Key insights from the conversation

What are AI agents and how do they differ from copilots?

• Copilots are personal AI assistants grounded in your work data, helping with day-to-day tasks (one-to-one relationship)

• Agents are expert AI assistants for specific business processes, working for teams, departments, or entire organizations (one-to-many relationship)

• Agents handle ambiguity and complex reasoning that traditional automation cannot manage

The massive agent opportunity ahead

• Microsoft projects 1.3 billion agents by 2028 - just 3 years away

• 1 million agents already created in Copilot Studio and SharePoint in the last quarter alone

• Agents will solve business inefficiencies that human labor couldn't scale to address

• They democratize advanced automation, making it accessible to non-technical users

Key capabilities transforming work

• Computer use: Agents can operate virtual machines, click through applications, fill forms, and handle pop-ups autonomously

• Multi-agent orchestration: Specialized agents can communicate and pass work between each other (like a banking system with separate agents for savings, mortgages, withdrawals)

• Human-in-the-loop: Maintains human oversight and approval at critical decision points

The human-agent team dynamic

• Concept of becoming an "agent boss" - managing teams that include both humans and AI agents

• Optimal human-to-agent ratios vary: 1:1 for complex autonomous agents, 1:10+ for simple on-demand agents

• Transferable management skills apply: clear direction, knowledge sharing, performance reviews

Three pathways to get started

• First-party agents: Microsoft-built, ready-to-use agents (Researcher, Analyst) - plug and play

• Third-party agents: Pre-built by partners like Workday, ServiceNow, SAP

• Build your own: Using no-code, low-code, or pro-code tools depending on technical expertise

Best practices for agent development

• Start small and modular - avoid creating overly complex agents with pages of instructions

• Break large processes into specialized agents that can communicate with each other

• Focus on business processes that require reasoning and ambiguity handling, not simple automation

• Iterate and improve over time - agents won't be perfect initially

Governance and security framework

• Zone 1: Low-risk productivity agents available to all employees

• Zone 2: Business-IT partnership for department-specific agents with more capabilities

• Zone 3: Fully IT-managed enterprise-wide agents with advanced integrations

• Three pillars: Enterprise data protection, access control, and lifecycle governance

Overcoming adoption challenges

• People and change management are the biggest hurdles, not technology

• Start with experimentation and hackathons to build familiarity

• Normalize AI use and provide safe environments for testing

• Focus on demonstrating potential rather than building perfect demos for every team

• Remember: we're still early - only 40% of people use AI at work currently

Skills for the future

• Curiosity and experimentation become critical competencies

• Soft skills (communication, critical thinking) become more valuable as they differentiate humans

• Business process knowledge remains essential for successful agent implementation

• The "maker" mindset - technical capability without deep coding knowledge

Getting started and learning more

• Use Microsoft's agent community and learning documentation

• Start with personal, fun projects to build skills before applying to work scenarios

• Don't wait for perfect technology - begin experimenting now to build habits and expertise

• Focus on real business problems, not just technology hype

Full transcript

Here is the full conversation transcript, edited for clarity and conciseness. Here’s also the link to the conversation on LinkedIn live.

Roberto: We are live. Hello, Jack.

Jack: Hey Roberto. How's it going?

Roberto: I'm super excited to do this live with you. first of all, I want to thank you. for finding the time especially because it's 8:00 AM in Redmond, you are here about to share amazing things about agent and all these exciting things that we have.

Thank you so much for that. I'm sure it'll be a very exciting conversation and lead an introduction, about who you are and I'll be following you for a long time. And you are senior product marketing manager for Copilot Studio. you got multiple awards, one of which just yesterday by the way, congratulations both on the technical and customer side, which is one of the things I think is the beauty of what you share.

You have a deep technical view and also, very close to the consumer. And you are also father.

Jack: Yes.

Roberto: You create amazing content. I did a Google search, about AI agent and identity AI in the, in the Google trends.

And I saw an exponential curve into the search for this work, which by the way, a company, but the great content that you share, you have more than 80,000 people following everything you share. you will probably do many more things in this space. I'm super excited to be here and thank you again.

Yeah. Thank you for the intro. Wow. I think I might save that and then use that for every other future call. Again, thank you. this is just the beginning. When we were doing the preparation, we say, okay, we are talking about the agent opportunity. And by the way, for the people who is connected to the chat, we are seeing already 16 people. Hi everyone. Thank you for joining.

I will close for now the chat window. If you, want to interact with us, drop a question, we will go back to the chat later and, have a conversation. But to go back to the beginning, the first question, we are talking about the agent opportunity. What is an agent?

Jack: Yeah, it's a good question. in the industry, the term agent is very broad. I like to keep it as simple as possible At Microsoft we maintain two. AI terms that you probably see copilot.

And agents. it might be best to describe the two side by side to then help you understand how we're differentiate from the agent perspective as well. So we have copilot, which is your personal AI system. grounded in your work data, your you, your meetings, your emails, all that good stuff.

And it knows you, it's personalized and it helps you do your day-to-day work. And so co-pilot's there to work for you one-to-one. And then we have agents that are kind of your experts, AI assistants for your specific business processes. they might have their own instructions, their own knowledge and tools, but ultimately they're there to complete a specific business process or task.

they're working not just for you, but for a team, a department or even the whole organization. it's more like a one to many relationship. They're doing much more across the whole organization that's goal oriented.

Roberto: It feels much more versatile. you can do infinite things.

