Unlock the Potential of Copilot Studio - Power CAT AI Webinar

00:54:16
https://www.youtube.com/watch?v=UJBx_zAd3Fg

概要

TLDRThe webinar focuses on unlocking the potential of Microsoft Copilot Studio for building AI agents that can transform business processes. It features speakers Ashley, Pav, and Vasvi, who discuss the importance of identifying specific business problems, potential impact areas, and the necessary data for successful AI solutions. The session includes a demonstration of creating various types of agents, such as retrieval, task-based, and autonomous agents, using simple instructions and integrating with existing data sources. Real-world examples highlight the significant ROI and efficiency gains achieved by organizations using these tools. The session emphasizes that even non-technical users can build effective AI agents, making it accessible for a wide range of business applications.

収穫

  • 🤖 Microsoft Copilot Studio enables easy AI agent creation.
  • 📈 Businesses can achieve significant ROI with AI agents.
  • 🔍 Identify specific business problems to solve with AI.
  • 💡 Non-technical users can build agents using natural language.
  • 🌐 Agents can be published across multiple channels.
  • 📊 High-quality data is crucial for effective AI solutions.
  • 🤝 Collaboration enhances AI solution effectiveness.
  • ⚙️ Autonomous agents can automate tasks based on triggers.
  • 📚 Real-world examples demonstrate successful implementations.
  • 🔗 Integration with existing data sources is seamless.

タイムライン

  • 00:00:00 - 00:05:00

    The webinar begins with introductions from the hosts Ashley, Pav, and Vasvi, who express excitement about the session focused on unlocking the potential of Co-Pilot Studio.

  • 00:05:00 - 00:10:00

    Pav discusses the agenda, which includes how businesses are transforming processes with AI agents, the tools Microsoft provides for building these agents, and a demonstration of agent creation.

  • 00:10:00 - 00:15:00

    The importance of prioritizing use cases for AI agents is emphasized, with three key questions to consider: the specific business problem to solve, areas of high impact, and the availability of necessary data and stakeholder buy-in.

  • 00:15:00 - 00:20:00

    Pav outlines trends in agent building, starting with simple retrieval agents that answer questions based on data, moving to task-based agents that perform actions on behalf of users, and finally to autonomous agents that can trigger workflows based on external events.

  • 00:20:00 - 00:25:00

    Microsoft's Co-Pilot Studio is introduced as a platform for building agents, highlighting its integration with Teams and the ability to create agents using natural language instructions without coding.

  • 00:25:00 - 00:30:00

    The session showcases customer success stories, including a retail fraud detection agent that significantly increased processing speed and a banking agent that effectively answered customer queries.

  • 00:30:00 - 00:35:00

    Vasvi begins a live demo, creating a furniture retail assistant agent that can answer product questions and check if items fit in specified spaces, demonstrating the ease of use of the agent builder.

  • 00:35:00 - 00:40:00

    The demo continues with the addition of knowledge sources from SharePoint, allowing the agent to provide accurate product information and respond to specific queries about furniture dimensions.

  • 00:40:00 - 00:45:00

    Vasvi enhances the agent's capabilities by integrating it with additional data sources, such as ServiceNow and Dataverse, to fetch customer order information and incidents, showcasing the flexibility of the platform.

  • 00:45:00 - 00:54:16

    The session concludes with a discussion on the potential of autonomous agents to handle customer requests triggered by emails, demonstrating how these agents can automate processes and improve efficiency.

もっと見る

マインドマップ

ビデオQ&A

  • What is Copilot Studio?

    Copilot Studio is Microsoft's agent-building platform that allows organizations to create AI agents for various business processes.

  • What types of agents can be built using Copilot Studio?

    You can build retrieval agents, task-based agents, and autonomous agents using Copilot Studio.

  • How can businesses benefit from using AI agents?

    AI agents can help automate processes, improve efficiency, and provide significant ROI by reducing manual tasks and enhancing customer interactions.

  • What tools does Microsoft provide for building AI agents?

    Microsoft provides a range of tools including Copilot Studio, Azure AI Foundry, and integration with Microsoft 365.

  • Can non-technical users build agents?

    Yes, users can build agents using simple natural language instructions without needing coding skills.

  • What are some examples of successful AI agent implementations?

    Examples include a retail fraud detection agent that increased processing speed by 677% and a banking agent that answered 90% of customer queries.

  • How do autonomous agents work?

    Autonomous agents can be triggered by external events, such as receiving an email, to perform tasks automatically.

  • What is the significance of data in building AI agents?

    High-quality data is essential for AI agents to function effectively and deliver accurate results.

  • How can agents be published for use?

    Agents can be published across multiple channels, including Microsoft Teams and other communication platforms.

  • What is the role of collaboration in building AI solutions?

    Collaboration across departments enhances the effectiveness of AI solutions by leveraging diverse insights and data.

