Beyond the Prompt: Architecting Multi-Agent AI Solutions with Microsoft Copilot & SharePoint with Reshmee Auckloo [MVP]
![Beyond the Prompt: Architecting Multi-Agent AI Solutions with Microsoft Copilot & SharePoint with Reshmee Auckloo [MVP] Beyond the Prompt: Architecting Multi-Agent AI Solutions with Microsoft Copilot & SharePoint with Reshmee Auckloo [MVP]](https://images.podpage.com/tr:w-1200,h-630,cm-pad_resize,bg-blurred_70/https://img.youtube.com/vi/eE9IJp6RARg/maxresdefault.jpg)
![Beyond the Prompt: Architecting Multi-Agent AI Solutions with Microsoft Copilot & SharePoint with Reshmee Auckloo [MVP] Beyond the Prompt: Architecting Multi-Agent AI Solutions with Microsoft Copilot & SharePoint with Reshmee Auckloo [MVP]](https://img.youtube.com/vi/eE9IJp6RARg/maxresdefault.jpg)
Artificial Intelligence is evolving far beyond single prompts and chatbots. In this episode of the M365 FM Podcast, Mirko Peters speaks with Microsoft MVP Reshmee Auckloo about how organizations can design enterprise-grade multi-agent AI solutions using Microsoft Copilot, Copilot Studio, SharePoint, Microsoft Graph, and Azure AI Foundry. Rather than focusing on prompt engineering, the conversation explores the architecture required to build AI systems that are scalable, secure, and trustworthy.
Reshmee explains why successful AI initiatives begin with well-governed data, structured knowledge, and clear information architecture. SharePoint plays a central role as the enterprise knowledge foundation, while Microsoft Graph provides the contextual intelligence that enables AI agents to understand relationships across people, documents, and business processes. The discussion also covers how multiple AI agents can collaborate to solve complex tasks, when orchestration is preferable to a single agent, and how organizations can balance automation with human oversight.
The episode highlights critical topics including AI governance, security boundaries, identity, permissions, compliance, and responsible AI design. Listeners will learn why clean data, metadata, and governance are prerequisites for reliable AI outcomes, and how Copilot Studio enables organizations to build specialized agents that work together without compromising control or security.
Whether you're an IT architect, Microsoft 365 administrator, SharePoint professional, or AI strategist, this episode offers practical insights into designing production-ready AI solutions that move beyond simple prompts toward intelligent, collaborative enterprise systems.
In today's fast-paced technological landscape, you encounter multi-agent AI solutions that revolutionize how businesses operate. These systems consist of multiple agents working together to tackle complex problems, enhancing efficiency and decision-making. By leveraging Microsoft Copilot and SharePoint, you can create effective multi-agent systems that drive productivity.
Consider this: nearly 90% of enterprises are increasing their AI budgets, and approximately 80% have active deployments of AI solutions. Microsoft Copilot and SharePoint contribute significantly to this trend by enabling intelligent automation and improving knowledge management. With these tools, you can unlock organizational knowledge and streamline workflows, ultimately leading to remarkable productivity gains.
Key Takeaways
- Multi-agent AI solutions use multiple agents to solve complex problems, improving efficiency and decision-making.
- Microsoft Copilot and SharePoint enhance collaboration and automation, leading to significant productivity gains.
- Decentralization allows agents to operate independently, contributing to a common goal without a central authority.
- Specialized agents can handle different tasks, leading to faster and more accurate problem resolution.
- SharePoint improves content management and real-time collaboration, making it easier for agents to work together.
- Implementing orchestration patterns helps manage how agents interact, streamlining workflows and enhancing efficiency.
- Security and governance are crucial for multi-agent systems, ensuring safe and ethical operations.
- Using Microsoft technologies can transform your AI projects, boosting productivity and decision-making.
What Are Multi-Agent AI Solutions?
Key Characteristics
Multi-agent AI solutions consist of multiple autonomous agents that collaborate to solve complex problems. These systems are essential in today's technology landscape, as they enhance efficiency and adaptability in various business contexts. Here are some defining characteristics of multi-agent AI solutions:
| Characteristic | Description |
|---|---|
| Decentralization | Autonomous agents operate independently using local data, contributing to a common goal without a central authority. |
| Local Views | Each agent has a limited perspective, accessing only the necessary information to perform its tasks, lacking a global view of the system. |
| Social Interaction and Coordination | Agents are equipped with protocols for negotiation, cooperation, and conflict resolution, enhancing their ability to work together effectively. |
| Proactivity | Agents can anticipate future needs and challenges by observing their environment and acting accordingly. |
| Communication Channels | Agents utilize structured protocols for effective information exchange, ensuring smooth operation within the multi-agent system. |
The role of multiple agents in problem-solving is crucial. Multi-agent systems excel at complex problem-solving by leveraging the collective intelligence of specialized AI agents working in harmony. Each agent operates independently, possesses its own knowledge base and decision-making capabilities, yet collaborates with other agents to achieve shared objectives.
- Agents specialize in specific domains or workflow stages.
- They communicate and share knowledge with one another.
- Agents can pause, resume, and hand off work dynamically.
This division of labor allows for handling complex or interdependent tasks. For example, in a help desk scenario, the Help Desk Manager initiates the process and assigns tasks. The Web Researcher searches for public solutions, while the Documentation Researcher reviews internal runbooks. This collaborative effort leads to faster and more accurate resolutions.
In business contexts, multi-agent AI solutions find applications in various areas, including:
- Automating business processes like workflow management and supply chain optimization.
- Large-scale data analysis and market trend prediction.
- Customer service management through chatbots.
- Marketing automation and customer relationship management.
These solutions provide significant advantages over single-agent systems. They allow for specialization, optimizing agents for specific tasks, leading to better performance in complex scenarios. Additionally, multi-agent systems ensure resilience; if one agent fails, others can continue to operate, preventing total system failure.
Copilot Features for Multi-Agent Solutions
Integration with Microsoft Tools
Microsoft Copilot offers a range of capabilities that enhance the development of multi-agent AI solutions. These features allow you to create systems that are not only efficient but also scalable and adaptable to your business needs. Here’s an overview of some core capabilities of Microsoft Copilot:
| Capability | Description |
|---|---|
| Multi-Agent Orchestration | A central orchestrator agent coordinates multiple specialized AI agents to complete complex tasks. |
| Agent-to-Agent Communication | Agents can delegate work to one another, enhancing collaboration and efficiency. |
| Governance Tools | Tools that ensure compliance and oversight across multi-agent systems. |
With Microsoft Copilot, you can build multi-agent systems that leverage the strengths of each agent. For instance, Copilot Studio enables you to create modular and scalable solutions. This low-code approach makes it accessible for users with varying technical skills. You can automate tasks, potentially reducing manual work by up to 75%.
Microsoft Copilot emphasizes the concept of agentic AI as a system of systems. This means that one agent can delegate tasks to another, utilize existing logic, and operate within a richer enterprise context. This streamlining of processes reduces redundancy and enhances overall efficiency.
The integration of Microsoft Copilot with other Microsoft tools is seamless. You can connect your orchestrator to existing agents built on Microsoft's agentic platform. This allows you to configure connections to Copilot Studio agents, Microsoft Foundry agents, and Microsoft Fabric Data agents. Managing your connected agent ecosystem ensures accurate orchestration and enhances the overall functionality of your multi-agent solutions.
| Feature | Description |
|---|---|
| Multi-Agent Orchestration | Facilitates coordination among various agents across business systems and third-party platforms. |
| Fabric-aware Reasoning | Enhances the ability of agents to operate with a richer context, improving decision-making and task delegation. |
| Agent2Agent (A2A) Interoperability | Allows agents to communicate and delegate tasks to one another, reducing redundancy in capabilities across teams. |
By leveraging these features, you can create enterprise-ready solutions that enhance trust and mitigate bias in decision-making. Continuous monitoring and integration strategies are crucial for reliability in enterprise applications. Microsoft Copilot supports these efforts by providing governance tools that ensure compliance and oversight across your multi-agent systems.
SharePoint for Collaboration

SharePoint plays a vital role in enhancing collaboration within multi-agent AI solutions. By leveraging SharePoint, you can create a robust environment where agents work together seamlessly. Here are some key ways SharePoint enhances collaboration:
Effective Content Management: SharePoint Knowledge Agent manages content efficiently. It automates workflows, reducing manual effort. This agent enriches documents with metadata, making them easier to find and use. It can even answer questions directly from SharePoint content, ensuring you have access to the information you need.
Automated Document Features: SharePoint offers powerful document management features that benefit multi-agent AI projects. For instance, the Knowledge Agent automatically extracts and structures metadata. This process transforms unstructured content into organized, searchable knowledge bases. As a result, your AI agents can access actionable data quickly.
Enhanced Collaboration Tools: SharePoint supports real-time collaboration. You can create and share documents simultaneously, improving teamwork across departments. This feature is crucial for multi-agent systems, where agents often need to work together on shared tasks.
| Feature | Description |
|---|---|
| Intelligent Workflow Automation | Automates routine tasks, allowing employees to focus on more strategic work. |
| Enhanced Content Management | AI Copilots assist in automatic tagging and metadata extraction, streamlining content organization. |
| Real-time Collaboration | Supports simultaneous content creation and sharing, improving teamwork across departments. |
To maintain collaboration and document integrity in SharePoint-based multi-agent AI projects, consider these best practices:
| Strategy | Description |
|---|---|
| Documentation | Keep a living document with agent design, purpose, and changes. |
| Version Control | Use ALM features to export and version agent configurations in a source repository. |
| CI/CD Integration | Automate export/import and testing of agent solutions for safe and reliable updates. |
By following these strategies, you can ensure that your multi-agent AI solutions remain effective and secure. SharePoint not only supports the integration of AI solutions but also enhances their functionality. With its capabilities, you can build a collaborative environment that drives productivity and innovation.
Remember, clear communication is essential. Plan your orchestration carefully to improve modularity and scalability. Regularly review your agent parameters to keep your AI system secure and up to date.
Multi-Agent Orchestration Patterns
Implementing Orchestration
When architecting multi-agent AI solutions, you must consider various orchestration patterns. These patterns help you manage how agents interact and collaborate. Here are some widely used orchestration patterns:
- Supervisor Pattern: This pattern centralizes command and control. A supervising agent manages the operations of other agents, ensuring they work towards a common goal.
- Adaptive Agent Network Pattern: This decentralized approach allows agents to collaborate without a central authority. Agents adapt to changes in their environment, enhancing flexibility.
- Custom Pattern: This pattern offers programmatic flexibility and control. You can design it to meet specific needs, allowing for tailored solutions.
In addition to these patterns, you can explore various orchestration styles, such as:
- Sequential pattern
- Concurrent pattern
- Group chat pattern
- Handoff pattern
- Hierarchical and federated variants
These patterns provide a framework for organizing your agents and their interactions. They help you streamline workflows and improve overall efficiency.
Embedded and Connected Architectures
You can implement multi-agent orchestration using embedded or connected architectures.
Embedded Architecture: In this setup, agents reside within a single system. They share resources and communicate directly, which simplifies coordination. This architecture is ideal for smaller systems where agents can easily interact.