Jack: You think about the alternative of, if we didn't have AI agents doing these specific business processes, it's a human going in to do CRM, input records into CRM. update ERP. There's so many different business processes in a broad sense, and agents are there to help facilitate and manage those for you.

Roberto: Yeah. And so connected to that, we can start to imagine, we said infinite opportunities of collaboration, of doing thing. How would you define the agent opportunity, as was the title of this conversation? Why is it this big opportunity?

Jack: There's a lot of inefficiencies and problems in businesses that maybe you haven't been able to solve or scale with human labor I think agents will unlock a whole new way of thinking about how you can approach business processes and actually start to focus on the problem and allow agents to assist that.

And I think that's gonna open up a whole new way of problem solving. It's gonna open up a whole new capacity. We talk about the idea of, being able to hire employees or agents to improve that intellectual pool. I think it's gonna do wonders. It's gonna help you drive increased efficiency by bringing in agents to help you scale.

It's gonna allow you to, build business processes for problems you've probably never been able to consider. Because now where AI and technology's going, it can actually solve problems that can understand much more deep reasoning and thought as well. And then where agents are also going as well.

Every single piece of technology that we've seen in the industry from mobiles, internet laptops have been democratized to make them more accessible. And agents are gonna take that exact same path where we are building the tool so anyone can go away and build agents for their specific business processes.

And so you talk about that scalability as well. That's exactly where agents are gonna help. So kind of threefold there.

Roberto: Yeah. And, that's amazing. And actually, this is probably one of the first time when we're a con technology, we have first as a consumer and then in the company. We all use one or other form of ai, be it copilot be it chatGPT cloud or, or whatever.

And now we're bringing this to the organization and we have these amazing opportunities. I'm curious about this. what do you think? we can have a way to automate everything that we ever wanted to automate. This is like, okay, I have this huge backlog of stuff that I wanted to do and now I finally can do it.

And there is another side, which I also see, which is maybe, we could rethink the way we do the things. Not just automate, but do it differently. What's your take on this?

Jack: Yeah, I think naturally automation, we do a lot of automation today, right? There's the standard API automation and UI automation, and it's helping humans focus on what's most important while allowing you to Do the boring work automatically with standard automation agents are just opening that pool to then do even more business processes that traditional automation couldn't do.

Maybe it would get stuck because it couldn't deal with the ambiguity. now you will see agents being able to do more creative work as well. And so it's being able to help people build the right skills and maybe become generalists that can make a much broader impact for example, myself, I'm a product marketing manager that can understand so much more in the marketing process thanks to my agents that I have for messaging, for licensing, for anything else is allowing me to, and to kind of increase my impact as well.

Roberto: This reminds me of one critical skill, which I think it's essential for this, which is curiosity and experimentation.

Yeah. Yeah.

Jack: The soft skills, you think about like what skills are gonna be really important for humans. The soft skills is what makes humans unique. The ability to communicate, the ability to do all these like critical skills, et cetera, I think they're gonna become increasingly more important because that will help them be more effective with humans, but also be more effective with how they scale their agents as well.

And so I think being able to think about what does your career look like and the skills you wanna build those will be really pivotal.

Roberto: And absolutely. And connected to what you said about being a generalist. There's one thing that, I heard from one of your posts and one of the keynotes fragments talking about the agent boss. Can you talk a little bit about that?

Jack: The concept of a human agent team is, a bit weird, I guess if I explain my day to day work. I'm in a team, I have a manager, I have colleagues as well. And then I also have a number of agents that I've built that do specific marketing processes

we talk about the optimal human to agent ratio? Is it one to one? Is it a one to five? Is it a one to 10? And so each organization might have a, their own ratio. it really depends on what the agents are doing.

If an agent is, and I say they kind of sit across a spectrum from simple agents that focus on just answering questions on demand to fully autonomous agents completing business processes autonomously. if you, Roberto, had, five autonomous agents. You are gonna spend time observing, maintaining, iterating, giving direction.

there's a bit more of an actual like management role there where you're having to run reviews, give them guidance, make sure they're doing the work. But if they're just simple agents, you could probably have 10 to 15 because they only work when you ask them to. so there's not that constant oversight that you need for them.

Roberto: And so you can see how like the human agent ratio will kind of depend on the complexity of what the agents have Wow. this also, connects straight with one thing that I saw in the keynote, which was the capabilities that this agent can have. And you mentioned the computer use, so this is something that completely blew my mind.

Jack: the technology is moving so fast. whenever I talk about new features that come out like computer use, I like to say, what was the problem statement before and what's it solving? I think that's really important because it's not just technology for technology's sake. It's actually real meaningful impact, that these business that they can help with business processes.

if you've built automation, maybe you've used tools like Power Automate, they're perfect for following a sequence of steps. And focus on working with applications that are in the cloud. They have APIs and so you, you can inter interface with them, but what if you wanna do a business process, but it just doesn't have an API to connect to.

Today's approach, you do something called RPA, robotic Process Automation, and it's essentially teaching a computer how to do these set of steps. So the automation would execute on a computer and say, follow these clicks, click round the screen. But then what would happen is, what if you had a pop-up?

And so the RPA is only trained to follow these steps, and if there's a popup, it can't deal with the ambiguities computer vision, uses generative AI to process what's on the screen, understand the instructions given and pivot to deal with that ambiguity to complete the workflow.

And so it's pretty impressive. They can spin up virtual machines, do all of their work, clicking around, copying data, filling in forms, et cetera, and then pass over to the agent to finish off the process.