ビデオをもっと見る

AIを活用したYouTubeの無料動画要約に即アクセス！
字幕
en
オートスクロール:
  • 00:00:03
    [Music]
  • 00:00:05
    hi
  • 00:00:06
    everyone welcome to the podcat AI
  • 00:00:08
    webinars first session uh unlocking the
  • 00:00:11
    potential of a co-pilot Studio I'm
  • 00:00:13
    Ashley I can see we have quite a number
  • 00:00:16
    of people we're so excited to have you
  • 00:00:18
    um so yeah uh no I just want to say
  • 00:00:22
    these are our first in the session and
  • 00:00:23
    we're excited to have you um I'm Ashley
  • 00:00:26
    and I'm one of the hosts um and I'll
  • 00:00:28
    just be answering questions in the
  • 00:00:30
    background and helping everything run
  • 00:00:32
    smoothly and we have two other speakers
  • 00:00:34
    for you today um we have pav Taria and
  • 00:00:38
    vasti and I'll let them introduce
  • 00:00:40
    themselves thank you Ash Ashley I'll go
  • 00:00:43
    first my name is pavari I'm a principal
  • 00:00:45
    product manager in the customer advisory
  • 00:00:47
    team focusing on copilot Studio growth
  • 00:00:50
    very excited to be kicking off this
  • 00:00:51
    inaugural session with all of you and
  • 00:00:53
    introducing uh over to you uh
  • 00:00:57
    vasvi hi everyone I'm Vas set I'm also
  • 00:01:00
    on the power cat team we are the
  • 00:01:02
    customer advisory Team all up for Power
  • 00:01:04
    Platform and many of you might have
  • 00:01:06
    interacted with us and through many
  • 00:01:08
    channels uh in my current role within
  • 00:01:10
    powercat I'm more focused on developing
  • 00:01:12
    content as well as managing content for
  • 00:01:15
    a scale programs like Kickstarter and
  • 00:01:17
    this one our AI webinars that we are
  • 00:01:19
    launching today and super excited for
  • 00:01:21
    this awesome no thank you guys we're
  • 00:01:24
    really happy to have you guys speaking
  • 00:01:25
    today we have a packed agenda today I'll
  • 00:01:28
    start by giving you the business
  • 00:01:29
    business you know like how businesses
  • 00:01:31
    are essentially
  • 00:01:32
    transforming uh you know their their uh
  • 00:01:35
    processes with agents uh I'll cover the
  • 00:01:38
    different tools Microsoft uh allows for
  • 00:01:41
    anyone to build these agents and how
  • 00:01:43
    these tools are better together we heard
  • 00:01:46
    about the customers and what they love
  • 00:01:48
    about building these agents and the
  • 00:01:50
    impact that they're seeing and then
  • 00:01:52
    finally our amazing uh V will uh
  • 00:01:55
    demonstrate uh how to kind of get
  • 00:01:57
    started building agents uh We've res
  • 00:02:00
    some time for the for answering
  • 00:02:01
    questions so if you have those questions
  • 00:02:03
    please feel free to uh put those in your
  • 00:02:05
    AMA tool as Ashley indicated earlier all
  • 00:02:08
    right let's get started so uh you know
  • 00:02:10
    our customers have been busy we've been
  • 00:02:12
    seeing these customers build amazing
  • 00:02:14
    agents that span a wide variety of use
  • 00:02:17
    cases right from helping customers selfs
  • 00:02:20
    serve and resolve issues themselves to
  • 00:02:22
    building Agents that increase efficiency
  • 00:02:25
    in various functions like HR it right
  • 00:02:28
    and there are many more there are a few
  • 00:02:30
    interesting things for you to kind of
  • 00:02:32
    look at on on this slide to just get
  • 00:02:34
    your ideas flowing um that and and all
  • 00:02:37
    of these have delivered some significant
  • 00:02:39
    Roi within your organization so here are
  • 00:02:41
    a few things um one of the things that
  • 00:02:43
    you must think about is how do you
  • 00:02:45
    prioritize the array of uh use cases
  • 00:02:48
    that you're seeing here uh that that
  • 00:02:50
    truly understand that they are ready to
  • 00:02:52
    be transformed within your organization
  • 00:02:53
    so you know how do you prioritize these
  • 00:02:56
    things well we I have three questions
  • 00:02:58
    that you probably want to ask uh for for
  • 00:03:00
    for deciding the use case you want to go
  • 00:03:02
    first what specific business problem or
  • 00:03:05
    inefficiency are you looking to solve
  • 00:03:06
    with AI you must clearly articulate the
  • 00:03:09
    problem who will benefit from this and
  • 00:03:13
    what does success look
  • 00:03:15
    like second which areas of the business
  • 00:03:18
    have the highest potential for impact
  • 00:03:20
    through some sort of conversational
  • 00:03:22
    automation is it the HR team is it the
  • 00:03:25
    support team right like there are many
  • 00:03:27
    departments that can benefit from this
  • 00:03:29
    cuz the cuz this use cases span across
  • 00:03:32
    look for areas that have workflows that
  • 00:03:34
    are resource intensive repetitive and
  • 00:03:38
    prone to errors and look beyond your own
  • 00:03:40
    Department as I indicated earlier as
  • 00:03:42
    well the best outcomes that we see
  • 00:03:44
    customers delivering success with is one
  • 00:03:46
    that requires you to collaborate across
  • 00:03:48
    your or boundaries and the third thing
  • 00:03:51
    is do you have the necessary data and
  • 00:03:54
    the right Buy in from your stakeholders
  • 00:03:56
    to kind of build these kind of AI
  • 00:03:58
    Solutions at scale
  • 00:04:00
    effectively success with AI really
  • 00:04:03
    depends on reasoning over high quality
  • 00:04:05
    data and requires buying from these key
  • 00:04:08
    stakeholders across the organizations
  • 00:04:10
    and some real you you'll see some
  • 00:04:12
    examples of real impactful use cases um
  • 00:04:15
    you know require this this kind of
  • 00:04:17
    collaboration that I'm talking about uh
  • 00:04:19
    I put a link to the implementation guide
  • 00:04:21
    which is a excellent resource for you to
  • 00:04:24
    start your planning and execution
  • 00:04:26
    Journey so as a company Microsoft is
  • 00:04:29
    committed towards our vision of
  • 00:04:31
    empowering every employee with a