Connected Architecture: This architecture allows agents to operate across different systems. They communicate through APIs or messaging protocols. This flexibility enables you to integrate various tools and services, enhancing the capabilities of your multi-agent solutions.
Both architectures have their advantages. Embedded architectures offer simplicity, while connected architectures provide scalability and interoperability.
MCP-Connected Tools and Their Benefits
MCP-connected tools play a crucial role in orchestrating multi-agent AI systems. They offer several benefits that enhance your orchestration efforts:
| Benefit | Description |
|---|---|
| Tool/Agent Reusability | Wrap once, reuse forever. Any agent or API exposed via MCP can be plugged into different workflows. |
| Separation of Concerns | MCP separates execution from planning and control, simplifying reasoning and debugging. |
| Observability & Debuggability | Every interaction is logged, versioned, and auditable, crucial for compliance and scaling. |
| Scalability | New tools can be added without breaking the existing architecture, enhancing modularity. |
| Interoperability | MCP abstracts differences in language, framework, and protocol, allowing seamless communication. |
By leveraging MCP-connected tools, you can create a more efficient and manageable multi-agent orchestration environment. These tools help you maintain control while allowing agents to work independently.
Embracing these orchestration patterns and architectures will empower you to build robust multi-agent AI solutions that meet your business needs.
Recent Updates in Copilot Studio
Microsoft Copilot Studio has recently introduced several exciting features that enhance the capabilities of multi-agent AI solutions. These updates empower you to create more efficient and effective systems. Here are some of the key features:
| Feature | Description |
|---|---|
| Multi-agent orchestration | Enhances the ability to manage multiple agents working together effectively. |
| Connected experiences | Facilitates seamless integration and interaction between different AI agents and systems. |
| Faster prompt iteration | Improves the speed at which prompts can be tested and refined, aiding in quicker development cycles. |
| ServiceNow and Azure DevOps connector | Enhances agents' understanding of operational queries, leading to more accurate and actionable responses. |
| Evaluation automation APIs | Allows for programmatic evaluations and integration of quality checks into CI/CD workflows. |
| Real-time meeting transcripts for Teams | Supports agents in providing relevant information and tracking decisions during meetings. |
| Model context protocol (MCP) apps | Expands agent capabilities to interact with external business systems beyond just providing information. |
| Additional model support | Introduces new models like Grok 4.1 Fast and GPT-5.4 Instant, offering more options for tuning AI experiences. |
The introduction of Multi-Agent Copilots marks a significant advancement in orchestration capabilities. These copilots allow agents to collaborate more effectively. They can share data and tasks based on their specific expertise. This collaboration leads to improved efficiency and effectiveness in handling complex tasks.
| Feature | Description |
|---|---|
| Collaboration | Agents can work together, sharing data and tasks based on their specific expertise. |
| Task Delegation | Tasks can be divided among agents, improving efficiency and effectiveness in handling complex tasks. |
| Scalability | The architecture supports growth by allowing multiple agents to handle increased workloads. |
| Performance | Enhanced response accuracy and reduced complexity lead to better overall performance. |
| Governance | Simplified management of domain-specific knowledge across various business functions. |
These updates significantly impact multi-agent architectures. They enhance the ability of agents to work together across tools and data sources. Improved integration allows agents to deliver real business value. Faster prompt iteration updates streamline the development process. Enhanced agent evaluations enable teams to validate performance using real scenarios. This ensures that you can confidently deploy agents for critical tasks, meeting quality and responsiveness standards before full-scale implementation.
With these advancements, you can trust that your multi-agent AI solutions will perform better and adapt to your business needs more effectively.
Key Principles for AI Architecture
When you design multi-agent AI solutions, you must adhere to several architectural principles. These principles ensure that your systems are robust, efficient, and capable of adapting to changing needs. Here are some key principles to consider:
- Coordination: Implement patterns like leader election and decentralized consensus protocols. These prevent conflicting actions among agents.
- Adaptation and Learning: Integrate reinforcement learning. This allows agents to adapt based on feedback in dynamic environments.
- Distributed Perception and Decision-Making: Enable agents to act on local observations. This capability enhances collective problem-solving.
To build effective AI architectures, follow these ordered steps:
- Begin with bounded workflows and clear success metrics.
- Define roles, interfaces, and guardrails for agents.
- Test with evaluation scenarios and monitor coordination failures.
- Optimize for cost and reliability.
- Adopt secure-by-design principles.
Scalability and Flexibility
Scalability and flexibility are crucial for the design of multi-agent AI architectures. They allow your systems to grow and adapt as your business needs change. Here’s how these concepts influence your architecture:
| Key Concept | Explanation |
|---|---|
| Scalability Through Distributed Processing | Divide tasks among agents. This enables parallel operation and horizontal scaling of the system. |
| Reusability and Modularity | Design agents as modular components. This allows for easy updates and replacements without disrupting the system. |
| System Architectures | Utilize various architectures (flat, hierarchical, microservice-style) to support different scalability and flexibility needs. |
| Key Benefits | Modularity allows for incremental evolution of AI systems, while scalability supports numerous task-specific agents. |
By focusing on these aspects, you can ensure that your multi-agent AI solutions remain effective as your organization grows.
Security and Governance Considerations
Security and governance are vital for deploying autonomous AI systems. You need to establish frameworks that ensure safe, ethical, and compliant operations. Here are some essential principles to consider:
| Principle | Description |
|---|---|
| Transparency & Explainability | Ensure that AI systems are understandable and their decisions can be explained. |
| Fairness & Non-Discrimination | Promote equitable treatment and prevent bias in AI decision-making. |
| Oversight & Accountability | Establish mechanisms for monitoring and holding systems accountable. |
| Trust & Safety | Focus on building user trust and ensuring safety in AI operations. |
| Privacy & Security | Protect user data and ensure secure operations of AI systems. |
| Robustness & Resilience | Ensure systems can withstand and recover from failures or attacks. |
Effective governance frameworks provide continuous visibility into agent actions and decisions. You should monitor for behavioral anomalies and policy violations. Tracking performance, cost, and security signals is also essential. Implementing input, process, output, and action controls will help limit risks and ensure compliance.
By adhering to these principles, you can architect multi-agent AI solutions that are not only powerful but also secure and trustworthy.
Practical Example: Service Desk Orchestration
Service desk orchestration using multi-agent AI solutions transforms how organizations manage customer support. This approach shifts from human-centric ticket handling to agent-orchestrated support systems. AI agents collaborate to detect intent, sentiment, and risk in real time. They auto-classify and route tickets, enhancing efficiency in customer support.
Overview of the Use Case
Consider a service desk scenario in a large organization. Here, multiple agents work together to streamline ticket management. For instance, organizations like Cleveland Clinic and Mayo Clinic utilize AI virtual assistants for appointment management and FAQ handling. Similarly, Humana has implemented AI support tools in call centers to improve response times and resolution rates. These real-world examples demonstrate the effectiveness of multi-agent orchestration in service desk operations.
Implementation Steps
To deploy a multi-agent AI service desk solution, follow these steps:
| Agent Role | Description |
|---|---|
| Triage Agent | Classifies incoming tickets and routes them to the appropriate specialist. |
| Knowledge Agent | Handles questions from documentation and resolves tickets without escalation. |
| Billing Agent | Manages invoice questions and has specific access to billing systems. |
| Technical Support Agent | Addresses product issues and can query system logs for diagnostics. |
| Response Composer + QA Agent | Composes customer-facing responses and ensures accuracy through a separate QA process. |
These roles ensure that each agent specializes in specific tasks, leading to a more efficient workflow.
Benefits of Multi-Agent Orchestration
Organizations that implement multi-agent orchestration in their service desks report significant benefits:
- Operational Efficiency: Processing complex workflows in parallel leads to faster ticket resolution.
- Cost Reduction: Seamless processes minimize the need for human oversight, resulting in cost savings of 15–30%.
- Improved Decision-Making: Coordinated agents provide comprehensive insights, enhancing overall decision-making.
Many organizations experience up to a 25% improvement in operational efficiency. They also see double-digit productivity growth and measurable reductions in error rates. Orchestrated agents eliminate redundant work and reduce processing time. Automation of multi-step processes lowers operational costs while improving accuracy. Enhanced decision-making occurs through collaborative analysis, breaking down data silos.
In summary, architecting multi-agent AI solutions with Microsoft Copilot and SharePoint offers significant advantages for organizations. You should prioritize security, governance, and structured information architecture when developing these systems. Creating specialized agents for specific business challenges enhances efficiency and effectiveness.
Consider integrating Microsoft technologies, as they play a crucial role in your AI projects. Embrace the architectural patterns and principles discussed to improve your solutions. By leveraging tools like Copilot Studio, you can streamline workflows and foster collaboration among agents.
Explore these tools for your projects and witness the transformation in productivity and decision-making.
FAQ
What is a multi-agent AI solution?
Multi-agent AI solutions comprise multiple autonomous agents collaborating to solve complex problems. These systems increase efficiency by allowing specialized agents to address different aspects of tasks simultaneously.
How does Microsoft Copilot enhance multi-agent systems?
Microsoft Copilot allows seamless integration of various agents, improving collaboration, governance, and task delegation. It enables organizations to build enterprise-ready solutions that automate workflows effectively.
What role does SharePoint play in multi-agent systems?
SharePoint facilitates collaboration among agents by managing documents and workflows. It streamlines communication, ensuring that agents have access to necessary information in real-time.
Can I customize my agents with Microsoft Copilot?
Yes, Microsoft Copilot offers tools like Copilot Studio for creating modular agents. You can tailor these agents to specific business needs, enhancing their functionality and effectiveness.
What are some common use cases for multi-agent systems?
Common use cases include customer service automation, workflow management, supply chain optimization, and data analysis. Multi-agent systems provide efficiency and improve decision-making in diverse business areas.
How do I ensure security in multi-agent AI solutions?
Implement governance frameworks that address transparency, oversight, and user privacy. Regular monitoring and compliance checks help keep your multi-agent systems secure and accountable.
Are there any best practices for deploying these solutions?
Yes, prioritize thorough documentation, version control, and continuous integration. Optimize agent roles and configurations to maintain system reliability and ensure smooth updates.
How can multi-agent orchestration improve operational efficiency?
By allowing agents to work collaboratively and delegate tasks effectively, orchestration enhances speed and accuracy in workflows, leading to significant operational improvements and cost savings.
🚀 Want to be part of m365.fm?
Then stop just listening… and start showing up.
👉 Connect with me on LinkedIn and let’s make something happen:
- 🎙️ Be a podcast guest and share your story
- 🎧 Host your own episode (yes, seriously)
- 💡 Pitch topics the community actually wants to hear
- 🌍 Build your personal brand in the Microsoft 365 space
This isn’t just a podcast — it’s a platform for people who take action.
🔥 Most people wait. The best ones don’t.
👉 Connect with me on LinkedIn and send me a message:
"I want in"
Let’s build something awesome 👊
1
00:00:00,000 --> 00:00:05,440
Hello everyone and welcome back to the M365FM podcast.
2
00:00:05,440 --> 00:00:10,720
Today we are diving into one of the most exciting areas and Microsoft ecosystem.
3
00:00:10,720 --> 00:00:15,760
We hear a lot about Microsoft co-pilot, AI agents and autonomous systems.
4
00:00:15,760 --> 00:00:20,900
But how do it act really build enterprise-ready solutions that business can trust?