Roberto: and what you mentioned now we are talking back technical and when you talk about virtual machine, for someone maybe who doesn't know, I saw the demo I tell a person that every time I send him or her an email, what they have to do is to take the content of the email to go to her website, fill a form and other row to an Excel and whatever it is.

they. Create their own computer. They do it on their own. They don't need to have you the computer on, which is great because I, I'm not depending on my computer to be online And then they come back with a result to validate, okay, this is what I found, this is what I do.

Give me an okay. Or an okay. It's like having a person. Actually, when I started working with copilot, I started as an intern with 100% of time available infinite resources, but they needed very clear direction and step by step guidance. And when we had the prompt, it's exactly the same, but much more advanced.

Jack: That's why I talk about the concept of the agent boss because transferable skills apply anywhere. Like if you were bringing an intern or a new colleague on board, you give them a lot of direction. You give them the knowledge they need. It's similar, to agents. You give them instructions, you give them , the data that they need to help 'em do their work.

And so it is funny when you said about the capacity side. We did the work trends index recently and brought out a report about the future of work and the idea of a frontier firm. It's a firm that's AI first in their business processes. some of the research showed that 80% of the global workforce didn't have the time or energy to do more.

Roberto: And I think that's where agents really start to fill that gap, by being able to help them scale where they haven't been able to scale before. yeah. It's like a whole new capability that we didn't have. And I'm curious, we are talking about rate of agent to human. It's very early probably to give estimate. But anyway, what's the estimate of the number of agents in the near future?

Jack: Oh. actually, yeah, that gives me a good point. You just reminded me. We worked with IDC and we're projecting about 1.3 billion agents by 2020 1.3 billion. Yes. not long at all, three years away, we expect to have that many agents. It shows

how we're investing in agents significantly. just to give you another, momentum stat, 1 million agents were created in copilot studio, our low-code agent tool in Microsoft and SharePoint in the last quarter. So 1 million total across those two technologies.

Yeah. And so you can see the pace is gonna accelerate exponentially as organizations start to deploy, build the right skills and start to see real value over that as well. So yeah, you can see the momentum.

Roberto: And then maybe for someone who hears this, okay, this is huge number, but I think there is one, one very important point because an agent has not to be a huge process. what I heard from one of your posts is that it can be very modular. So you can have an engine for every single task and then combine them.

How could this work?

Jack: Yeah, that's a really good, it is a best practice that we're trying to emphasize as much as possible when you first start building your agent, the number one advice, I say most people just start to write this long list of things that they want this agent to do, and they'll write like a CV, pages and pages.

And then they'll send it to the agent and they'll go, why can't you do the work? there's so much complexity there. If you give that to a human and just went, do it, there's so much ambiguity. They needs to understand the question. They need to understand the intent. the concept of the multi-agent, the multi-agent or ATA has come out agent to agent, that's our solution to helping break up enormous agents into specialized agents that kind of focus in their experts in this specific business process.

But the most important part, they can communicate together so they can pass context, they can share information as needed to complete a much more round process without having to have so much instructions that there's kind of overlap, there's conflicting, So that's really our best practice approach for that.

Roberto: So in a way, it's something similar to what we did in the early day when we were talking with LLM. You do not, you don't give them the LLM a huge distraction. Maybe you separate in pieces, but here we have the bonus that once we do it, it's saved. So you create your agent for a specific function of a job, then you have it and you can train it, you can improve it.

And what I also hear from you is that everyone in the organization can use that piece.

Jack: Yeah. You can start to share that across your organization so they can work for multiple people. and the whole team as opposed to just one person. it is incredible where the technology's going. one thing I'm excited about is multi-agent and agent to agent.

These are open source technologies that multiple companies can use. So agent brought, Google brought out agent to agent. Microsoft is using some of that to help show that integration. agents across different platforms can talk together as well. maybe a Microsoft agent can talk to a Google agent as an example.

Like that is, is really important. 'cause then it doesn't mean that customers are locked in. They can use the agent tools that work for them.

Roberto: Yeah, exactly. and what I hear is that the more people use it, the more the value, like the network, if the internet was only between two people was useful. But the more connection we have, the more the value for everyone. So it's amazing.

Jack: Exactly. one example I wanted to give just on that multi-agent, 'cause I, you kind of referenced it. At Microsoft Build we showed a demo, of a banking agent think about the different services a bank may offer, like savings, mortgages, withdrawals

each agent would be an expert in those different parts. So you speak to an agent, say, I'd love to take out some money, please, goes to the savings agent to help manage that. Actually I want to explore buying a house and I'm looking for a mortgage. So it passes it over and transfers you to the next agent.

So it is just the seamless experience for the end user, but behind the scenes it's all managed coordinating and orchestrating

Roberto: Yeah. and I can also imagine that the human can intervene at any point because if something, goes another level, or maybe if there is a rule, you can put it in place. So you can say, okay, I will automate everything, but then keep this human interaction. This human touch is not 100% automated if you want, so you can always choose.

Jack: Exactly that goes back to that spectrum of agents that I just spoke about. the simple agents that are on demand answering questions. Humans always in the loop because the human's asking for the work and they're just doing it for you. So you can review the content and then approve it and go as needed.

But in the autonomous world, human in the loop is so important 'cause you wanna see kinda what's happening, why they're thinking this way, why they're orchestrating. the example I always give is that we worked with a retail pet company in the uk, to build one of their first autonomous agents.