  • 00:04:34
    co-pilot and to transform every business
  • 00:04:37
    process with
  • 00:04:40
    agents and our tools or our range of
  • 00:04:43
    tools enable organizations to build
  • 00:04:45
    agents that vary in levels of complexity
  • 00:04:48
    and capabilities that depend on your
  • 00:04:50
    organization's needs when we look at all
  • 00:04:53
    these customers building agents we see a
  • 00:04:55
    few Trends
  • 00:04:56
    emerge many customers start their AG
  • 00:04:59
    building Journey with building simple
  • 00:05:01
    retrieval agents that can kind of Reason
  • 00:05:03
    over and answer questions based on your
  • 00:05:07
    data based on your knowledge think of
  • 00:05:09
    these as those HR policy agents that can
  • 00:05:12
    answer benefit questions or those retail
  • 00:05:14
    agents that can answer customer
  • 00:05:16
    questions and find the right products
  • 00:05:19
    and services that your organizations
  • 00:05:20
    have to
  • 00:05:22
    offer the next type of Agents we see
  • 00:05:24
    customers build are task-based agents
  • 00:05:27
    these agents allow them allow the
  • 00:05:29
    customer customers to build agents that
  • 00:05:31
    act on behalf of their end users imagine
  • 00:05:34
    placing orders making appointments
  • 00:05:36
    booking a vacation all through the power
  • 00:05:39
    of natural language these two are table
  • 00:05:43
    Stakes for any conversational AI
  • 00:05:46
    platform but why stop there and we
  • 00:05:49
    aren't stopping there we are now
  • 00:05:50
    enabling customers to build autonomous
  • 00:05:53
    agents ones that can be triggered
  • 00:05:55
    through external events say an employee
  • 00:05:57
    has joined your team or a customer
  • 00:06:00
    initiates a return autonomous agents can
  • 00:06:03
    create autonomically Dynamic plans to
  • 00:06:07
    run through multi-step workflows imagine
  • 00:06:10
    these agents that can process orders do
  • 00:06:13
    inventory management trigger orders when
  • 00:06:15
    your with your suppliers when your stock
  • 00:06:17
    levels are low or even connect employees
  • 00:06:20
    on a recurring basis with the right
  • 00:06:22
    mentors and training to suit their
  • 00:06:25
    development plans the good news is that
  • 00:06:28
    Microsoft has the right tools for you to
  • 00:06:30
    build these types of
  • 00:06:32
    Agents copilot studio is the agent
  • 00:06:34
    building platform at Microsoft and we're
  • 00:06:36
    integrated this products capabilities to
  • 00:06:39
    offer a range of building experiences to
  • 00:06:41
    suit your organization's expertise and
  • 00:06:44
    maximize the impact within the
  • 00:06:46
    organization for example for information
  • 00:06:49
    workers agent Builder powered by copilot
  • 00:06:52
    studio is fully embedded within teams
  • 00:06:54
    and enables these work information
  • 00:06:57
    workers to build agents using simple
  • 00:06:59
    instr ruction for our makers the full
  • 00:07:02
    co-pilot Studio experience enables you
  • 00:07:04
    to build agents with your own knowledge
  • 00:07:06
    wherever it may be trigger actions on
  • 00:07:09
    your own internal apis or even deploy
  • 00:07:12
    these agents across channels where your
  • 00:07:14
    customers are
  • 00:07:16
    communicating and for developers that
  • 00:07:18
    are looking to build and integrate
  • 00:07:20
    custom AI components for tailored
  • 00:07:22
    Solutions Azure AI Foundry and copala
  • 00:07:26
    Studio are better together and is the
  • 00:07:28
    preferred approach to building these
  • 00:07:30
    agents the best part about the tools
  • 00:07:33
    that I talked about earlier is that they
  • 00:07:34
    are better together so we all know about
  • 00:07:38
    M365 copilot which instantly transforms
  • 00:07:42
    information workers to be their personal
  • 00:07:44
    best in
  • 00:07:46
    productivity um and and when they are
  • 00:07:49
    ready to create very specific repeatable
  • 00:07:52
    task or function using an agent like
  • 00:07:55
    writing coach or a market survey agent
  • 00:07:58
    they can use agent Builder powered by
  • 00:08:00
    copilot Studio to reason over their own
  • 00:08:03
    Enterprise data or we or public
  • 00:08:06
    websites once these agents need to
  • 00:08:08
    integrate with other knowledge sources
  • 00:08:10
    like service now sap or need to be
  • 00:08:13
    hosted on multiple channels um these
  • 00:08:17
    customers our customers can build those
  • 00:08:19
    experiences into copilot
  • 00:08:22
    studio and then when those developers
  • 00:08:25
    want to build those custom AI uh you
  • 00:08:27
    know Solutions and integrate them into
  • 00:08:30
    copilot Studio or make copilot studio
  • 00:08:32
    available in their Pro code
  • 00:08:34
    Solutions these Solutions all work
  • 00:08:36
    Better Together We believe the best
  • 00:08:39
    place for you to build these agents in a
  • 00:08:41
    world where you will be building a lot
  • 00:08:42
    of them is in copilot Studio which is
  • 00:08:45
    our low code fully hosted agent building
  • 00:08:49
    platform whether it's building for um
  • 00:08:52
    you know your your um your own channels
  • 00:08:55
    your own um you know websites or
  • 00:08:57
    extending uh and through 365 co-pilot
  • 00:09:01
    you can start by simply describing what
  • 00:09:03
    you want you can add triggers so that
  • 00:09:05
    these agents can run uh you know
  • 00:09:07
    autonomously and orchestrate multi-step
  • 00:09:11
    workflows um you can reason over uh
  • 00:09:14
    knowledge and create those retrieval
  • 00:09:16
    agents that I talked about whether
  • 00:09:18
    whether the knowledge is a public
  • 00:09:19
    website a file a database or even a
  • 00:09:23
    third party
  • 00:09:24
    service Beyond knowledge you can even
  • 00:09:27
    create rule-based topics to ensure that
  • 00:09:30
    your agents can not only adapt to your
  • 00:09:32
    business processes but also complete
  • 00:09:34