5
00:00:20,900 --> 00:00:26,100
What separates the cool demo from a product-ready AI solution?
6
00:00:26,100 --> 00:00:32,160
The answer to answer this question, I'm joining by someone who spend their days helping
7
00:00:32,160 --> 00:00:37,080
organization make M365 secure, governed and AI ready.
8
00:00:37,080 --> 00:00:42,280
Our guest is Microsoft MVP, Modern Workplace Consultant, International Speaker and an
9
00:00:42,280 --> 00:00:46,440
active contributor to the Microsoft P&P Open Source community.
10
00:00:46,440 --> 00:00:52,060
She specialized in Microsoft 365 governance, SharePoint, co-pilot extensibility, co-pilot
11
00:00:52,060 --> 00:00:58,180
studio and the one of the hottest topics today, architecting multi-agent AI solutions.
12
00:00:58,180 --> 00:01:03,380
So please welcome, wish me out.
13
00:01:03,380 --> 00:01:05,380
Thank you for the kind introduction.
14
00:01:05,380 --> 00:01:10,340
I don't need to introduce myself anymore.
15
00:01:10,340 --> 00:01:14,340
Awesome.
16
00:01:14,340 --> 00:01:20,500
Yeah, can you tell us a little bit about you and especially your journey into Microsoft
17
00:01:20,500 --> 00:01:22,500
technologies?
18
00:01:22,500 --> 00:01:30,500
Wow, that's been quite a long, a long, a long while since I started working at Microsoft
19
00:01:30,500 --> 00:01:38,620
technologies back when I finished my degree in, while I was still like, emerges.
20
00:01:38,620 --> 00:01:44,140
And I need just happened in my first job was like a C# developer.
21
00:01:44,140 --> 00:01:47,140
Actually, not it was VB, VB.NET.
22
00:01:47,140 --> 00:01:55,300
VB.NET.NET.NET.NET.NET.NET.NET.NET.NET.NET.NET.NET.NET.NET.NET.
23
00:01:55,300 --> 00:02:00,060
I see sharp developer and I've done a few on, on ELP systems at that time.
24
00:02:00,060 --> 00:02:07,580
And then when I moved to the UK, that's when I kind of spend the onto SharePoint.
25
00:02:07,580 --> 00:02:13,220
And working, I was kind of working as a, at a particular small company, which is visioned
26
00:02:13,220 --> 00:02:20,420
as in project server. And when you do project server online, it, it's kind of depends on having
27
00:02:20,420 --> 00:02:28,060
a, be SharePoint site and all the platform, the, the active directory at that time.
28
00:02:28,060 --> 00:02:29,780
Now, it's all kind of changed.
29
00:02:29,780 --> 00:02:37,220
So I kind of had like, like from the old school, like on-prem, and then, and then soon after,
30
00:02:37,220 --> 00:02:44,820
it was like the Azure cloud, which was like a game, like another, a, a, a, era, another
31
00:02:44,820 --> 00:02:49,460
innovation called everyone trying to move onto the cloud.
32
00:02:49,460 --> 00:02:55,340
And, and yes, and then, and then a kind of stage onto a particular platform, but nearly,
33
00:02:55,340 --> 00:03:03,460
since I was like in the UK, so it's more like 15 years, 15, 16 years, so seeing it all,
34
00:03:03,460 --> 00:03:08,140
building solutions and some prime and on, on the cloud.
35
00:03:08,140 --> 00:03:18,220
And now, looking, and to have like to adopt, be, implement like all like, the, the AI functionality
36
00:03:18,220 --> 00:03:22,580
at big business, get the best out of the platform.
37
00:03:22,580 --> 00:03:33,220
So yes, it's been a very, a very kind of evolving career in that space, never really static.
38
00:03:33,220 --> 00:03:42,180
And right now, it just feels so quite overwhelming in the sense, like things are changing so quickly.
39
00:03:42,180 --> 00:03:47,860
And to keep up, it just feels impossible to be honest, even when you try to like to build
40
00:03:47,860 --> 00:03:54,820
a solution, it's like, you think like you get it right, and then suddenly you're here, like,
41
00:03:54,820 --> 00:04:01,580
or, or, they have like a new completely new experience, a new interface, and it's just like,
42
00:04:01,580 --> 00:04:07,620
it's, it's, it's just, it's just a fun game.
43
00:04:07,620 --> 00:04:08,620
Yeah.
44
00:04:08,620 --> 00:04:13,340
Or like to keep all on your toes.
45
00:04:13,340 --> 00:04:22,180
You have come known for helping organizations adopt Microsoft 365 responsibility, especially
46
00:04:22,180 --> 00:04:24,380
the topics, governance and compliance.
47
00:04:24,380 --> 00:04:30,860
You are really famous for what, what do you, for responsibility, adoption actually mean?
48
00:04:30,860 --> 00:04:38,300
Um, I would say like, um, yeah, security is very, very, very important.
49
00:04:38,300 --> 00:04:45,500
And then when you consider about co-pilot, um, roll out, to be able to do it securely, it's,
50
00:04:45,500 --> 00:04:52,140
it's kind of simple just to turn, I mean, just switching it on, it's, it's easy.
51
00:04:52,140 --> 00:05:02,980
But like really, the, really like to, to be able to do it in a secure, compliant way, so
52
00:05:02,980 --> 00:05:10,260
that you to prevent any data leakage, for example, like, like, like, a pilot, like, how, how
53
00:05:10,260 --> 00:05:18,460
it is, it can easily, it will kind of, um, surface any data and user have access to, but
54
00:05:18,460 --> 00:05:26,740
then it'd be data itself, but it's not like properly, um, it has, doesn't have like the right
55
00:05:26,740 --> 00:05:32,820
permissions, uh, like following the least, uh, privilege, uh, principle.
56
00:05:32,820 --> 00:05:41,140
It can kind of, um, really, um, put your organization at risk by surfacing all those data,
57
00:05:41,140 --> 00:05:47,940
so you need to have like a good platform hygiene, uh, in terms of ensuring, like your, for
58
00:05:47,940 --> 00:05:55,260
example, your team, channel, shape, on side, have a correct, uh, security, apply to them,
59
00:05:55,260 --> 00:06:04,020
and myself, I kind of wrote a lot of, um, uh, scripts to have, um, all those, uh, try, like,
60
00:06:04,020 --> 00:06:10,180
um, to, to find out, uh, if, um, if ever, we have like, the correct, uh, permissions,
61
00:06:10,180 --> 00:06:15,900
because one of the thing, which is quite hard with shape, born still, then we have co-pilot
62
00:06:15,900 --> 00:06:24,620
is to be able, like, to find out, like, um, on, on a, on, on the side, like, uh, where, um,
63
00:06:24,620 --> 00:06:28,620
if you want to like, to highlight the food inventory or what permissions have been applied
64
00:06:28,620 --> 00:06:34,660
at, at the site, at the file level, at the folder level, at the library level, it's, it's,
65
00:06:34,660 --> 00:06:40,140
it's very hard out of the box, so I kind of, um, contributed a lot of, uh, power, PNP
66
00:06:40,140 --> 00:06:47,900
power, shared scripts, uh, to VEP and B, uh, repository, and I, I know, many, uh, uses have,
67
00:06:47,900 --> 00:06:54,140
um, reused the, uh, those scripts to have to find out whether, uh, uh, they're, they're
68
00:06:54,140 --> 00:06:59,660
tenant, uh, all the shape, one side, or secure, and it's quite, quite nice, like, to receive
69
00:06:59,660 --> 00:07:04,780
a lot of messages, like, uh, thank you, like, my spot really have, but, uh, as you do have,
70
00:07:04,780 --> 00:07:11,660
like, over, like, for party tools, you could use, but, um, but it's not, uh, that means it comes
71
00:07:11,660 --> 00:07:15,900
like, addition and cost, and you have shape on advanced management, but again,
72
00:07:15,900 --> 00:07:23,340
shape on advanced management, it can be a little bit, um, in terms of customizations, it can
73
00:07:23,340 --> 00:07:28,380
be a little bit more limited, but if you, no, have to write on code, and you have, like,
74
00:07:28,380 --> 00:07:34,460
with PNP power, shared, you can do a lot, which is based on your particular requirements,
75
00:07:34,780 --> 00:07:40,140
yeah, so I would say, yeah, permissions, but then you have, like, other things, like, um,
76
00:07:40,140 --> 00:07:46,540
like, using a purview, just to be able, like, to audit monitoring, like, whose age of, um,
77
00:07:46,540 --> 00:07:56,860
of co-pilot, of, um, nipc.org, uh, AI tools in your organization, uh, that's very important as well,
78
00:07:56,860 --> 00:08:02,300
and you can kind of, um, readies, um, what you want to protect, as you send Steve Data,
79
00:08:03,020 --> 00:08:09,580
and, and really like using purview, you need to have to, kind of, put all those guard raids in place
80
00:08:09,580 --> 00:08:16,140
before you start drawing co-pilot, that's what I would say, and it's the same, like, um, yeah, with, um,
81
00:08:16,140 --> 00:08:21,420
with, with, with, with, with, with, you say, uh, AI, and everything, you, you, it's very important, like,
82
00:08:21,420 --> 00:08:29,900
to think, security governance, uh, before trying, like, to go for the speed, um, to, to deployment,
83
00:08:29,900 --> 00:08:37,180
onto production. Yeah, that's, that's cool. Um, when, when customers approach you today,
84
00:08:37,180 --> 00:08:47,660
what are their biggest concerns regarding AI? I would say, like, um, um, it's kind of the, uh,
85
00:08:47,660 --> 00:08:55,100
general kind of, um, concern that we've seen, because right now, people, most people are aware,
86
00:08:55,100 --> 00:09:01,580
like, what is AI, compared, like, to one, two, three, two years ago, um, and, and they're a bit more
87
00:09:01,580 --> 00:09:09,020
conscious, so, and when they come, they, they, it's a no longer about experimentation, but more about
88
00:09:09,020 --> 00:09:17,260
building solutions, uh, to solve, like, a particular problem, but of course, they still have, like, uh,
89
00:09:17,260 --> 00:09:24,380
the, uh, we come so, like, um, knowing, like, we're the mutation of AI, um, about, like, the, uh,
90
00:09:24,380 --> 00:09:32,300
in, uh, in accuracy, hallucinations, uh, and, and again, it's about, yeah, the, the security
91
00:09:32,300 --> 00:09:39,900
risk, whether they are compliant as well, is, um, especially, like, I'm based in the UK, say, there's a lot, uh,
92
00:09:39,900 --> 00:09:50,620
of, um, of strike, um, idea of events like, do, EU, AI, Arc, or DPR, or Hippau, or Salk,
93
00:09:50,620 --> 00:09:57,740
say it's all those, um, and, and I think people are generally kind of, but, we really think about
94
00:09:57,740 --> 00:10:05,660
the security and rest first, before they try, like, do, um, build, like, the, uh, solution, and I think
95
00:10:05,660 --> 00:10:13,660
the other one is, um, is sometimes, um, depending on what they want to do, is, doesn't mean like,
96
00:10:13,660 --> 00:10:19,900
you actually need to be the AI solution, needs to look at other ways, like, automation, um, if
97
00:10:19,900 --> 00:10:26,380
automation can fix, um, reduce case, monographer, automation, like, does it always needs to be an AI
98
00:10:26,380 --> 00:10:32,460
agent? So it's always like, about the education and trying to find, like, what's the best, uh,
99
00:10:32,460 --> 00:10:41,740
solution for the current? Um, is, uh, AI, it's, it's, uh, especially Microsoft, they, they drop,
100
00:10:41,740 --> 00:10:52,140
uh, I feel every day, uh, some new, uh, updates, or, or stuff, um, how, how did you personally keep,
101
00:10:52,140 --> 00:11:01,180
keep up? I'm, uh, I'm a little bit over there, actually. Uh, I, um, the, fortunately, I do feel
102
00:11:01,180 --> 00:11:07,420
like, well, a great, uh, link to now it's, and that's like, a easier way just to keep up to date,
103
00:11:07,420 --> 00:11:13,820
because even like, we had, uh, build, which was three weeks ago, but again, build was more focused
104
00:11:13,820 --> 00:11:22,300
on this time on, uh, developer space, like, on Foundry, um, uh, and GitHub, uh, very much less
105
00:11:22,300 --> 00:11:28,140
on Copa, Cedrio, and set a week, they had like, we released like a whole new experience, which
106
00:11:28,140 --> 00:11:33,900
became into Pravew, and then one week off, they became into a journey of availability. And, and really,
107
00:11:33,900 --> 00:11:40,780
it's like, um, yeah, it's just like, everyday, I'll go into LinkedIn and just like, look at the feed,
108
00:11:40,780 --> 00:11:49,660
um, I'll, I'll post by people I followed, and that's like, I kind of like, oh, what, a new user
109
00:11:49,660 --> 00:11:55,260
interface, and now it's becoming a G, and then one week after, again, when you interface, um,
110
00:11:55,260 --> 00:11:58,960
that became into, back into preview,
111
00:11:58,960 --> 00:12:01,280
which was, like, quite of a relief,
112
00:12:01,280 --> 00:12:02,440
but the same times,
113
00:12:02,440 --> 00:12:06,360
because I was presenting at a particular conference
114
00:12:06,360 --> 00:12:10,760
in CAVAR Studio, and I was like, "Mm, seems like
115
00:12:10,760 --> 00:12:13,400
one of the presenting is already out of date.