And this was for fraud protection. when a customer buys a product, and maybe they're trying to get a refund for a product they haven't bought, or, any of those suspicious activities.

How do you scale a customer service team to review every single case? It's impossible to have the human labor to be able to do that. And so having that agent do that first step of understanding that whole process review and the images, et cetera, and then providing it to a human to make that final call with the human in the loop as an example, it's helping them scale significantly.

I think that's a key part to think about.

Roberto: And now that you mentioned a real use case, which sounds amazing and gives a scale and another dimension, how do you start? Where do you start? If you are never done anything with agent, what would you bet on it?

Jack: So at Microsoft we have this three phased approach to how you can start with agents. We say, you've got first party agents that Microsoft builds. we have engineers building products and agents just for you. you can use them, deploy them out of the box and start seeing value quickly.

Plug and play. Got researcher, analyst, many other agents that you can use and you can just see value. They're all managed, the instructions are managed for you. So all you do is give them a prompt and they start doing the work. And you can do some pretty incredible things. I dunno if you've managed to use research yet.

Roberto: Yes. And it's amazing. I was very positively surprised

Jack: The depth of review, the ability to use your work data to inform that as well. I've done product marketing strategy reports and market research so that's the first way, really simple, low barrier to entry. Just get started using those first party ones.

Of course there's then third parties that we work with that popular tools like Workday, ServiceNow, SAP, are building agents you can then plug into co-pilot and use from that experience, taking advantage of those ones. And then some organizations say, I have unique business processes.

I don't work like the average organization. I need to do this step before every other step until they go and build their own. And so they need to look at what tools are available for them to build their own agents as opposed to the stuff out of the box. What I think is special about Microsoft is that we offer the spectrum of tools to meet the makers where they are.

So whether you are no code, you just wanna build an agent with a conversation like, hi agent, I'd love you to help me do it or low code where there's a bit more drag and drop, a bit more visual. You need to bit under, make a, you'd have to be really technical, but you need to have a bit more of an understanding to how these things work all the way to Pro Code.

So like no code, low code is Copilot Studio Pro Code is, GitHub Visual Studio. More of the traditional pro code experiences as well. So wherever you are, you can go from all of those.

Roberto: And looks like we have complete flexibility and there is one word that you mentioned, and I love it, which is the maker. What's the definition of a maker for you and why this are such an important role, and especially today?

Jack: So I've been in the low code space for five plus years, and I describe myself as a maker because I don't write code. I don't understand the complexities of that side, but I am someone that wants to make a difference. I'm someone that wants to make an impact with the technology that I can use and I want to show how I can build solutions without the technical depth.

And so I think it's someone that's technically capable, without having to code to be able to build solutions that help a business or help you do more essentially. So, it's quite an ambiguous, term. I guess the more well-known term was something like a citizen developer, someone that's not a core developer, but just wants to build solutions and get hands on.

Yeah. Curious, another skill that you mentioned, like being able to learn and iterate as well.

Roberto: Yeah. And beautifully connects with what you said in the beginning that we don't have enough people to do all the things that we want to do. Even with AI, even with agents, and it feels like much more accessible. So you don't need to be an expert in coding. For example, my case, I've never programmed in Python and in the days I programmed in visual basic, SQL, I love programming, but I thought, okay, I miss the train of Python.

I cannot do it because it's too huge investment for me at this point. But then when chatGPT came out, I say, wow, I can finally program in Python and that night I couldn't sleep actually, because I say, how think can I do that I ever, always wanted to do? But then of course you have limited time, so the tools will be better and better and better.

And finding this mix between the completely low code and where you want to go deep.

Jack: Exactly, and that interoperability as well. they're not in silos. It's not just like, here's low code, here's pro code. You build separately and don't speak to each other. It's how can you integrate the two and work? I talk about the idea of a fusion team. you're in a company, trying to build a solution.

You've got the low coders closest to the problem, the business users, and then you've got the pro coders and the it the real technical people that build the more complex solutions you can work together. the example Copilot Studio is low code.

We manage the out of the box models for you, but actually what if I want to fine tune custom model from Azure will go get the pro developer to build that custom, model and plug it into Copilot Studio. So it's that perfect world where you kind of all build together.

Roberto: Yes and going to the people also it's much more fun because now it's not it and the business, so it can be closer to the business and the business people can be closer to it it's not rocket science. You can do little experiments. you can speak a common language and then you can build things together, which is much more fun.

Jack: That's one of the things I always hear is that people wish they had better lines of communication between IT and the business these tools really facilitate that and help you be more successful.

Roberto: I think this is one of the pieces, not that the agent, but in general the AI opportunity of creating this bridge between IT and business people also, because we are human. We have our lives and we want to know this technology. So both sides, are interested and we are all in the same. it's not one or the other.

Jack: Yeah. the point you mentioned at the very beginning really resonated with me about the idea that these technologies became consumer first before they went to the enterprise. That tackles one of the biggest challenges when you think about bringing new technology into an organization.

It's the people, it's the change management, it's the understanding the value, it's the training and enablement. It's so that by having consumers, the average person that just wants to go on and do this outside of work and use chat GPT and all these other fantastic tools, you're building the skills and then you can apply them to the business.

I think that's a really great way to think.

Roberto: Yeah, absolutely. I read recently a blog post from Ethan Mollick and he was saying that we have huge tools in our hand, but still we are not realizing the potential.

only 40% of people are using AI at work or say they use AI at work. Even if the productivity of the people may increase, the organizational productivity is not yet made a leap.