    tasks on behalf of your customers and
  • 00:09:37
    once your agent is ready publish it to
  • 00:09:40
    the channel where your customers are at
  • 00:09:42
    and with builtin Rich analytics to get
  • 00:09:45
    deeper insights on how your knowledge is
  • 00:09:47
    performing and what questions your agent
  • 00:09:49
    is able to respond to you have the
  • 00:09:52
    ability to fully fine-tune what your
  • 00:09:54
    agent does and and performs uh for your
  • 00:09:57
    organizational needs
  • 00:09:59
    and we're also the only company that
  • 00:10:01
    offers this Continuum of agent building
  • 00:10:04
    tools that work better together so your
  • 00:10:06
    investment in copilot is safe even if
  • 00:10:09
    your requirements exceed current product
  • 00:10:13
    capabilities let me switch gears a bit
  • 00:10:16
    let me talk about the momentum we're
  • 00:10:17
    seeing from our customers already over
  • 00:10:20
    100,000 companies are using copal Studio
  • 00:10:23
    to create their own agents for
  • 00:10:25
    Enterprise transformation in just about
  • 00:10:28
    every language every country and every
  • 00:10:31
    industry agents for their employees
  • 00:10:34
    agents for customers agents for B2B
  • 00:10:37
    scenarios and more and across all those
  • 00:10:40
    use cases there are a few of them a few
  • 00:10:43
    positive themes originate what customers
  • 00:10:45
    love uh is that they can build these
  • 00:10:48
    agents and deploy them really quickly
  • 00:10:50
    they love the significant return on
  • 00:10:52
    investment and the cost savings that
  • 00:10:53
    they can achieve as a result they love
  • 00:10:56
    that they can connect to their own
  • 00:10:57
    knowledge sources and existing backs
  • 00:11:00
    without having to move their data they
  • 00:11:02
    love how easy it is to build the stateof
  • 00:11:05
    thee art generative system without
  • 00:11:06
    having to train custom models or going
  • 00:11:09
    deep into the complexities of
  • 00:11:12
    AI they love that generative AI exists
  • 00:11:15
    but they also love the fact that they
  • 00:11:17
    can keep tight control over specific
  • 00:11:20
    flows and curate them to meet their
  • 00:11:23
    business needs and finally they love how
  • 00:11:26
    all of this comes packaged together in a
  • 00:11:29
    geod distributed secure and compliant
  • 00:11:32
    SAS that can be governed uh using
  • 00:11:35
    Enterprise grade Solutions and full
  • 00:11:37
    security and admin controls let me take
  • 00:11:39
    an example of a few customers pets at
  • 00:11:42
    home first use case was that they built
  • 00:11:44
    a retail fraud detection agent that
  • 00:11:47
    autonomously process data in real time
  • 00:11:50
    to prevent fraud this single agent
  • 00:11:53
    increase their velocity in processing
  • 00:11:56
    cases by
  • 00:11:57
    677 per. they're now looking to bring
  • 00:12:01
    this technology to other parts of the
  • 00:12:03
    business virgin money needed a way to
  • 00:12:06
    help their customers feel comfortable
  • 00:12:08
    building and completing digital banking
  • 00:12:11
    actions and so their award-winning agent
  • 00:12:14
    called red eye was able to answer 90% of
  • 00:12:18
    the customers
  • 00:12:21
    questions BYU pathway was able to start
  • 00:12:25
    saving 150 human hours with the agent
  • 00:12:28
    that they build in just a single week
  • 00:12:32
    and clex finally um went and created a
  • 00:12:36
    customized agent that went beyond the
  • 00:12:38
    traditional chatbots to assist their own
  • 00:12:40
    customer requests to reduce handle time
  • 00:12:44
    from the five to 15 minutes per request
  • 00:12:47
    to you know less than a minute typically
  • 00:12:50
    30
  • 00:12:51
    seconds there are many more stories on
  • 00:12:54
    this on the website link that I put
  • 00:12:56
    below all right we're now ready to we've
  • 00:13:00
    talked enough we're now ready for VAs to
  • 00:13:01
    come and show us how to build these
  • 00:13:04
    agents and how to get started in copilot
  • 00:13:06
    Studio posie over to you perfect thank
  • 00:13:10
    you so much Pavan this was like great
  • 00:13:13
    introduction to copilot studio and I'm
  • 00:13:15
    really excited to take some of these uh
  • 00:13:18
    topics that we just covered more
  • 00:13:20
    specifically about agents and show them
  • 00:13:22
    in action so I'm going to do live demos
  • 00:13:25
    of how you can actually create these um
  • 00:13:28
    ret re based agents task agents
  • 00:13:31
    autonomous agents all tied to a real
  • 00:13:33
    world example so with that let me get
  • 00:13:36
    started with my screen share and go into
  • 00:13:40
    the live
  • 00:13:42
    demo um before I move forward just want
  • 00:13:45
    to confirm is everybody able to hear me
  • 00:13:47
    okay any
  • 00:13:49
    concerns all good see a couple of Thumbs
  • 00:13:53
    Up perfect thank you so what I'm going
  • 00:13:56
    to try and do here is I imagine like uh
  • 00:13:59
    I'm working in a Furniture retail store
  • 00:14:01
    and I want to build a Furniture retail
  • 00:14:03
    assistant what uh I'm expecting here my
  • 00:14:07
    uh Furniture assistant to do is maybe be
  • 00:14:09
    able to answer very simple questions
  • 00:14:11
    about like the furniture products
  • 00:14:13
    available I also want my agent to be
  • 00:14:15
    able to help answer uh more specific
  • 00:14:18
    questions like hey Will a particular
  • 00:14:20
    sofa fit in a requested or a mentioned
  • 00:14:22
    space so without much delay let's go
  • 00:14:25
    ahead and get started uh as you can see
  • 00:14:28
    here the screen I'm sharing here uh I've
  • 00:14:30
    opened the M365 chat experience uh and
  • 00:14:34
    what I'm going to show here is how you
  • 00:14:36
    can build an agent using this agent
  • 00:14:38
    Builder experience in the M365 chat on
  • 00:14:41
    my right pane I see I have options where
  • 00:14:44
    I can either get existing agents added
  • 00:14:47
