. It’s becoming the way we use them.
Let's start
creating together.
We're always up for a chat, talk to us about your next project.
About the author
Rowan Schaaf
Rowan heads up client engagement and strategy at Pattern. With over three decades of experience in the technology sector, he has worked with a range of organisations from startups to some of the world's biggest brands.