So we're still very early.

Jack: That's good to know. I think it's the habit forming as well, when I first started using it. I had to think AI first when I was trying to implement that into my day-to-day. 'cause naturally, it's, let's go, just put my heads down, let's start doing the work myself.

And then actually using that as a sounding board, as a brainstorm, as a anything else to help you do your work net. And now, every, every day, being able to just jump to it. If I have a question, if I want to validate something, if I need, just need some inspiration, I've now built that habit. And I, I look to that as an assistant, to help me.

Roberto: Exactly. 100% agree with you. And actually now we have two questions. Can I do this with AI and personal assistant? And the second is, is it worth to build an agent to do it if it's a current task or how can I integrate this in the tools that I already have? So it's, making this muscle and everything that we think, see how can we do this in a different way better.

Jack: Yeah. One other question I'm gonna add here, which is should it be an agent as well? Maybe that's the third question. I think naturally agents, the term in the industry, it's being used for nearly every single description right now.

And so I think the real honest question is, does it need to be an agent every time. I speak to a lot of organizations and when they describe their use case, it just sounds like traditional automation, like traditional AI automation, that doesn't need to be an agent. It's just, Hey, we have these set 10 sets of steps.

They will always follow the same 10 set of steps. They don't need to do anything else. There's no ambiguity. And I'm like, okay, well that's not an agent because it's not dealing with the ambiguity, it's not dealing with, doesn't need deep reasoning, it just needs to follow these 10 steps. So just go for your automation path.

And so just knowing when and why, it's an agent is really

Roberto: Yeah, absolutely agree. And thank you for naming this because we have a lot of hype. Of course, we are excited. But there's a lot of hype because it's an exciting technology and one thing connects to this, which is how important it is the business and process knowledge of the people, because it's not just about the technology.

The people who may be using that 60% that didn't use AI or not so much, it's very early. We're not late because still the business knowledge is so important to make this decision.

Exactly, and you'll find that the people that understand the business processes uniquely will have more successful agents and will have more successful outputs. So that is still a continued value of a human, is that they truly understand how the business works and they can share that expertise. Absolutely. Time flies. It's already half an hour. Lemme lemme check if you have some question in the chat. hi Ami, I see here, Laman. it's a very, good question from blacks, man. He said, with so many quick changes happening in AI how can an enterprise adopt it because by the time you implement something, you have a new, better technology

what would be your answer? Jack.

Jack: Yeah, I guess it depends on the type of technology that you are using. If you are just trying to build out your own or just trying to use LLMs, you're seeing that every week a new LLM comes out, but the pace of change and the differences will move so fast, tools like Copilot Studio that we use, manage the models for you, so we're continually iterating and improving them behind the scenes, and so that kind of future proofs you, that you don't lose everything you've built in the past while still taking advantage of the latest

Considering the tool that you're using, if you're just doing LLM development and adding some prompts on top, The LLM models will move very fast. you'll see different iterations in modality speed and token size, from an agent perspective think about how you can incorporate the latest features that get added to a single platform.

And so you're still taking advantage of them, but you're doing the right thing. The best thing to do is to get started, like start to think about iterate, learn fast, and it's only gonna get better from there as well.

Roberto: Yes. to your point of getting better, which is absolutely true, I will add one thing, even if it doesn't work today, don't throw it away when someone asks me, can copilot do this? Can copilot do that? And I answer, not yet, but keep it.

Please keep it. And in three to six months, try it again. Because maybe you'll be surprised as I was surprised with the researcher when I tried something and say, oh, wow, now it, now it works. This thing didn't work before. Now it works.

Jack: Exactly, and the troubleshoot inside as well. I think it's knowing the expectations of what to expect from these technologies. they're never gonna be a hundred percent perfect, they can't be a hundred percent perfect. They're trained on human data where humans make errors too.

Agents will fail at times and you have to know that you can continue to iterate. They're not gonna be perfect the first time. They take time. You think about when we used to build and develop AI and conversational AI five years ago. You as an organization would spend 12 months training a model and spending a lot of money to get to a point of a prototype and seeing value.

And now we're in the world where we're so lucky and privileged to have these sorts of tools available that when we try them for the first time, if they're not a hundred percent perfect, we give up. And so being able to iterate know why it hasn't worked, is it because the data wasn't good enough in the first place?

Is it because the LLM instructions just weren't accurate enough? Like being able to be persistent and iterate is a really important skill. And I think that sometimes people don't give it enough chance. And so I think that's one of, one of the key takeaways is continue to use and improve over time.

Roberto: Yes. To add to your point, someone could say, okay, I will not do anything. I will wait infinite time because the model will be gather better, so I will just not do anything. But on the contrary, while you are experimenting with the initial version, you gain knowledge and you mentioned the people.

The people get used and they build this habit, which then they will help increase the speed of adoption when the new model or tool comes out. Absolutely.

Jack: Exactly. And I love the passion for companies to be a frontier firm and be on the cusp of innovation using the latest models. But we also have to understand that many organizations aren't there. Like they're still at the early stages. Some companies still use AI tools for search replacement.

How can I use this instead of traditional search engines? Understanding how you take a company from: I will use an AI tool to help me search for content and information, the them actually autonomously doing business processes. That is a handholding journey and enablement journey for a company, and it still takes time for them to catch up.