    and use them for my day-to-day job or I
  • 00:14:50
    have the ability to go create an agent
  • 00:14:52
    here and I'm going to go click this
  • 00:14:54
    create agent and in this experience you
  • 00:14:57
    see how I can not just configure
  • 00:15:01
    directly I can with with natural
  • 00:15:03
    language I'm able to interact in this
  • 00:15:05
    experience provide my description and go
  • 00:15:07
    create a new agent of my choice so
  • 00:15:10
    without much further delay let me tell
  • 00:15:13
    in this experience that hey your
  • 00:15:15
    furniture retail assistant and exactly
  • 00:15:18
    the scenario that I was mentioning about
  • 00:15:21
    so as I sent this response the interface
  • 00:15:25
    here is reasoning over the description
  • 00:15:28
    that have provided it comes back with
  • 00:15:31
    like hey do you want Furniture assistant
  • 00:15:32
    as the name that sounds good so let me
  • 00:15:34
    confirm on the name
  • 00:15:37
    here and of course it needs further
  • 00:15:40
    instructions on what this furniture
  • 00:15:42
    assistant is supposed to do I have my
  • 00:15:44
    instructions free ready I'm telling it
  • 00:15:46
    like hey you have a knowledge and
  • 00:15:48
    SharePoint source that you're going to
  • 00:15:50
    use for your uh for building uh for for
  • 00:15:53
    your U um you know experience here and
  • 00:15:56
    giving more specific instructions here
  • 00:15:59
    and in a moment let let us go into the
  • 00:16:01
    SharePoint site where I do have the
  • 00:16:04
    scoso products in my document library
  • 00:16:07
    and what I've tried creating is all my
  • 00:16:09
    sample products uh are in the contoso
  • 00:16:12
    product data and I'm also providing
  • 00:16:14
    extensive information about uh you know
  • 00:16:18
    like whether pickup is available uh
  • 00:16:20
    information about uh Insurance warranty
  • 00:16:23
    return replacement refund policies or if
  • 00:16:26
    you want to go buy secondhand products
  • 00:16:27
    right all this data is available for me
  • 00:16:29
    in SharePoint so here now the agent is
  • 00:16:33
    uh ready up and uh running so I can
  • 00:16:36
    further go into the configure tab you
  • 00:16:38
    see how it picked the name Furniture
  • 00:16:41
    assistant that it was recommending based
  • 00:16:42
    on my confirmation it created the
  • 00:16:45
    description it added all the
  • 00:16:47
    instructions that I provided and now I
  • 00:16:50
    can point this agent to the knowledge
  • 00:16:52
    that I was talking about earlier in the
  • 00:16:54
    SharePoint site so I'm able to quickly
  • 00:16:57
    browse my SharePoint Library and select
  • 00:16:59
    this particular koser products folder
  • 00:17:02
    which contains all the data related to
  • 00:17:04
    my furniture products and that's that's
  • 00:17:07
    it right that's exactly what my agent
  • 00:17:10
    needs now it has everything ready and
  • 00:17:12
    you see how based on the instructions
  • 00:17:15
    based on the knowledge that I've
  • 00:17:17
    uploaded it's come up already with some
  • 00:17:19
    uh you know outof the boox prompts I can
  • 00:17:22
    either use these or use more specific
  • 00:17:25
    prompts where I want to ask my agent
  • 00:17:27
    like hey what kind of furniture do you
  • 00:17:28
    have
  • 00:17:30
    and you see how my agent is reasoning
  • 00:17:32
    over the content that it was provided
  • 00:17:35
    it's reasoning over the instructions So
  • 00:17:37
    based on that and the data that is
  • 00:17:39
    available it gave me back the results
  • 00:17:41
    from my uh you can see how it references
  • 00:17:45
    the document that I was using to get all
  • 00:17:47
    the information about the available
  • 00:17:49
    products super easy right I I didn't
  • 00:17:51
    have to like do any coding with just
  • 00:17:54
    some basic instructions I was able to
  • 00:17:56
    get my agent to work and give me the
  • 00:17:58
    data that I need
  • 00:17:59
    now we did give our agent an additional
  • 00:18:02
    instruction so let me try this like Hey
  • 00:18:05
    will this modern coffee table actually
  • 00:18:07
    fit in a room that is 100 by
  • 00:18:10
    100 you see how quickly my agent came
  • 00:18:12
    back with a response saying yes it's
  • 00:18:14
    going to fit because the coffee table
  • 00:18:16
    actually only measures so and so you see
  • 00:18:19
    how the agent is intelligently able to
  • 00:18:22
    respond it's taking the instructions
  • 00:18:24
    that I gave and it knows kind of how to
  • 00:18:27
    answer to a specific question and if you
  • 00:18:29
    go back and see my instruction I just
  • 00:18:31
    told you need to do this but I didn't
  • 00:18:32
    tell you like how to go uh learn
  • 00:18:35
    mathematics or calculate and everything
  • 00:18:37
    right so uh this is a super quick way of
  • 00:18:40
    how you can go do
  • 00:18:45
    this perfect we have everything ready
  • 00:18:48
    here and
  • 00:18:52
    now perfect so technically I should be
  • 00:18:56
    able to click create here uh seems like
  • 00:18:59
    I have exceeded some limits here so
  • 00:19:02
    let's just remove that and get our agent
  • 00:19:05
    to save in a
  • 00:19:08
    moment okay
  • 00:19:12
    perfect yes so I'm going to go click
  • 00:19:15
    create and we have our agent being
  • 00:19:18
    created in the background and once this
  • 00:19:20
    agent is created you can also see how
  • 00:19:23
    easily I'm able to share this agent with
  • 00:19:26
    with uh it's available for me to just
  • 00:19:29
    use as it is or imagine in a very simple
  • 00:19:32
    scenario right like here I'm coming in
  • 00:19:34
    exploring trying to build something that
  • 00:19:36
    works for me but maybe I want to once I
  • 00:19:39
    feel confident I want to uh share this
  • 00:19:41
    agent with a few specific folks in my
  • 00:19:43
    team so I can do that through security
  • 00:19:45
    groups or I can share it with anyone in
  • 00:19:48
    in my organization if I think that this