116
00:12:13,400 --> 00:12:14,560
"Why shall I do?"
117
00:12:14,560 --> 00:12:16,620
So I felt a little bit of pressure,
118
00:12:16,620 --> 00:12:20,700
like, trying, like, to get off to his fade,
119
00:12:20,700 --> 00:12:23,800
like, to play up, to play with the new,
120
00:12:23,800 --> 00:12:26,580
into a phase and try to get something,
121
00:12:26,580 --> 00:12:29,080
go build some acudemmo,
122
00:12:29,080 --> 00:12:30,480
like, showing the difference,
123
00:12:30,480 --> 00:12:31,920
like, between the old,
124
00:12:31,920 --> 00:12:36,920
classic experience versus the new experience.
125
00:12:36,920 --> 00:12:40,880
Especially when you talk about multi-agent architecture,
126
00:12:40,880 --> 00:12:43,560
it was, like, kind of, quite,
127
00:12:43,560 --> 00:12:45,360
quite, because my session,
128
00:12:45,360 --> 00:12:48,080
X-Astrak had, like, showing,
129
00:12:48,080 --> 00:12:50,280
like, the difference between Charlie and Agent,
130
00:12:50,280 --> 00:12:51,420
and Connected Agent,
131
00:12:51,420 --> 00:12:53,040
and then when you're in experience,
132
00:12:53,040 --> 00:12:55,040
you don't have, like, concept of Charlie,
133
00:12:55,040 --> 00:12:57,480
and any more, when you're just like, "Mm."
134
00:12:57,480 --> 00:12:59,820
But I had to still kind of explain it,
135
00:12:59,820 --> 00:13:01,780
because all the documentation,
136
00:13:01,780 --> 00:13:05,160
and all, at lots of the content videos
137
00:13:05,160 --> 00:13:07,160
was still kind of talk about it,
138
00:13:07,160 --> 00:13:09,160
so it's good just to show, like,
139
00:13:09,160 --> 00:13:12,080
the Charlie agent, where he's, like,
140
00:13:12,080 --> 00:13:13,380
in a classic experience,
141
00:13:13,380 --> 00:13:15,280
in the new experience, like,
142
00:13:15,280 --> 00:13:17,000
you don't have Charlie, and any more,
143
00:13:17,000 --> 00:13:19,320
but it's something different.
144
00:13:19,320 --> 00:13:21,600
You could use, like, skills, for example,
145
00:13:21,600 --> 00:13:24,120
to replicate some of the functionality,
146
00:13:24,120 --> 00:13:27,480
or even you could use, like, workflow,
147
00:13:27,480 --> 00:13:29,840
the workflow, again, last night, another thing,
148
00:13:29,840 --> 00:13:31,440
because you have, like, the workflow,
149
00:13:31,440 --> 00:13:36,440
which is the new way, like, instead of Agent Flow,
150
00:13:36,440 --> 00:13:39,960
which is a classic flow,
151
00:13:39,960 --> 00:13:42,080
and then you have the new flow,
152
00:13:42,080 --> 00:13:44,320
which is a workflow, again,
153
00:13:44,320 --> 00:13:47,160
which is kind of different thing,
154
00:13:47,160 --> 00:13:48,440
and then we've been to the workflow,
155
00:13:48,440 --> 00:13:50,000
what you could do, you can hold, like,
156
00:13:50,000 --> 00:13:53,320
in line agents, which is completely different,
157
00:13:53,320 --> 00:13:55,120
in line agents where you can,
158
00:13:55,120 --> 00:13:57,760
it's kind of similar concept to TA agent,
159
00:13:57,760 --> 00:14:01,080
because it will further the same lifecycle of a workflow,
160
00:14:01,080 --> 00:14:03,680
but then it's not that reasonable.
161
00:14:03,680 --> 00:14:05,880
So, yeah, so it's kind of,
162
00:14:05,880 --> 00:14:11,080
it's hard, it can be quite stressful as well,
163
00:14:11,080 --> 00:14:15,280
but with your best, like, to just do,
164
00:14:15,280 --> 00:14:17,440
however we could, we can,
165
00:14:17,440 --> 00:14:21,840
and just like getting help, like, from all those,
166
00:14:21,840 --> 00:14:25,000
to share the knowledge, which is very important,
167
00:14:25,000 --> 00:14:27,200
so you may ear at age,
168
00:14:27,200 --> 00:14:31,240
where everything is changing so quickly, just to keep off.
169
00:14:31,240 --> 00:14:33,320
- Yeah, you were all over. - And thank you too, Rene.
170
00:14:33,320 --> 00:14:35,880
- Who was on these amazing events,
171
00:14:35,880 --> 00:14:37,800
like the blur Microsoft build?
172
00:14:37,800 --> 00:14:41,640
What surprised you most about Microsoft's AI strategy, actually?
173
00:14:45,960 --> 00:14:50,160
- I think the way, like, one of the announcement
174
00:14:50,160 --> 00:14:55,160
was around the scout, which was the new,
175
00:14:55,160 --> 00:14:58,760
like, what we call autopilot,
176
00:14:58,760 --> 00:15:02,640
which is, again, quite, which is quite surprising,
177
00:15:02,640 --> 00:15:05,160
but it's quite, like, interesting,
178
00:15:05,160 --> 00:15:08,760
because we used to talk about co-hilot being UI for AI,
179
00:15:08,760 --> 00:15:10,600
for the Microsoft platform,
180
00:15:10,600 --> 00:15:13,440
and we keep, when we try to explain co-pilot,
181
00:15:13,440 --> 00:15:16,320
it's co-pilot, and not autopilot,
182
00:15:16,320 --> 00:15:18,960
and so, really, now we have autopilot.
183
00:15:18,960 --> 00:15:23,520
So, again, it's something, yeah, and it's called,
184
00:15:23,520 --> 00:15:26,360
it's meant, like, how am I supposed to position it?
185
00:15:26,360 --> 00:15:28,680
It's, like, for everyone, like, to use,
186
00:15:28,680 --> 00:15:31,440
in a way, to co-pilot,
187
00:15:31,440 --> 00:15:36,680
and it's running, like, in the background,
188
00:15:36,680 --> 00:15:38,680
as your person assisted,
189
00:15:38,680 --> 00:15:43,080
keeping tracking everything that you're doing,
190
00:15:43,080 --> 00:15:48,080
and hoping with all your tasks, like, in the background,
191
00:15:48,080 --> 00:15:54,120
I personally didn't try it, but it's something,
192
00:15:54,120 --> 00:15:56,840
it's something which is powered by the OpenCool,
193
00:15:56,840 --> 00:15:59,920
but it has, like, all the guardrails in place,
194
00:15:59,920 --> 00:16:04,520
like, in Microsoft, there's a show, like, a demo,
195
00:16:04,520 --> 00:16:09,520
where someone tried to ask the autopilot
196
00:16:10,080 --> 00:16:13,560
to clean up, like, the whole desktop,
197
00:16:13,560 --> 00:16:17,200
but that, of course, it just, it kind of,
198
00:16:17,200 --> 00:16:19,840
didn't work, so it's kind of,
199
00:16:19,840 --> 00:16:24,360
it's going back to work within the 13 boundary,
200
00:16:24,360 --> 00:16:29,360
but, of course, yeah, you need, you can't really,
201
00:16:29,360 --> 00:16:33,000
you need to have, like, an in-tune, in-eability device,
202
00:16:33,000 --> 00:16:34,760
and the GitHub Enterprise,
203
00:16:34,760 --> 00:16:38,640
yeah, it will be quite good to see, like, the evolution,
204
00:16:38,640 --> 00:16:42,840
like, when people are, like, using it, and try, like,
205
00:16:42,840 --> 00:16:47,840
to, to, to comment, like, on any, any particular concern,
206
00:16:47,840 --> 00:16:52,840
but they could have, yeah, so, that was kind of, uh,
207
00:16:52,840 --> 00:16:55,160
the scout, and, and the other one,
208
00:16:55,160 --> 00:16:57,760
which was quite interesting, was around Microsoft,
209
00:16:57,760 --> 00:17:01,840
having their own, uh, models, language models,
210
00:17:01,840 --> 00:17:05,520
I don't know, I like to pronounce it, my, my, my, uh,
211
00:17:05,520 --> 00:17:10,040
my, yeah, my, possibly my, yeah, all the different languages,
212
00:17:10,040 --> 00:17:12,840
which was, again, they're looking forward,
213
00:17:12,840 --> 00:17:15,080
like, to be, would like to use them,
214
00:17:15,080 --> 00:17:18,800
is a way that Microsoft is moving, is more,
215
00:17:18,800 --> 00:17:21,000
like, especially, to, so, billing,
216
00:17:21,000 --> 00:17:22,760
and you look at, it, help, copy,
217
00:17:22,760 --> 00:17:26,120
it, um, and now, uh, with, um, copy,
218
00:17:26,120 --> 00:17:29,400
the, uh, studio, as well, um,
219
00:17:29,400 --> 00:17:33,720
a lot, kind of change, like, to be, consumption-based billing,
220
00:17:33,720 --> 00:17:38,200
instead of having, like, uh, one line sits for everything,
221
00:17:38,200 --> 00:17:43,200
so, so, based on the consumption, your, your costs,
222
00:17:43,200 --> 00:17:45,480
but, kind of, very, say, okay,
223
00:17:45,480 --> 00:17:47,720
it's something, like, keeping an eye on,
224
00:17:47,720 --> 00:17:49,840
it's quite easy when it was just, like,
225
00:17:49,840 --> 00:17:51,920
emphry spike, or, by the way, since,
226
00:17:51,920 --> 00:17:54,920
one, uh, one fixed fee,
227
00:17:54,920 --> 00:17:58,720
but now, you have to be a bit more careful,
228
00:17:58,720 --> 00:17:59,960
related to that.