So I think something like myself, I really take the job on that. I will absolutely talk about the latest and greatest technology that comes out. It's exciting, it's gonna make a huge impact, but we also have to continue to talk about what's available today that isn't in a research preview that you can use and are generally available and you can make a difference of.

Roberto: Exactly. Thank you. Let's see if we have a few more question. Where to find a readily available agent developed by Microsoft, which an enterprise can use. Laman is asking. Yeah I think she hear first to the one that you mentioned out of the box.

Jack: Out of the box. If you have, Microsoft Copilot or Microsoft 3 6 5 copilot, so this is for your work AI at Work essentially experience. You'll notice that on the left when you use copilot Chat, you have the all agents button, called our Agent store, in there you will see managed agents that we offer first party out of the box, like researcher, analyst, and then you'll also see third party ones, like SAP, et cetera.

And then you'll see a button that says Build your Own Agent with Agent Builder. So you've got the three stages I spoke about all available in that one agent store.

Roberto: Oh wow. That sounds very interesting. And I'm curious, there is one question that probably will arise when you talk about third parties and integration, how does, the governance and security, especially in company, comes with this?

Jack: Yeah, so I mean, I can talk from a Microsoft perspective as well, but there's kind of, there's three pillars that I think are important. The first one is enterprise data protection. just making sure that the way the models you interact with, et cetera, aren't training the data you use in the underlying models, that they're encrypted across that journey.

And all the key important things like responsible AI safeguards as well. So being able to manage and filter out some of the negative uses of ai. So that's the first one. The second one is access control, making sure that. When you are building agents or using agents, the it are in the front, in the driver's seat of who can create, who can share, who can use, and then being able to manage, manage and monitor agent performance.

So there's data, access, and governance. How do you manage the process of an agent being created from creation to retirement, set policies of what data can be used, audit that and have all the operational insights as well. So there's those three stages.

I'd love it if you're interested to share how organizations are actually applying this in the day-to-day

Roberto: Yes, very much because the first step comes by default because it comes with a tool. But the second and the third depends on the organization. So it's human. They have to implement that.

Jack: Exactly. And so those three pillars that I just spoke about will vary depending on if it's a first party agent fee, when it's a build your own. Because naturally if you're building your own, there's a few more, bells and whistles, you want to change and iterate as an IT person to make them more relevant to you.

How we apply this even at using agents in our company, we use a zoned governance model. And so the idea is that there's a three step or three zoned areas of how agents will be used in the company. The first one is zone one, which is really focus on how every employee in the company has access to agents for low risk, quick win productivity based scenarios.

So if you've used the Microsoft 3 6 5 agent builder experience today, it's like it's really easy to use. You ground it in the data that you have access to today, and you can just start seeing value quickly. I've built tons of those that I share with my colleagues now.

The research on the analyst, everyone can use them. Very safe. free for everyone. Exactly. so as these agents show increased potential or need more connectivity to systems that maybe you wouldn't give to every single employee, they progress up these to zone two and zone three. And so zone two is about partner development between the business user and the developer to build an agent specific for you.

There's more governance and control, but it's given to a specific, more granular audience that can use and build this application. And then all the way to zone three, which is fully owned by, What would be an example of, an agent that crosses the threshold and needs to go to this stage? Yeah. So let's say, a market research agent. I've built a really simple agent that does some market research on agents. I share that with my colleagues and we all see value, which is great, that use case could be used across any product marketing team in a company.

maybe I'm gonna upgrade this to a zone two, where I can work with the IT team, some developers and add more advanced capabilities to that market research agent. Like deep reasoning, maybe compile it all into a report for me.

Roberto: okay.

Jack: so we do that in a more secure environment. And then so that I could share that to a couple more teams. But if that agent showed potential to be relevant for every single employee, I'd give that to the IT team and they would manage and build all of it themselves, that'd be zone three, where they democratize it for use

no one could build it, but everyone can use it.

Roberto: So you can see as the potential increases, as the access to advanced tools increases, it becomes more IT owned. But if it's lower risk, the business owns, and in the middle they partner together, as needed. It's the perfect model and many companies use this today, to allow that upgrade path. Yeah, as you mentioned, when we talk about low code or full code you start with something very simple, then you can add one thing, then you can progress a little bit. Then you see how it evolves. Wow. Super.

Jack: Exactly. Applying that back to the governance and security we spoke about, it can implement rules and processes to help automate that. They can set up policies. A business user could submit a report and say, this is why I want the agent into this new environment.

These are the data policies that I need. Then you could use application lifecycle management to automatically move the agent into this new environment, allocate the new policy. So you can really start to get clever from an IT perspective on if you're managing thousands of employees, how can you scale your IT operations.

Roberto: Yeah, exactly. As we mentioned before, we can maybe have one single agent that works for everyone and not replicate the same agent in every single team. You have to think about if we have to create an agent for this, maybe it's not an agent, maybe it's just pure automation of the process.

Don't create an agent if you don't need that. instead of creating 10. 100 or whatever, doing the same thing. I can imagine myself some use case that I would build on my own. Then maybe some calling would also need, so, yeah, absolutely. The business area has to be in contact with IT, so they have to have a dialogue constantly.

Exactly. One of the big, processes IT will manage is inventory management. Like what's been built across our company. Is there opportunities to consolidate and do something really well? If 10 people in the company have built an onboarding agent, maybe we need a company wide onboarding agent that it and pro devs can manage.

Jack: That's why we have that zone model of the perfect upgrade path.