  • 00:19:50
    is going to benefit for everybody for
  • 00:19:52
    now we'll just leave it at here but as
  • 00:19:55
    you can see once I've created the agent
  • 00:19:57
    the agent will also start appearing ing
  • 00:19:58
    in the right pain for me to go and
  • 00:20:01
    further interact it's probably just
  • 00:20:03
    going to take a second I'll refresh it
  • 00:20:05
    uh for for
  • 00:20:09
    us this is a live demo so I'm a little
  • 00:20:12
    nervous I'm constantly praying to the
  • 00:20:14
    demo Gods hoping that you know it it
  • 00:20:16
    works as expected so uh very quickly
  • 00:20:19
    hopefully yes here is a Furniture
  • 00:20:21
    assistant that we just created it's
  • 00:20:23
    available for me to test and use it and
  • 00:20:26
    ask more questions and it's just going
  • 00:20:28
    to work fine so this is the M365 agent
  • 00:20:32
    Builder experience that we all have been
  • 00:20:35
    talking about we saw how quickly it is
  • 00:20:38
    easy to build this agent with just some
  • 00:20:41
    instructions and now let me actually
  • 00:20:43
    switch over to kilot Studio but before
  • 00:20:46
    that I as I'm here right so when I click
  • 00:20:49
    on this create an agent um and look at
  • 00:20:51
    my furniture assistant here eventually
  • 00:20:55
    you will see an option where you can
  • 00:20:57
    take this agent that you build in your
  • 00:20:59
    M365 experience and edit it in copilot
  • 00:21:02
    studio uh as far as I know it should
  • 00:21:04
    light up very soon I don't have that
  • 00:21:05
    available in My Demo environment but you
  • 00:21:08
    see how easy it is to start with
  • 00:21:10
    something small in the agent Builder
  • 00:21:11
    experience and switch over to copilot
  • 00:21:15
    Studio okay so here I'm in my copilot
  • 00:21:18
    Studio experience to save time for us I
  • 00:21:21
    haven't gone in and started building the
  • 00:21:23
    same agent in copilot Studio but
  • 00:21:26
    technically you can build uh you can
  • 00:21:28
    continue to edit that same agent in
  • 00:21:30
    copilot studio in near future but for
  • 00:21:33
    now I went ahead and manually added some
  • 00:21:35
    description instructions and I have the
  • 00:21:37
    exact same agent uh knowledge added from
  • 00:21:40
    Koso products everything ready here for
  • 00:21:43
    it for me to uh get going so in my first
  • 00:21:48
    scenario where I called out about this
  • 00:21:50
    furniture retail assistant I was only
  • 00:21:52
    trying to get some information from
  • 00:21:53
    SharePoint as a knowledge Source but if
  • 00:21:57
    you want to enhance your agent with
  • 00:21:59
    additional knowledge sources additional
  • 00:22:01
    actions or even publish them beyond your
  • 00:22:03
    SharePoint Channel you can easily switch
  • 00:22:06
    to copilot studio and continue enhancing
  • 00:22:08
    your agent so I I won't go super deep on
  • 00:22:11
    some of the things but you see how like
  • 00:22:13
    you know you can even design custom
  • 00:22:15
    topics in this experience and going into
  • 00:22:18
    knowledge uh Beyond SharePoint earlier
  • 00:22:20
    we saw only SharePoint was our option
  • 00:22:22
    where you could browse the files and uh
  • 00:22:24
    you know point it to a particular
  • 00:22:25
    location but here I can either point to
  • 00:22:28
    uh public websites or I can add data ver
  • 00:22:31
    also as one of my knowledge source using
  • 00:22:34
    structured data so first let's go with
  • 00:22:37
    data ver but imagine in my scenario
  • 00:22:40
    Beyond getting the product information I
  • 00:22:43
    also want to be able to use my agent uh
  • 00:22:45
    as I'm interacting with the agent right
  • 00:22:47
    I I want to quickly be able to say like
  • 00:22:49
    hey get me order information or get me
  • 00:22:51
    information about a customer so I want
  • 00:22:53
    to be able to easily do that with my
  • 00:22:55
    agent and similarly I may want to look
  • 00:22:58
    at you know incidents in service now so
  • 00:23:02
    you clearly see how going into the
  • 00:23:03
    advanced tab I can add more Enterprise
  • 00:23:06
    data connections if you have knowledge
  • 00:23:08
    in uh externally in Azure you can bring
  • 00:23:11
    that into your copilot Studio agent and
  • 00:23:13
    get your agent to reason over it and
  • 00:23:14
    answer any questions similarly even with
  • 00:23:17
    service now uh I I'm able to quickly go
  • 00:23:20
    in and create a connection to my service
  • 00:23:23
    now instance uh I have a bunch of
  • 00:23:26
    incidents in service now already created
  • 00:23:29
    and I'm able to just go in here I have
  • 00:23:32
    the connection established already I'm
  • 00:23:34
    just saying
  • 00:23:36
    next and as I'm adding my service now
  • 00:23:38
    connection uh let let me tell you what
  • 00:23:40
    I'm trying to do here right so I I I
  • 00:23:43
    look in the incident table basically I'm
  • 00:23:46
    adding a step to my agent where I can
  • 00:23:50
    connect my agent to the incident table
  • 00:23:52
    and service now and get my agent to
  • 00:23:54
    reason over some of the
  • 00:23:56
    incidents you see how
  • 00:23:59
    um give me a second so I make sure I
  • 00:24:01
    have the right prompt here perfect so
  • 00:24:05
    this knowledge description is super
  • 00:24:07
    important in the sense that your
  • 00:24:10
    agent will understand when to use this
  • 00:24:13
    particular knowledge especially in the
  • 00:24:15
    runtime right so uh without it always
  • 00:24:19
    good to you know go here review the
  • 00:24:21
    description that you provide to the
  • 00:24:23
    agent so your agent knows when to
  • 00:24:25
    trigger this particular knowledge when
  • 00:24:27
    to look for the uh data or content
  • 00:24:29
    within this knowledge so uh it's good as
  • 00:24:32
    it is but I I want to explicitly go call
  • 00:24:34
    out like hey use this knowledge Source
  • 00:24:36
    when you want to fetch particular
  • 00:24:37