229
00:17:59,960 --> 00:18:02,280
So, I'm hoping, like, with the Microsoft,
230
00:18:02,280 --> 00:18:07,080
but, language models, it will be a little bit cheaper
231
00:18:07,080 --> 00:18:10,880
to, uh, to run those models, hopefully,
232
00:18:10,880 --> 00:18:13,880
and, and will have, like, to reduce some of the costs,
233
00:18:13,880 --> 00:18:17,720
uh, of course, um, before picking any model,
234
00:18:17,720 --> 00:18:19,960
you need to do your own evaluation,
235
00:18:19,960 --> 00:18:22,320
whether that's, like, the best, uh,
236
00:18:22,320 --> 00:18:24,200
the best, uh, options,
237
00:18:24,200 --> 00:18:27,360
both to choose for your, for use case,
238
00:18:27,360 --> 00:18:29,920
and it's always tricky, because even, like,
239
00:18:29,920 --> 00:18:32,560
emphry, it's quite a good one, then, when you think,
240
00:18:32,560 --> 00:18:34,520
there's, like, thousand,
241
00:18:34,520 --> 00:18:36,720
nomen, thousand on models,
242
00:18:36,720 --> 00:18:40,000
and, but emphry, we do have, like,
243
00:18:40,000 --> 00:18:44,440
something, like, to allow you to evaluate the models,
244
00:18:44,440 --> 00:18:47,240
based on certain benchmarking,
245
00:18:47,240 --> 00:18:49,880
but, again, like, trying, like, to know,
246
00:18:49,880 --> 00:18:53,200
whether that's the best, uh, model, like, to use,
247
00:18:53,200 --> 00:18:56,160
uh, it's, um, it's tricky, I would say.
248
00:18:56,160 --> 00:18:59,480
Yeah, maybe, uh, loads of announcements from Bird,
249
00:18:59,480 --> 00:19:02,960
as well, just, like, you could only take the whole session.
250
00:19:02,960 --> 00:19:04,800
(laughs)
251
00:19:04,800 --> 00:19:08,840
And, um, yeah, if someone asks you, we will co-pilot.
252
00:19:08,840 --> 00:19:11,400
What, what's the first question you, uh, yeah,
253
00:19:11,400 --> 00:19:12,240
will ask them?
254
00:19:12,240 --> 00:19:21,680
Um, um, if they want to co-pilot, um,
255
00:19:21,680 --> 00:19:27,680
I know, a lot of it is kind of being driven by, uh,
256
00:19:27,680 --> 00:19:32,200
by what, like, over what's happening, like, in the market,
257
00:19:32,200 --> 00:19:36,160
and everyone wants, like, to be, um, on the lead,
258
00:19:36,160 --> 00:19:39,440
or not in the lead, but at least, like, not to feel, like,
259
00:19:39,440 --> 00:19:43,040
they are falling behind, and try, like, to keep up,
260
00:19:43,040 --> 00:19:47,720
and try, like, to, uh, to, to, kind of, embrace it.
261
00:19:47,720 --> 00:19:54,600
I would say that's a good decision, that they'd try, like, to, uh, to, to, to, uh,
262
00:19:54,600 --> 00:19:58,040
to go on and then board co-pilot for end-users,
263
00:19:58,040 --> 00:20:01,040
because of all the benefits that it provides.
264
00:20:01,040 --> 00:20:04,240
Um, but again, what we could see is, um,
265
00:20:04,240 --> 00:20:08,120
a lot of organisations, not like, really, you're ready,
266
00:20:08,120 --> 00:20:11,480
like, just to roll out co-pilot, yeah, so,
267
00:20:11,480 --> 00:20:14,760
it could be because we are, like, um, still on-prem,
268
00:20:14,760 --> 00:20:17,320
like, on the, um, the, guessy platform,
269
00:20:17,320 --> 00:20:20,760
which, as it's really, um, co-pilot,
270
00:20:20,760 --> 00:20:23,880
and, and really, the first thing would be, like, um,
271
00:20:23,880 --> 00:20:28,360
to move, um, the whole, like, the platform, um,
272
00:20:28,360 --> 00:20:33,400
that they have, of course, it's, it's kind of a lengthy process onto the end-
273
00:20:33,400 --> 00:20:38,440
and, and for 65, and, and, of course, um, setting all, like, we, um,
274
00:20:38,440 --> 00:20:42,600
be bound to raise the guardraids, our arms, so that they can use, uh,
275
00:20:42,600 --> 00:20:46,680
co-pilot safely, before we actually roll out co-pilot.
276
00:20:46,680 --> 00:20:51,400
So, yeah, so the first thing would be defining out whether the kind is ready for it,
277
00:20:51,400 --> 00:20:55,640
and then you have to go, like, for a readiness, uh, kind of, um,
278
00:20:55,640 --> 00:21:00,120
workshop trying to see where they are at and, and just have in that journey.
279
00:21:00,120 --> 00:21:04,840
Yeah, what would you say is, it's a, uh, co-pilot, or, uh,
280
00:21:04,840 --> 00:21:11,000
I roll out, um, more, um, I keep project, or is it more,
281
00:21:11,000 --> 00:21:13,960
a business transformation initiative?
282
00:21:13,960 --> 00:21:19,960
I would say it has to be a whole business transformation initiative,
283
00:21:19,960 --> 00:21:25,880
rather than IT-led, because the people know their, their tasks, uh,
284
00:21:25,880 --> 00:21:29,000
they will be able, like, to set up the benchmark,
285
00:21:29,000 --> 00:21:33,160
in as, like, trying, like, to identify, like, the return on investment.
286
00:21:33,160 --> 00:21:39,640
It's, like, the end, the business people would be better equipped to make all those, um,
287
00:21:39,640 --> 00:21:45,800
assessments rather than an IT person, and, and, and really, it's, um,
288
00:21:45,800 --> 00:21:56,040
But everyone will have to be on board and trained to get, like, the best out of co-pilot,
289
00:21:56,040 --> 00:22:05,480
say, and the 80 out there just as someone who's going like, to support, like, they're all out.
290
00:22:05,480 --> 00:22:15,320
And support the business, and, yeah, and just help out. But, yeah, it's all, like,
291
00:22:15,320 --> 00:22:20,840
a business. And really, we need to have, like, the leadership buying as well, who believes in it,
292
00:22:20,840 --> 00:22:29,320
like, all in all its potential, and trying, like, to, yeah, to kind of, uh, drive it.
293
00:22:29,320 --> 00:22:39,640
Yeah. Let's move into one of your specialty. Yeah, it's co-pilot extensibility. What are the exactly
294
00:22:39,640 --> 00:22:48,760
as, as this mean? Okay, so co-pilot is, um, whatever Microsoft provides out of the box, say,
295
00:22:48,760 --> 00:22:54,840
could be like, anything girl, which you see, like, on the Amphits bike, a pilot chart,
296
00:22:54,840 --> 00:23:02,840
on SharePoint, it could be like, in your apps, uh, upload everywhere. And then you have,
297
00:23:02,840 --> 00:23:08,200
what used to be called, like, maybe one year ago, two years ago, custom co-pilot.
298
00:23:09,240 --> 00:23:18,760
And that's the call, like, agents sound. So co-pilot extensibility is, um, aren't really, like, building agents.
299
00:23:18,760 --> 00:23:27,320
And, um, yeah, I know we'd be terminologies, but different when you talk about it, or, or, or, or,
300
00:23:27,320 --> 00:23:34,280
or agent, but it's kind of same. And it can be, like, anything, anything custom either you build,
301
00:23:34,280 --> 00:23:40,920
like using, um, agent builder, which is kind of simple from the Amphits bike pilot,
302
00:23:40,920 --> 00:23:47,320
or you can have, like, from SharePoint as well, but you can create, like, SharePoint agents,
303
00:23:47,320 --> 00:23:53,800
quite, like, simple use case, like, which can be done by any end user, just, about, just using
304
00:23:53,800 --> 00:23:58,920
a little bit of prompting, and they can build something. And why you want, like, to build, like,
305
00:23:58,920 --> 00:24:06,280
agents instead of using, um, my chart is, uh, because you have, like, a reusable prompt, but, you know,
306
00:24:06,280 --> 00:24:10,840
like, if you kind of taping the same prompt over and over again, like, to get the same output,
307
00:24:10,840 --> 00:24:15,720
for example, you want, like, a weekly report, you might want, like, to, to put that agent for, like,
308
00:24:15,720 --> 00:24:21,400
to be as simple as, uh, SharePoint agent. I'd like to do a report if all your data is already in SharePoint.