Roberto: I can imagine how in this 1.8 billion, you can go crazy if you want and create thousands of agents. And then how do you know what's the good one like in the documents? I have version 1, 2, 3, 4, 5. What's a good one? No one knows.

Jack: Yeah, exactly, and I think that comes down nicely to the second part, which is knowing what the agents are doing, having the right analytics to know how and why they're performing and the usage, but also then translating that to ROI like the, that connection, as you kind of mentioned, it's really important for organizations not just to look at activity, but the impact of that agent.

How does that really contribute to the bottom line? How does that help reduce costs? How does it generate new revenue? That's the bit where I think companies need to spend more time on the attribution to the real business impact, as opposed to, Hey, this agent has saved me five minutes a day. But it doesn't really contribute to the broader business goals and outcomes.

And so I think that's a mindset shift that we have to think.

Roberto: Yeah. I can see how that can be a challenge. I'm curious now what can be the most significant challenges that the company face when they say, okay, I want start as early as possible, but also do it with some, criteria. And do it well.

Jack: I think I only have one example, which I see happen the most, and it's not necessarily the technology, as someone mentioned in the comments, it's not the speed of innovation that is going incredibly fast and will continue to iterate, I think with every single project in a company if I was doing a new CRM implementation in my company, the number one challenge is the people.

How do you help them with the change management? How would you help 'em understand the why, the value? How do you help them understand the new ways of working and how do you get them bought into, using it in the day to day? I think that's the number one right now. And so we spend a lot of time helping people ideate, do hackathons to start to brainstorm agents that work just for them.

How do you do, showcasing, when you do lots of different demos to your different departments to show them the potential? And one of the piece of feedback that I've had when companies try to drive adoption in their company is that users can get really specific about, if this doesn't work exactly, just for my scenario, I'm not going to use it.

So the demos, it's impossible to build a demo for every single person. If I was it and I was trying to show potential in HR or in IT and finance, in sales. You need to show the potential of the demo, not build them specifically for every team, because you won't be able to scale.

And then you'll be stuck in this analysis paralysis where you've kind of got this massive inventory of potential, but not actually building any. That's another challenge that can be overcome through enablement. Open communication, leadership communication is very important as well top down about why we're doing that, what's the value, how it's gonna help us be more competitive, and all those things.

Roberto: And one thing that also come to mind, which is normalizing the use of ai, because sometime maybe someone heard, okay, be careful what you do, because okay, I'm not going to do that. But everything we put in copilot is 100% safe.

We set up the incentive to use it. We've set up space, we give you, time resources just to experiment. And then there's one thing because we are all figuring this out, as you said, I show you what you can do with this technology, but then you as a business user know your process.

It's your turn to try and see how it work because no one knows how to do it.

Jack: Yeah. Exactly. when I do demos for customers, I always showcase three different steps of what these agents can do. If I was speaking to HR or finance, et cetera, I couldn't go away and build each individual agent. 'cause I don't know. I'm not the closest business problem solver to finance or hr.

I show three things and it usually sticks with them. The first demo I always show is an agent's ability to retrieve information. being able to answer questions over your data. That could be any type of data, and this is how I apply it to them when I speak to 'em. So you're in finance.

This could be your finance processes, documents. You're seeing how this agent is able to look over data and retrieve. You want agents to do tasks for you, right? You don't want 'em to just answer questions. You maybe want 'em to complete workflows and start some automation.

Here's this agent that does it for suppliers, but that could be for your financial approval process. It could be for the onboarding process. Then finally, the first two are more chat experiences. What if this agent could be more autonomous in nature? Are there processes that you just want to happen behind the scenes?

This is how this could apply to your business. getting them to have that spark and then get the in the hands of them that know the problem. To start ideating is good.

Roberto: Yeah. It's like having the pieces of a Lego that you build whatever you want, You don't have to use the same Lego piece for everything, but then you have first thing what you want to build and if you have the appropriate pieces.

Jack: This isn't the first time that we've been through this sort of technology, evolution. The technology is very unique and innovative, but the process of applying a technology like low-code technology like agents, we've done the same with, application development with products like power apps.

We've done the same with automation, with Power automate. We've done the same with dashboarding with Power Bi. So all of those learnings of how do you take a low-code tool and scale it in your business is there today. It's very well ridden and people know a lot of good learnings from there. And so we can apply that to this agent adoption and learning journey as well.

We can use all of that goodness to help, improve the process of building agents.

Roberto: And now that you talk about learning journey, if someone is curious and want to learn more, where to start?

Jack: Yeah. So you referenced Ethan Mollick even he, struggles to keep up with the pace of change because there's so much that goes on. There's announcements all the time. The best part is think about the organization or the tools that you use today. Look at the learning documentation online. There's some incredible YouTube, playlists as well on understanding the fundamentals.

So I think it's kind of depending on who you are, are you a, system developer? Are you a full on developer? depending on your audience, find the adoption pages of the company. So Microsoft has some great agent adoption websites that has simple learning content, some use case inspiration as well, some training development and hands-on demo videos.

So just search online the product name, learning documentation and that usually has a lot of the great stuff. But they also look at some of the great creators online as well that are actually creating some of this independent content, because they're doing and they're really learning fast and sharing their experiences.