    information about incidents and then
  • 00:24:40
    here I'm able to just go in and add
  • 00:24:42
    service now connection uh service now as
  • 00:24:45
    a knowledge to my agent this might take
  • 00:24:48
    a couple minutes and similarly
  • 00:24:51
    um I thought it will at least close here
  • 00:24:54
    but similarly I'll show you a step where
  • 00:24:57
    you can add uh data ver so once you have
  • 00:25:00
    data ver as well
  • 00:25:03
    um okay so similarly I was talking about
  • 00:25:06
    order and customer information that I
  • 00:25:08
    have in data so I have this product
  • 00:25:10
    order table that I've created in data
  • 00:25:12
    which contains information about all the
  • 00:25:14
    orders uh I have two orders in here
  • 00:25:17
    placed by two different customers one is
  • 00:25:19
    pan and one more is uh Emily uh sorry
  • 00:25:22
    that I don't have the order uh the
  • 00:25:25
    customer name but I have more
  • 00:25:27
    specifically just the customer customer
  • 00:25:28
    ID in here uh and along with this I can
  • 00:25:32
    even just add the data from customer
  • 00:25:34
    table but for now let me just go ahead
  • 00:25:36
    and add the order table here similarly
  • 00:25:39
    we can go in and add the customer table
  • 00:25:41
    which contains the customer information
  • 00:25:44
    what I'm going to do here this probably
  • 00:25:46
    is going to take a moment uh here so
  • 00:25:48
    using the exact same steps I have
  • 00:25:51
    pre-built part of this agent so that we
  • 00:25:54
    can you know Skip uh waiting for the
  • 00:25:57
    service now connected to get ready as
  • 00:25:59
    well as skipping through the processes
  • 00:26:01
    of like you know adding U data was
  • 00:26:03
    customer and Order table so let me show
  • 00:26:06
    you in the knowledge previously we just
  • 00:26:08
    had condos products as our knowledge but
  • 00:26:10
    I've now added the product order
  • 00:26:13
    customer tables from data was and I've
  • 00:26:15
    been able to add the incident table in
  • 00:26:18
    service now additionally I want to
  • 00:26:21
    instruct my agent saying like hey now
  • 00:26:24
    you can go use the data from uh service
  • 00:26:27
    now and return and display that data so
  • 00:26:29
    I just went in and added this one
  • 00:26:31
    instruction and when specifically a
  • 00:26:35
    customer comes in and ask for orders
  • 00:26:37
    placed by so and so customer I want to
  • 00:26:39
    be able to use that customer information
  • 00:26:42
    to go and fetch the related order Ting
  • 00:26:44
    so all of this is purely in my
  • 00:26:46
    instructions and you see how my agent is
  • 00:26:48
    ready with these steps and let's just go
  • 00:26:51
    and test them so I have this question
  • 00:26:55
    like where I'm asking my agent
  • 00:26:59
    because I'm dealing more with the pro
  • 00:27:02
    with the furniture I want to go fetch
  • 00:27:04
    all incidents related to any product
  • 00:27:06
    missing part and you can see how my
  • 00:27:10
    agent is going to reason over the
  • 00:27:12
    knowledge that is available to it in the
  • 00:27:15
    activity map the great the good part is
  • 00:27:17
    it immediately starts showing you what
  • 00:27:19
    your agent is doing right you clearly
  • 00:27:22
    see how transparent it is uh in terms of
  • 00:27:25
    how your agent is working behind the
  • 00:27:26
    scenes it I identified all the available
  • 00:27:29
    knowledge sources and uh and it it
  • 00:27:33
    reasoned over all these knowledge
  • 00:27:35
    available and it knows that it has to go
  • 00:27:37
    look in the incident that is where you
  • 00:27:39
    see how the output is from the incident
  • 00:27:42
    uh which is our service now knowledge
  • 00:27:44
    source and it was able to find three
  • 00:27:47
    incidents related to product missing
  • 00:27:49
    part and how easily it was able to get
  • 00:27:51
    that data for us right so this is how we
  • 00:27:54
    were uh Beyond SharePoint we were able
  • 00:27:57
    to connect with service now and also
  • 00:27:59
    reason over the data in service now and
  • 00:28:01
    our agent is easily able to handle that
  • 00:28:04
    and uh similarly we we added the
  • 00:28:08
    customer and Order tables and data was
  • 00:28:10
    and let's quickly see uh I have one of
  • 00:28:13
    my customer as Pavan taparia in my
  • 00:28:15
    system so I want to check what were the
  • 00:28:17
    orders placed by Pavan in the last 30
  • 00:28:20
    days so I just want to kind of just want
  • 00:28:24
    to pause here and I want to celebrate
  • 00:28:26
    what V was able to do uh in just a few
  • 00:28:29
    clicks she was able to bring in a whole
  • 00:28:32
    bunch of knowledge sources across the
  • 00:28:35
    organization connect it all together
  • 00:28:37
    with simple instruction and now she's
  • 00:28:40
    able to Simply reason over all of that
  • 00:28:43
    knowledge with single prompts that is
  • 00:28:47
    power all right back to you thank you P
  • 00:28:51
    yeah I was really nervous and PN and a
  • 00:28:54
    few other folks like know how nervous I
  • 00:28:56
    was to get this running uh live since
  • 00:28:59
    yesterday right like since last couple
  • 00:29:01
    of days so super excited to see how well
  • 00:29:03
    our agent is performing here uh so
  • 00:29:05
    similarly for the second example I
  • 00:29:07
    wanted to fetch information about orders
  • 00:29:10
    placed by PN in the last 30 days and you
  • 00:29:12
    see how it looked at the knowledge
  • 00:29:14
    sources available to it it and it was
  • 00:29:17
    able to identify that it needs to go
  • 00:29:19
    into the data was customer and Order
  • 00:29:22
    table because it used the customer name
  • 00:29:24
    fetch the customer information like ID
  • 00:29:27
    it used use the customer ID and then
  • 00:29:30
    fetch the related product order a and
  • 00:29:32
    there's no magic here right I did not
  • 00:29:34
    write a flow so if you see in here I'm
  • 00:29:37