309
00:24:21,400 --> 00:24:28,520
Um, or you can have co-pilot studio as well, which is more, like, towards the, um, local, um,
310
00:24:29,480 --> 00:24:36,440
power, uh, the make, but the maker is a system, the movements, um, but again, it's a little bit, uh,
311
00:24:36,440 --> 00:24:42,920
I will say it's, like, completely low code because it can have, but you can get a little bit more complex
312
00:24:42,920 --> 00:24:49,320
with, uh, code when you use connectors and other than you experience your hard skills, your skills,
313
00:24:49,320 --> 00:24:56,280
you can have, in-built, scrap, spit, um, script. And as well, um, when you look at some,
314
00:24:56,280 --> 00:25:05,160
anthropic skills, we have, um, a lot of, um, references to, uh, to, to over, kind of, um,
315
00:25:05,160 --> 00:25:12,600
script files, like JavaScript or Python or something. It can be quite, quite complex, even if you say,
316
00:25:12,600 --> 00:25:19,560
cover a studio, being a local, doesn't always have to be. And you have another one, which doesn't, uh,
317
00:25:19,560 --> 00:25:27,880
get talk very, um, a lot is the agents talk it. agents talk it is an extension within VS code,
318
00:25:27,880 --> 00:25:32,520
which allows you to build like with the creative agent. And that mean like, the creative agent in
319
00:25:32,520 --> 00:25:38,680
that sense is the agent will use the same orchestrator, which go pilot use, but again, you can
320
00:25:38,680 --> 00:25:46,760
bring your own instructions, you can bring your own, like, um, plugins. So on cp servers, um, you can
321
00:25:46,760 --> 00:25:53,800
quite, you can do quite a lot with agents talk it. And then if you go towards a pro code, um,
322
00:25:53,800 --> 00:26:02,120
options, you have a Azure Foundry, and you can use like, um, um, over options, like,
323
00:26:02,120 --> 00:26:10,280
the log change, the Microsoft agent framework. And it's, it's, um, it's just like your imagination is,
324
00:26:10,920 --> 00:26:17,240
is your image really? And you can, you can, you can't, you can't, when you go towards that level,
325
00:26:17,240 --> 00:26:23,480
you're choosing like your own model, because, uh, with go pilot, you're kind of restricted with, uh,
326
00:26:23,480 --> 00:26:29,080
unlimited, uh, models that are provided onto the platform. And when you go to a frondry,
327
00:26:29,080 --> 00:26:36,680
you can never reach the thousand of, you can pick any other model. And as well as you can build your
328
00:26:36,680 --> 00:26:44,040
own orchestration using the long change, it will be Microsoft agent framework. Um, yeah, so it's
329
00:26:44,040 --> 00:26:52,760
quite a lot. So be global access, the day again. So very, uh, huge, broad platform. And, and you just
330
00:26:52,760 --> 00:27:01,560
kind of need to know, like one, two, use which depending on the use case. And, and what, what will,
331
00:27:01,560 --> 00:27:14,200
there was Microsoft Graph, uh, place? Yes, uh, Microsoft Graph, one of the thing is, um, it's, um,
332
00:27:14,200 --> 00:27:22,120
it's free to make, um, uh, cold, like press API, cold, Microsoft Graph. And, um, like, we've been
333
00:27:22,120 --> 00:27:28,760
in Kaba studio, if he's agents, okay, um, or even if he, I'm not such a, uh, foundry, but I'm pretty
334
00:27:28,760 --> 00:27:35,480
sure there should be a way of connecting, uh, to Microsoft Graph, reconnectes, uh, all plugins in
335
00:27:35,480 --> 00:27:43,720
agent's toolkit. But one of the thing, uh, what Microsoft kind of, um, announced, uh, in, Microsoft
336
00:27:43,720 --> 00:27:52,200
Ignite last year was about the, um, IQ, all the different types of IQ, uh, the work IQ, fabric IQ,
337
00:27:53,400 --> 00:28:01,080
and, um, the foundry IQ, lots of IQ and, and during build, it's, uh, three weeks ago, the
338
00:28:01,080 --> 00:28:07,880
announced that the web IQ and the Microsoft IQ, which kind of, I'm combustible of it. So, Microsoft Graph
339
00:28:07,880 --> 00:28:15,560
is still very important because, um, when you look at the work IQ and all, like, the, uh, endpoints,
340
00:28:15,560 --> 00:28:24,760
which are available, it doesn't give you the full spectrum to be able, like, to do everything you
341
00:28:24,760 --> 00:28:31,000
want. So, Microsoft Graph is there, so there's a bit more rich in, in terms of, what the endpoints
342
00:28:31,000 --> 00:28:37,560
it provide. But still, I would say Microsoft Graph doesn't do everything because a lot of times,
343
00:28:37,560 --> 00:28:43,000
when I have to do something quite specific, I still need to rely on the, uh,
344
00:28:43,000 --> 00:28:48,760
Shebhorn, Rassapri, all the C-Some, like, back on the days, the client side of a checker model.
345
00:28:48,760 --> 00:28:57,160
I know, sorry, old school, but it's good, like, to still know the foundation and how everything,
346
00:28:57,160 --> 00:29:05,720
is kind of stuck, uh, and worked together. But of course, like, the, the model is everything
347
00:29:05,720 --> 00:29:17,480
towards, um, the MCB and work IQ do provide, um, ready, available MCB, which, uh, to, to do quite a lot of
348
00:29:17,480 --> 00:29:24,440
different things. But again, it just depends on what you actually wants to do. It might not always
349
00:29:24,440 --> 00:29:33,160
be, um, be feasible to just rely on work IQ, MCP, all Microsoft Graph is still need to know, but Shebhorn
350
00:29:33,160 --> 00:29:39,720
Rassapri, uh, and see some, unfortunately, it is like, it is, like, it's only like this year, for
351
00:29:39,720 --> 00:29:48,520
example, Microsoft Graph had it end point to create Shebhorn side, and you just have to think like,
352
00:29:48,520 --> 00:29:57,160
what, what should we, like, go from, from ages ago, but it's still very slow, like, to catch up,
353
00:29:57,160 --> 00:30:05,800
and a lot of the thing probably, literally, um, in the Rassapri or CSOF first before it gets, um,
354
00:30:05,800 --> 00:30:12,520
added to the Microsoft Graph, all, all probably to work IQ. But we'll see, I think shebhorn,
355
00:30:12,520 --> 00:30:20,520
which is always good, you know, um, all the possible, um, all the possible routes, I can
356
00:30:20,520 --> 00:30:28,280
give me half back to build a customization. Yeah, um, actually when I go into LinkedIn,
357
00:30:28,280 --> 00:30:34,120
there's, it's one topic I come by, my timeline is completely flooded with, it's, it's multi-agent
358
00:30:34,120 --> 00:30:47,800
solutions. Uh, so what, what is, um, multi-agent system? Um, yeah, so, say, it's, um, it's having,
359
00:30:47,800 --> 00:30:55,080
like, you think, like, a team of people, each one, uh, each one is specialised in their specific
360
00:30:55,080 --> 00:31:03,400
domain, um, knowledge, uh, and able, like, to do setting activities. It's similar way, like,
361
00:31:03,400 --> 00:31:10,200
thinking, because one particular person can't know everything, can't do everything, so, so trying,
362
00:31:10,200 --> 00:31:17,960
like, to build a multi-agent solutions will have towards, um, towards getting, like, a more
363
00:31:17,960 --> 00:31:25,080
accurate and more powerful than solution. Is imagine if you're building one, um, agent,
364
00:31:25,080 --> 00:31:32,280
which does everything in your organization, from IT, sales, marketing, and then you cannot
365
00:31:32,280 --> 00:31:39,880
equip it with all the different tools, but it needs to be able to do those, and the knowledge,
366
00:31:39,880 --> 00:31:48,520
as well, a lot of different knowledge, your agent is going to struggle, you'd be able to reason,
367
00:31:48,520 --> 00:31:55,640
to find, like, the correct route of actions, but it needs to do for a particular, for a particular
368
00:31:55,640 --> 00:32:01,960
task, because how's that to enumerate for all the different tools, all the different knowledge,
369
00:32:01,960 --> 00:32:08,600
and sometimes it may be confused if the knowledge of the tools named have similar name or,
370
00:32:08,600 --> 00:32:15,960
um, or not the same name, but it's kind of almost a similar name. So, so it can be quite hard.
371
00:32:15,960 --> 00:32:23,880
So having, like, smaller agent, just specialised, like, in its specific knowledge, with a subset of,
372
00:32:23,880 --> 00:32:31,480
uh, tools, subset of knowledge, subset of instructions, much less, it's going, like, to perform,
373
00:32:31,960 --> 00:32:39,320
better, and then what, when you do multi-agent solutions, uh, one of the thing is, you'll need,
374
00:32:39,320 --> 00:32:46,760
like, to think of the orchestrator, or parent agent, so you will have to have, like, um, like,
375
00:32:46,760 --> 00:32:53,000
like, if you consider, like, with the team of people, you might have, like, a manager, or a team leader,
376
00:32:53,000 --> 00:33:00,760
who cannot delegate the jobs to, to its team members, so similarly, you have to think of the parent
377
00:33:00,760 --> 00:33:08,120
agent, orchestrator agent, and then base, like, intent of the user, it will kind of determine
378
00:33:08,120 --> 00:33:15,720
which agent that trigger, and your parent is a needs to be very, um, very, kind of, it is a needs to
379
00:33:15,720 --> 00:33:23,560
be very, um, loaded with loads of instructions, which, which are agents, will, all the connected agents
380
00:33:23,560 --> 00:33:30,280
will have, like, everything it needs, uh, to do, uh, once it gets triggered. So your parent agent,
381
00:33:30,280 --> 00:33:37,000
will be there just to orchestrate, like, one each of the connected agents gets triggered, and,
382
00:33:37,000 --> 00:33:45,480
they'll, and, and, and, and then everything. Sorry, we get cannot be executed, and, uh, as per the
383
00:33:45,480 --> 00:33:54,280
instructions of our connected agent. So, so, yes, so, we have integers of having, um, like, a multi-agent
384
00:33:54,280 --> 00:34:01,560
solutions with allow you to scale, um, um, and have, like, more accurate results, and as
385
00:34:01,560 --> 00:34:10,280
well as it's easier to maintain, to test, uh, because it's smaller, just with a few instructions,
386
00:34:10,280 --> 00:34:17,080
uh, tools, knowledge, say it's kind of easier, like, to test as well, and if you think, like, in terms of
387
00:34:17,080 --> 00:34:22,760
plug and play, if something, you think like, a agent is not doing as it is, you can take it out,
388
00:34:22,760 --> 00:34:30,280
and you can put, like, another agent in easily. And it, in terms of multi-agent architecture, it's quite,
389
00:34:30,280 --> 00:34:38,440
it's not only, um, only on the same platform, so if you can still, like, go by the studio, you can,
390
00:34:38,440 --> 00:34:46,920
connect cross-platform as well. For example, you could have, like, um, agent, like, in fabric, um, which is,
391
00:34:47,480 --> 00:34:53,960
which kind of, which is like, on the platform, someone already built that agent, and instead of
392
00:34:53,960 --> 00:35:00,280
rebuilding it, you can easily create a connection to that fabric agent. It's similarly, like,
393
00:35:00,280 --> 00:35:08,840
you could have, like, um, agent hosted in, uh, Microsoft Foundry, and from Copa Las Studio, you can
394
00:35:08,840 --> 00:35:16,920
create a connection to that, uh, Azure Foundry. So, like having a multi-agent architecture, it
395
00:35:16,920 --> 00:35:25,800
allows, like, to bring, um, to, to kind of make use of, um, to bring all of the, like, the agent across
396
00:35:25,800 --> 00:35:32,200
different platforms, like, it's working together, because you might have different reasons why you want
397
00:35:32,200 --> 00:35:39,960
to have, like, something quite specific in Foundry, uh, doing some quite, uh, complex, um, tasks, for example,
398
00:35:41,400 --> 00:35:49,160
that's a, for insurance claim, um, kind of scenario. So, the back end, you might want to do, like, um,
399
00:35:49,160 --> 00:35:56,120
to have an agent to, to detect fraud, for example. So, you might be using, like, something in Foundry,
400
00:35:56,120 --> 00:36:04,280
which is, like, a different model, which is not, uh, really available in Copa Las Studio, and, and,
401
00:36:04,280 --> 00:36:10,920
and then from Kobayost Studio, you do have the capability to select your model,
402
00:36:10,920 --> 00:36:23,080
but the list is quite still limited. So you have to think, separately, you might want to
403
00:36:23,080 --> 00:36:33,720
build an agent on different platforms, even like on the Empress 5, Kobayost chat using Agent Builder,
404
00:36:33,720 --> 00:36:39,560
so you'll have different agents built on different platforms, but you want to bring all together,
405
00:36:39,560 --> 00:36:44,280
memory agent architecture, kind of help as well in that particular instance.
406
00:36:44,280 --> 00:36:54,680
One topic is also when we talk about agents as autonomy, how much autonomy should an enterprise agent
407
00:36:54,680 --> 00:36:56,040
receive from your perspective?
