As AI agents become central to how organisations automate tasks, assist employees, and streamline processes, one challenge consistently slows teams down: testing how an agent actually behaves in a real user interface. 

Until recently, early-stage agents lived in prototypes, isolated sandboxes, or technical testing environments. That meant product teams were making critical design decisions such as interaction flows, prompts, actions and UI patterns without ever seeing how users would experience the agent inside the tools they rely on every day. 

Microsoft is changing that. 

With the new capability to publish AI agents built in Microsoft Foundry directly into Microsoft 365, organisations can now experiment, validate, and iterate on an agent’s UI in the live environment where it will eventually operate. 

This is not just a convenience upgrade. 
It’s a major leap forward in AI product design, usability testing, and experience refinement. 

Why UI Experimentation Matters for AI Agents 

An AI agent isn’t just a back-end flow or a reasoning model. 
Its value is determined by how effectively users can interact with it. 

UI and experience flaws often emerge only when an agent is placed into a live workflow: 

• Does the user know what the agent can do? 

• Are the prompts intuitive? 

• Does the response appear in the right place? 

• Is the agent interruptive, helpful, or confusing? 

• Are actions easy to follow? 

• Do users trust the responses enough to act on them? 

These are not questions a prototype or isolated test environment can answer. 

They require real users in real applications. 

From Prototype to Reality: What Foundry → M365 Publishing Enables 

Publishing an agent from Microsoft Foundry directly into Microsoft 365 unlocks the ability to perform true UI and UX experimentation, including: 

1. Real-world interaction testing 

Agents can be surfaced inside: 

• Microsoft Teams 

• Outlook 

• SharePoint 

• M365 Copilot experiences 

This means teams can observe actual user behaviour, not simulated intent. 

You see how users click, what they expect, what confuses them, and how they naturally attempt to interact with the agent. 

2. Rapid iteration based on immediate feedback 

Because the agent is live in the M365 interface: 

• UX issues become visible instantly 

• Product teams can refine prompts and flows quickly 

• Updated versions can be republished without heavy engineering 

• Changes are validated with real usage data 

This turns AI experience design into an agile, iterative process. 

3. Testing with controlled user groups 

By publishing demo agents to specific M365 audiences (e.g., a pilot team or early adopters), organisations can: 

• Observe usage safely 

• Prevent premature rollouts 

• Gather targeted feedback 

• Keep governance strict 

• Validate workflow suitability 

This is essential for regulated industries where safety, compliance, and oversight are non-negotiable. 

4. Understanding agent behaviour across different UX contexts 

AI agents behave differently depending on where they appear: 

• A message extension in Teams 

• A Copilot plugin inside Outlook 

• A sidebar interaction in SharePoint 

• A structured workflow in a line-of-business app 

Publishing into M365 environments allows teams to compare and refine experiences across these contexts, helping determine which UI surface delivers the best outcomes. 

Why This Matters for Product, UX, and AI Teams 

This capability bridges the biggest gap in AI agent development: 

The gap between how designers expect users to interact with an agent and how users actually interact with it. 

Foundry → M365 publishing enables teams to: 

• Reduce guesswork 

• Validate UI assumptions 

• Catch UX issues early 

• Test at low cost and low risk 

• Build agents that people will actually use 

Most importantly, it brings AI experimentation closer to real work, ensuring agents are shaped by genuine human behaviour, not theoretical design. 

Governance & Control Remain Central 

Since publishing is handled through Microsoft 365: 

• Identity and access stay governed by Entra ID 

• Access can be limited to pilot groups 

• Telemetry and usage insights feed back to admins 

• Data stays within organisation boundaries 

• Compliance teams get visibility and reassurance 

Innovation doesn’t require compromising safety or oversight. 

A New Era of AI Experience Design 

Microsoft’s approach signals a shift in how AI agents will be developed going forward. 

Rather than building agents in isolation and hoping they work in the final UI, organisations can now: 

• Experiment early 

• Validate with real people 

• Design AI experiences that feel natural 

• Iterate fast 

• Avoid costly redesign later 

Agent publishing from Foundry into Microsoft 365 makes UI experimentation and UX refinement a first-class part of AI development. 

It’s not just about building smarter agents, it’s about building agents people love to use.