I'm sure you've seen lots of white papers online on agent adoption practices and all that good

Roberto: Actually it's impossible to keep track of everything one thing I would add when you mentioned, sharing is the community, which I, I 100% believe that having a support of a community, internal, external, so critical because you're not alone, as you mentioned, I'm not in finance alone wanting to pull data

Many people want to pull data and, but it's different data. And having community, it can be a power support. And actually I just saw today during prep that there is a Microsoft community for copilot studio and agents.

Jack: There is great plug. The Copilot Studio Agents community, it's a single place where we have loads of different agent contributors that have built content. So you can click on their profile, see what they've created. And we've also consolidated all our latest agent content as well.

So. A really good place to start. The one thing I just want you all to do is just start small, build that as we talk about habits, learning and spending time to build these skills. these are skills that are gonna help you in your future. I talk about the idea that the future of recruitment isn't just, this is my cv.Look at the agents that I've built and the impact that they've made. These skills that you build are gonna help you in the future of your career as well.

Roberto: Exactly. I would add also one thing, which is having fun because we're such an exciting moment. We can build anything. We don't have to build anything, but we can almost anything. So if it's not, it's not yet. So who knows what we can build in one year, in two years? Just start now and then you'll see. Absolutely.

Jack: Yeah, that's really good, when I do the demos for these different functions and showcasing the potential, I will showcase a general scenario that shows the possibility of the technology. But when the user has fun and applies it to something that makes it relevant to them, like, go build an agent about you love cats.

Go build an agent that answers questions about the different cat breeds and does something else. when you make it really personal to you, you start to have so much more fun and explore the capabilities.

Roberto: Exactly. And then you transfer that capability to other domains because it's the same, power, and then you share with others. Of course it takes time. So again, 100% with you. When you say just start, don't wait We are not late still early.

Jack: Exactly. Like the other example, MCP, the Model Context

Protocol is a new protocol that came out to help with data integration for agents. a colleague was working on announcing MCP and copilot studio. They built a simple MCP that anyone could use that just told jokes.

Roberto: It showed that you, an agent could easily call out these servers that could then take data and then search and retrieve and index and spit it out to the appropriate users. But it's a fun way of doing it. Exactly. It's a brilliant example. And by the way, for people who don't know, as I didn't know until, very little time, what is MCP?

Jack: Oh yeah. Model, context, protocol. in the simplest fashion, back to the, what was the problem statement? People that would build agents today would have various different approaches to how they bring data into their agent. Some would maybe use retrieval, augmented generation to be able to do search over that data.

Some would use APIs. MCP was a new protocol to help you search and bring in data in a more consistent manner. And that MCP approach can be used across many different platforms as well. So not just Microsoft. It could be used Google, it could be used in other third parties. It originally came from Anthropic.

And so it's just this consistent way. No matter what tool we use. If you built an MCP server, you can take advantage of it across all these different platforms. back to that idea of open sourcing, this sort of technology, it kind of levels the opportunity for everyone.

Roberto: Exactly. And as we mentioned about network effects, it's the same if more people can plug more data, more value for everyone, more adoption, more fast innovation, and it gives bigger and bigger. Absolutely. Time flies. I see. Maybe we have time for one more question before wrapping up. okay. I will go jump with the last one.

From Laman, being a Microsoft 365 architect, what's your guidance to grow more on the copilot and agen ai? Are there any certification in place? Interesting on training.

Jack: Good question. I'm not sure. I don't have the specific list of, the different accreditations that you can do today. Microsoft Learn is a great place to start using and going for each of the free courses. So there's lots of free courses to use and learn from there. I can imagine that we will have and bring more accreditations that's needed from our learning team.

Check out Microsoft Learning. It allows you to filter by roles. So if you're an architect, it also suggests which roles are relevant for someone that's trying to build co-pilot skills as an architect. check out those learning paths. they iterate from free learning all the way to accreditation.

So I need to check if there is one available specifically for those different tools. I can absolutely add them in this chat as needed.

Roberto: Beautiful. I think we could go on for two hours more. Still you are just beginning your day. Remember it's 9:00 AM in Redmond. I would like to ask you one more question, Jack, to close this beautiful conversation. And I'm very curious about this. And for me would be my question.

Jack: What would you like to see more and less? Yeah. I would love to see, oh, I wasn't expecting this question. I just wanna see more people be problem solvers and be willing to put themselves out and try these new tools. don't be scared of failing. just try, innovate and see if something works for you. we are at a really interesting time in the technology industry, once in a lifetime.

There's maybe been a handful of times where we've gone, this is really gonna change the way that we work and live. take advantage of these agent opportunities as much as possible. Challenge content you see online don't just assume it's a hundred percent correct and validate it yourself and see if it's applicable to, to you.

There's a lot of hype around technology and agents focus on what's truly important and the real opportunity and don't get distracted by the hype and, noise that's something I'd take away.

Roberto: 100% it's like two sides of the same coin of the experimentation. You do stuff and you make your own opinion. So you have your own.

Jack: Your opinion is important and it's important to vocalize that when you see a lot of content online.

Roberto: Exactly. Don't give up agency. We keep your agency, we decide what to take, what not, and what to experiment with. Beautiful.

Jack: exactly.

Roberto: Thank you so much, Jack. It was super fun. And, thanks. You also for the people who are connected who asked questions.

Thank you. for experimenting with this, LinkedIn live format. I'm looking forward to seeing more of your content and I will experiment, a lot.

Jack: Sounds good. you have to let me know how you get on. Thank you so much.

Roberto: Bye everyone.

Jack: See you later.

Where to find Jack and his work

• LinkedIn

• YouTube