408
00:36:58,840 --> 00:37:05,480
Yeah, that's like another, it depends on the use case, I would say.
409
00:37:05,480 --> 00:37:14,760
For example, one simple use case is like, if an email is received into
410
00:37:14,760 --> 00:37:23,320
email inbox to kind of process that email, for example, you think of a support desk system,
411
00:37:23,960 --> 00:37:33,000
you might want to have an agent to go for an inbox and do a triage to see what to determine
412
00:37:33,000 --> 00:37:41,320
the severity or which teams need to handle it and to assign you to back to the team,
413
00:37:41,320 --> 00:37:50,120
and as just determining the time that it will take to kind of process that particular request,
414
00:37:50,120 --> 00:38:00,040
and to send the outcome or response back to the user who sends the email initially.
415
00:38:00,040 --> 00:38:08,120
So it depends on use case, this kind of example of autonomy, it gets triggers when the email is
416
00:38:08,120 --> 00:38:15,080
received, so you could have other triggers, for example, on your shareboard place, for example,
417
00:38:15,080 --> 00:38:23,640
you might have like an other application, which, for example, if someone sent, let's say,
418
00:38:23,640 --> 00:38:31,960
submit like a receipt for travel expense, for example, and then once it's submitted to your database,
419
00:38:31,960 --> 00:38:38,040
which could be shareboard, or it could be anything else, and then have a trigger which creates
420
00:38:38,040 --> 00:38:45,800
an approved task for someone to be able to approve it, or you can even have like a, be able to
421
00:38:45,800 --> 00:38:56,520
approve it as well, depending like if a request meets all your travel expense policy document,
422
00:38:56,520 --> 00:39:02,120
and yeah, why not, it can be like, but to approve as well, so you could have an agent,
423
00:39:02,120 --> 00:39:09,320
which does it, and then the payments get triggered and sent to the user who raised the travel expense,
424
00:39:09,320 --> 00:39:16,920
and of course, if it's a little bit more, if it doesn't meet the travel expense policy,
425
00:39:16,920 --> 00:39:23,160
then someone will have to money your review and try to find out what is happening and try to ask for
426
00:39:23,160 --> 00:39:30,760
rest of it, so it has to save a lot of time, like, right, by the route we think the amount of work,
427
00:39:30,760 --> 00:39:37,800
like if someone has, you know, an approved, everyone of them will be approved, the request will be
428
00:39:37,800 --> 00:39:42,840
reduced, I'm lucky to be a burden because we have an agent, which does it, of course, you have to
429
00:39:42,840 --> 00:39:48,920
define the guard read, and the boundaries under which it can automatically be approved,
430
00:39:48,920 --> 00:39:53,720
so it does have like, its use case for it to know me, but of course, everything will have to be assessed
431
00:39:53,720 --> 00:40:03,800
so about its use for this. Another topic, when we do talk about multi-agents, it's the model context
432
00:40:03,800 --> 00:40:12,520
protocol, or an NCP, what is it, and what is the different, I say, to in microservice or to an API?
433
00:40:12,520 --> 00:40:22,840
I would say like, NCP, it's kind of a wrapper around, like, all like the different APIs,
434
00:40:22,840 --> 00:40:30,440
like, for example, I mentioned about the work IQ, NCP, which is provided by Microsoft
435
00:40:30,440 --> 00:40:36,920
out of the box, so you don't have to build your own NCP, if you want to utilize it, of course,
436
00:40:36,920 --> 00:40:45,000
if you do have like, another application, which doesn't have like, its NCP yet, your custom
437
00:40:45,000 --> 00:40:54,680
application, you can build your own NCP server for it. So it allows, like, one of the advantage of
438
00:40:54,680 --> 00:41:04,280
using NCP is it has a lot of different tools, readily available, so actually kind of, instead of
439
00:41:04,280 --> 00:41:15,080
creating, like, for some, if you look at Kovar Studio, if you have different APIs, so you'll have
440
00:41:15,080 --> 00:41:23,560
to create different connections, like independent, okay, not independent, but like a custom connector,
441
00:41:23,560 --> 00:41:33,000
to API call and then define all your security boundaries, but if you use NCP, Microsoft IQ, NCP,
442
00:41:33,000 --> 00:41:40,680
for example, let's say, outlook, and I'm not sure what work IQ, email, NCP, I think it's
443
00:41:40,680 --> 00:41:46,920
quite email, let's see if it all could be something else. So it does have, give you a lot of the
444
00:41:46,920 --> 00:41:56,760
tools ready to be used, so you can tap in easily onto those. So it kind of makes it a bit easier
445
00:41:56,760 --> 00:42:07,560
for the developer, who is building the solution to leverage NCP instead of API, but we're only
446
00:42:07,560 --> 00:42:15,560
thinking, I would say, because NCP has, it can bring a lot of noise into your applications in terms of
447
00:42:15,560 --> 00:42:26,120
one NCP, like, can have like 10 to 20 or 30 tools, can create some noise, so again, I would say
448
00:42:26,840 --> 00:42:34,760
on that instance, and show that you're unable to tool in NCP that you actually want your
449
00:42:34,760 --> 00:42:44,760
agents to be able to do, for example, if you have like my email, NCP, if you want your agents to use it,
450
00:42:44,760 --> 00:42:53,960
just send email, you want to disable everything, like, create email, or do delete email, you want to
451
00:42:53,960 --> 00:43:01,640
disable it onto your NCP tools, so that your agents is allowed, has only access to the tools it needs
452
00:43:01,640 --> 00:43:05,880
to do its right, so it's something like just to be, to be aware of.
453
00:43:05,880 --> 00:43:16,840
I think a little bit now everyone can build their multi-agent systems, but what implication
454
00:43:16,840 --> 00:43:26,360
had this on, yeah, I think there are kind of new employees, everyone had, what roles do identity,
455
00:43:26,360 --> 00:43:35,480
or enter ID play, and what roles do governance, especially peer view, become in these systems?
456
00:43:35,480 --> 00:43:45,080
Yeah, what's it like, especially like since Ignite last year, Microsoft announced like the agent
457
00:43:45,080 --> 00:43:57,800
Physics 5, and now it became into GUE, now I think, response, or three weeks ago, which is
458
00:43:57,800 --> 00:44:06,680
really, really interesting, so the agent Physics 5 will have like to solve some of the concern that
459
00:44:06,680 --> 00:44:14,680
users have related to governance for agents, you know, exactly what they are doing, what data they
460
00:44:14,680 --> 00:44:23,240
have access to, and they have like their own identity, which you can kind of control,
461
00:44:23,240 --> 00:44:30,120
similar way, like to a user having an identity controlling access, so today they are even blocker
462
00:44:30,120 --> 00:44:38,840
agent in even like, an entral ID as well, because in the entire ID you will see the agent
463
00:44:39,480 --> 00:44:46,360
similar way, do you see it like a user, and you can kind of have a little bit more control now,
464
00:44:46,360 --> 00:44:53,480
like without agent Physics 5 it's going to be harder, and the agent Physics 5 is integrated,
465
00:44:53,480 --> 00:45:03,000
that with peer view defender, the entribe, or like the tools that need to be able to like to see,
466
00:45:04,280 --> 00:45:11,240
you have like one source of truth, one control pane, where you can do all those things,
467
00:45:11,240 --> 00:45:23,240
so again, and the based on what is happening, you can take actions that you cannot protect,
468
00:45:23,240 --> 00:45:29,560
your data, proven, the page, extra, so yeah, so identity and governance, that was something
469
00:45:31,160 --> 00:45:40,360
which was unorthodoxy, nothing in that last year, which is that created, take it, and a big leap forward
470
00:45:40,360 --> 00:45:47,240
into ensuring like everyone, or we are building like, the agents which are secure and scary.
471
00:45:47,240 --> 00:45:55,160
And what role do you set a meta play, or some of the play actually in this agent world?
472
00:45:56,920 --> 00:46:01,080
Sorry, would be the question again? Yeah, what role do you set a meta that have some
473
00:46:01,080 --> 00:46:08,680
the play in this agent world? Yeah, if I was something which was a little bit
474
00:46:08,680 --> 00:46:16,840
a miss, I think right from the beginning, where like, special and good part, it would be able to
475
00:46:16,840 --> 00:46:26,200
to reason on the contents of a file, but not actually take into account like the metadata,
476
00:46:26,200 --> 00:46:32,680
which are assigned to the file, like all your content types, or any custom columns you might have
477
00:46:32,680 --> 00:46:43,880
or onto your libraries or list, the metadata do play a very good deal. It does, it's very important
478
00:46:43,880 --> 00:46:48,840
because that's how like views is searched, and that's how like right from the beginning, especially
479
00:46:48,840 --> 00:46:56,920
with the role out of Ampsix 5 or SharePoint, having like a strong information architecture is something
480
00:46:56,920 --> 00:47:06,120
like we always advise the client and metadata, custom type, taxonomy, and everything are very
481
00:47:06,120 --> 00:47:14,120
important, and then we be co-pilot when it seems like to be in your ring, those, but I think now,
482
00:47:15,480 --> 00:47:22,040
I didn't try to co-pilot, but I need to try to say it, but definitely like now you have like co-pilot
483
00:47:22,040 --> 00:47:26,920
in SharePoint, which is used to be called the only SharePoint, previously used to be called knowledge
484
00:47:26,920 --> 00:47:34,040
knowledge agent, and before I was called syntax, it has lots of different names, you don't know what's
485
00:47:34,040 --> 00:47:43,480
going to be called in the future. It's very powerful, it does reason on metadata, for example, if you
486
00:47:43,480 --> 00:47:53,560
have a list of invoice, and you are said to provide you a sum of all like the invoice which happened
487
00:47:53,560 --> 00:47:59,400
the last week, it is quite a very good job, like to reason on those metadata and to give you a good
488
00:47:59,400 --> 00:48:04,840
response. It's getting better, so maybe not always give you a more accurate answer, but always check,
489
00:48:04,840 --> 00:48:12,120
but it's improving, but again, into some metadata, I'm not so sure, it does it, but the way I get
490
00:48:12,120 --> 00:48:19,400
around it is using Azure AR Search, so using Azure AR Search you can create a connections to your
491
00:48:19,400 --> 00:48:26,360
SharePoint library, and within that you can define the metadata you want the Azure AR Search to
492
00:48:26,360 --> 00:48:33,240
capture the reason on, and you can kind of, and then you can connect the Azure AR Search
493
00:48:33,240 --> 00:48:41,000
to your co-pilot studio, and you can ask questions on those metadata, and it will reason on those.
494
00:48:41,000 --> 00:48:50,360
So that's a way, like a workaround, I found like getting from a metadata details, but of course there
495
00:48:50,360 --> 00:48:55,640
might be other ways, but we have to see, or even like out of the box, I'm not so sure, that that's
496
00:48:55,640 --> 00:49:03,400
something that metadata is very important. And information architecture, a lot of companies, I think,
497
00:49:03,400 --> 00:49:12,840
it's the fat boy on the football game that's sitting outside all the time, is this information?
498
00:49:12,840 --> 00:49:16,360
Hashtag shall now be coming, yeah, a superhero?
499
00:49:16,360 --> 00:49:32,280
I would say, yes, it's still relevant, and it's still like the backbone of your document
500
00:49:33,000 --> 00:49:39,160
management system, if ever you're moving, because otherwise it'll become difficult for users
501
00:49:39,160 --> 00:49:50,440
to query, to search, and it will become kind of a big mess. So it's better, like even before
502
00:49:50,440 --> 00:49:56,920
like starting, just put in place like a strong information architecture, and it will eventually help
503
00:49:56,920 --> 00:50:06,840
your ER and co-pilot out in the ultimately, one of the metadata, or a kind of being
504
00:50:06,840 --> 00:50:14,120
index room, the co-pilot, but we'll see, about that, but it's still for now, there's still ways
505
00:50:14,120 --> 00:50:23,800
like you can kind of make sure you're utilizing your strong AI, of the framework or design,
506
00:50:23,800 --> 00:50:32,360
of what you put in place. So keep a focus on AI, which is very important.
507
00:50:32,360 --> 00:50:42,520
Oh, we're running a little bit on Skyvercy. So, yeah, I have an every interview I do a quick
508
00:50:42,520 --> 00:50:48,600
fire round, so I ask a short question, and you say, what comes in your mind?
509
00:50:50,440 --> 00:50:54,120
So my first one is, what's your favorite SharePoint feature?
510
00:50:54,120 --> 00:50:59,000
Sorry, SharePoint? What's your favorite SharePoint feature?
511
00:50:59,000 --> 00:51:09,640
Oh, SharePoint feature. Oh, I think I'm enjoying the co-pilot in SharePoint, because it's
512
00:51:10,680 --> 00:51:19,640
even if it's new, it's kind of take it to another level in terms of what she can do.
513
00:51:19,640 --> 00:51:28,360
Like, for the last SharePoint hackathon, I've used it a lot, like, to do a lot of the manual stuff I would
514
00:51:28,360 --> 00:51:35,800
do, like creating the libraries, creating the color, creating the list, like even generating test data,
515
00:51:36,920 --> 00:51:45,880
and even like to add skills to co-pilot in SharePoint to be able, like, to, for example,
516
00:51:45,880 --> 00:51:52,440
to generate a report based on the data about how, I know it's just a demo, but like,
517
00:51:52,440 --> 00:52:01,800
the potential is great, but again, I would say use it responsibly, because again,
518
00:52:01,800 --> 00:52:10,280
co-pilot, SharePoint, you can easily kind of create a lot of chaos as well, if it not used for a birthday.
519
00:52:10,280 --> 00:52:19,240
Yeah, RAM gets the very expensive, and you get a new laptop, and it's the only
520
00:52:19,240 --> 00:52:32,680
came around one Microsoft technology, which one should it be on your laptop?
521
00:52:32,680 --> 00:52:47,160
Weany because it's a way that I can easily do calls to other people. I do have Zoom as well,
522
00:52:47,160 --> 00:52:55,720
sometimes because depending on what people like, to kind of the, the, the preferred methods of
523
00:52:55,720 --> 00:53:01,960
communication, that I think, Teams is the one which I use the most as like the external
524
00:53:01,960 --> 00:53:11,320
collaborations, so I would still like to have it. Yeah, and such a Nadalya comes to you today and say,
525
00:53:11,320 --> 00:53:16,600
I give you unlimited engineering, you source of money and so on, for one feature, what will you
526
00:53:16,600 --> 00:53:27,480
build? Oh my God, that sounds like, really interesting. I've seen one of the pain points of customers,
527
00:53:27,480 --> 00:53:36,760
as I can see, is having like a strong, like governance tool, even like Microsoft, is trying to
528
00:53:36,760 --> 00:53:43,160
plug the car, but like we show up on a dust management, agent physics, but still doesn't do
529
00:53:43,160 --> 00:53:49,000
everything you want because the clients might have like really, really custom requirements for the
530
00:53:49,000 --> 00:53:59,960
ones like to be like to police, for example, like a user created a lot of libraries, which doesn't
531
00:53:59,960 --> 00:54:07,720
have like, be, be corrected data or something, that could be like a simple use case, but again,
532
00:54:07,720 --> 00:54:14,840
is I would say like to build like a strong like governance, so policing agent to help
533
00:54:14,840 --> 00:54:19,240
clients do a better job to manage, to manage the tenant.
534
00:54:19,240 --> 00:54:23,000
Edroy, I finally or quote palette studio.
535
00:54:23,000 --> 00:54:30,040
Um, not everything needs to be an agent to publish,
536
00:54:31,800 --> 00:54:39,240
there's one other thing about having like a strong governance of policing agent is
537
00:54:39,240 --> 00:54:48,280
I'll be able to access space the signals and data, which is not, you don't really have
538
00:54:48,280 --> 00:54:55,720
already the available, that you need to use a little bit of automation in the background to
539
00:54:55,720 --> 00:55:04,600
get all those data, put it somewhere as a, as a central pull on information, have your agent
540
00:55:04,600 --> 00:55:09,800
reason over it, though I would say it would be a combination of both like automation and,
541
00:55:09,800 --> 00:55:16,040
and AI, it could be perhaps in foundry or copilot studio, because one of the things, like
542
00:55:16,040 --> 00:55:21,480
differentiator between copilot studio and foundry is perhaps if you use copilot studio,
543
00:55:22,280 --> 00:55:26,920
both potentially you will bridge like a limit where you can't do anything behind,
544
00:55:26,920 --> 00:55:28,840
and you might have to switch to foundry.
545
00:55:28,840 --> 00:55:36,520
Yeah, when I think people come to, to you, to the UK, what's the, what should they eat?
546
00:55:36,520 --> 00:55:37,720
What's the best food there?
547
00:55:37,720 --> 00:55:45,880
But so interesting, because it all depends, like I think the, I live in London,
548
00:55:45,880 --> 00:55:54,600
they, it's very multicultural, so you can get their cuisine from all over the world, like Chinese,
549
00:55:54,600 --> 00:56:03,880
Italian, Vietnamese, they're, love African, European, French, it's like, you're kind of very
550
00:56:03,880 --> 00:56:14,360
pampered, so it all depends like on what they would like to try, but I know, like fish and chips
551
00:56:14,360 --> 00:56:24,440
are very popular, and the, like again, burgos, hot dogs, but again, that's something very American,
552
00:56:24,440 --> 00:56:33,880
but, but yeah, so, so I would say it depends on what the person wants to try, and yeah, London is
553
00:56:33,880 --> 00:56:38,280
a good place, like to find anything you want, you don't need to go to any other country,
554
00:56:38,280 --> 00:56:42,680
you can try Italian, you can try Chinese, you can try French, and then we,
555
00:56:44,120 --> 00:56:50,440
yeah, the last time I have the, I wasn't UK, I got huggers.
556
00:56:50,440 --> 00:56:57,560
That's when you go to Scotland, yeah, possibly in London also, in my view places where you can get
557
00:56:57,560 --> 00:57:03,400
huggers, I won't say, like you dog get huggers in London, most potentially, yes, yeah, I'm never,
558
00:57:03,400 --> 00:57:13,240
I don't know, I just like, I never felt brave enough to try it yet when I went to Scotland,
559
00:57:13,240 --> 00:57:21,560
because it do have it as part of the breakfast menu, like if you go and stay in a Scottish hot dog,
560
00:57:21,560 --> 00:57:27,800
it do have it, and you're just like, oh, I need to be brave, it's a lot to encourage me to do it.
561
00:57:27,800 --> 00:57:35,720
But when I say, I was the more lucky guy because my daughter had seen this black pudding,
562
00:57:35,720 --> 00:57:39,080
she has tried this, she don't eat it.
563
00:57:41,240 --> 00:57:54,440
Yeah, so when you did deploying a co-pilot in the company, what's wondering, coffee, tea or red water?
564
00:57:54,440 --> 00:58:06,920
I would say tea because I don't really drink much coffee or red food, so I just like,
565
00:58:06,920 --> 00:58:17,160
oh, I just have a crow with water. Just keep hydrated, I would say, yes. But I think, yeah,
566
00:58:17,160 --> 00:58:22,360
the city on co-pilot is very easy, yes, it's just like trying to get ready for a roll out,
567
00:58:22,360 --> 00:58:27,720
which is a little bit more painful, a little bit more hard work to do.
568
00:58:27,720 --> 00:58:36,440
Then my last question is, yeah, you do a lot of community contributions. What is the, where you most
569
00:58:36,440 --> 00:58:49,800
proud of? I would say since October last year, I was invited to be part of a, to be part of
570
00:58:49,800 --> 00:58:58,840
a maintenance team for PNP Barcher. I've been contributing to PNP Barcher in terms of
571
00:58:59,880 --> 00:59:10,600
scrap samples as well as adding new components for quite a few years now and to people like to be part
572
00:59:10,600 --> 00:59:18,120
of a maintenance team to be quite close. It's something, I'm forever grateful, like, to go to them,
573
00:59:18,120 --> 00:59:24,680
I say to invite them to be part of the maintenance team and to people like you to work, like,
574
00:59:24,680 --> 00:59:32,520
we owe in, which is a further, alternative, our share, and Koon, it's just incredible experience,
575
00:59:32,520 --> 00:59:39,160
and we are just, I think, fantastic. I look forward to each, or our bi-weekly meeting, where we catch
576
00:59:39,160 --> 00:59:47,400
up, which is, yeah, it's something I'm really proud of. Yeah, then thank you, Rich, for being here,
577
00:59:47,400 --> 00:59:54,360
and spend the time with me and share all the insights. I really enjoy this conversation,
578
00:59:54,360 --> 00:59:59,720
because we moved beyond the hype and into the real engine during challenge of enterprise AI.
579
00:59:59,720 --> 01:00:05,800
We explore governance, co-pilot extensibility, multi-agent architecture, grass, and a little bit
580
01:00:05,800 --> 01:00:12,760
security and security and compliance topics. So, yeah, I think, yeah, there was a lot of big takeaways
581
01:00:12,760 --> 01:00:19,560
for the listeners, and yeah, thank you for being here, and yeah, I hope I see you in next time in
582
01:00:19,560 --> 01:00:29,480
another episode. Thank you so much for being with me as well, it's been brilliant talking to you.

Founder of m365.fm, m365.show and m365con.net
Mirko Peters is a Microsoft 365 expert, content creator, and founder of m365.fm, a platform dedicated to sharing practical insights on modern workplace technologies. His work focuses on Microsoft 365 governance, security, collaboration, and real-world implementation strategies.
Through his podcast and written content, Mirko provides hands-on guidance for IT professionals, architects, and business leaders navigating the complexities of Microsoft 365. He is known for translating complex topics into clear, actionable advice, often highlighting common mistakes and overlooked risks in real-world environments.
With a strong emphasis on community contribution and knowledge sharing, Mirko is actively building a platform that connects experts, shares experiences, and helps organizations get the most out of their Microsoft 365 investments.

Microsoft MVP
Reshmee Auckloo is a Microsoft MVP, Modern workplace consultant, speaker, and contributes to open source PnP repositories. She helps organizations adopt Microsoft 365 securely and responsibly by combining governance, security, compliance, and practical business outcomes. Reshmee speaks at technology conferences and shares real-world guidance on AI readiness, Copilot extensibility using Agents toolkit, Copilot Studio, etc...