    not calling any flows uh all I did was
  • 00:29:40
    add these four knowledge sources and I
  • 00:29:43
    have just provided clear instructions of
  • 00:29:45
    what my agent needs to do when somebody
  • 00:29:47
    asks about orders placed for a customer
  • 00:29:50
    so you see how simple it is to take
  • 00:29:52
    simple uh language and convert that into
  • 00:29:55
    outcome here perfect and with this we we
  • 00:29:59
    kind of like are able to Showcase what
  • 00:30:01
    we are able to do with retrieval agents
  • 00:30:03
    right you are able to fetch the data
  • 00:30:05
    needed and uh share that with uh uh the
  • 00:30:08
    agent is able to fetch the data that you
  • 00:30:10
    need for for
  • 00:30:12
    you I want to show how with these agents
  • 00:30:16
    right as you build them you can even
  • 00:30:18
    publish them across multiple channels we
  • 00:30:22
    in the first demo like where I was
  • 00:30:24
    showcasing the agent Builder experience
  • 00:30:25
    you started off in the M360 five chat
  • 00:30:28
    space but here in copilot Studio you
  • 00:30:30
    have the option to publish over any
  • 00:30:32
    channels including the teams plus
  • 00:30:34
    Microsoft 365 publishing to teams
  • 00:30:37
    channel has always existed but I do want
  • 00:30:39
    to highlight this new option where you
  • 00:30:42
    can now make your agent available in
  • 00:30:43
    M365 co-pilot by selecting this which I
  • 00:30:47
    have already selected here you actually
  • 00:30:50
    uh can see this agent now in your M365
  • 00:30:54
    chat experience so I'll just continue
  • 00:30:57
    here on web so technically what I should
  • 00:30:59
    be doing here is um as I have my agent
  • 00:31:02
    ready I can click on share and say I've
  • 00:31:05
    shared this with uh pavn uh so pavn
  • 00:31:08
    would just have a link and using that
  • 00:31:10
    link pavn is able to access this agent
  • 00:31:14
    in his M365 chat experience here but
  • 00:31:17
    otherwise I can also go into teams here
  • 00:31:21
    um and I've already added this on my
  • 00:31:24
    M365 chat experience so I can just open
  • 00:31:27
    it here or in copilot sorry I I may have
  • 00:31:30
    uninstalled it but otherwise I I can let
  • 00:31:34
    me take a step back I might I might be
  • 00:31:36
    super confusing here so go into your
  • 00:31:39
    teams plus M3 365 make the agent
  • 00:31:42
    available in M365 co-pilot and you
  • 00:31:46
    should be able to add that as an agent
  • 00:31:48
    in your teams and and as well as like by
  • 00:31:52
    just sharing that agent you have access
  • 00:31:54
    to that agent and you can go open in uh
  • 00:31:56
    use it in your m 65 experience so
  • 00:32:00
    quickly switching back from here you see
  • 00:32:02
    how easy it is to build these custom
  • 00:32:04
    engine agents uh either using the agent
  • 00:32:07
    Builder experience or co-pilot studio
  • 00:32:09
    and you can get them to show up on the
  • 00:32:12
    right side pane for any of the users who
  • 00:32:14
    have access can go leverage those agents
  • 00:32:17
    for their day-to-day jobs with that let
  • 00:32:20
    me quickly switch over to uh some of
  • 00:32:23
    like the key takeaways here before we
  • 00:32:26
    move into a poll and continue with our
  • 00:32:29
    demos for the next set of task-based
  • 00:32:32
    agents and autonomous
  • 00:32:34
    agents
  • 00:32:35
    cool so in the first part of using the
  • 00:32:38
    agent Builder experience and copala
  • 00:32:40
    Studio to create these M365 chat
  • 00:32:42
    experience agents we learned how even
  • 00:32:45
    simple information workers without any
  • 00:32:47
    coding knowledge can build these agents
  • 00:32:51
    all by using very simple natural
  • 00:32:53
    language description you saw how these
  • 00:32:55
    agents were grounded in your inter price
  • 00:32:58
    data with irrespective of right you know
  • 00:33:01
    we started off with SharePoint we
  • 00:33:02
    explored data was we explored service
  • 00:33:04
    now it could be sales force or it could
  • 00:33:06
    be in Azure you can bring in the data
  • 00:33:08
    from anywhere uh within your Enterprise
  • 00:33:11
    and get your agent to leverage that and
  • 00:33:13
    the best part here is this is included
  • 00:33:15
    as part of M365 copilot or it was
  • 00:33:19
    available to you with copilot chat with
  • 00:33:21
    with more metered pricing so super easy
  • 00:33:24
    to build definitely I'm pretty sure a
  • 00:33:27
    lot of us uh you know want to build
  • 00:33:29
    agents for our day-to-day purposes where
  • 00:33:32
    you can easily go build these so you
  • 00:33:36
    don't have to
  • 00:33:37
    manually uh you know search inside a
  • 00:33:40
    document in terms of like you know
  • 00:33:42
    whether it's the furniture available or
  • 00:33:43
    for specific data you want somebody to
  • 00:33:45
    get you this data more
  • 00:33:48
    easily oh and one more important thing
  • 00:33:51
    that I missed uh calling out while
  • 00:33:53
    building here right so you also see how
  • 00:33:56
    I'm not using the default environment
  • 00:33:58
    but actually using a different
  • 00:34:00
    environment and able to build this agent
  • 00:34:02
    and still publish into the M365 uh chat
  • 00:34:05
    space so this was a question uh as we
  • 00:34:09
    were doing some Kickstarter workshops or
  • 00:34:10
    in general right given the challenge
  • 00:34:13
    with default environment where every
  • 00:34:15
    organization might have stricter data
  • 00:34:17
    loss prevention policies applied there
  • 00:34:19
    you see how you can build in any
  • 00:34:20
    environment but still open that up in
  • 00:34:22
    the M365 chat experience for for your
  • 00:34:25
    employees so moving
  • 00:34:28
    on and u a we we covered what we can do
  • 00:34:33
    with retrieval agents and let me tell
  • 00:34:36
    you based on what we see from what our
  • 00:34:38
    customers are doing it's not always
  • 00:34:41
    retrieval agents that our customers are
  • 00:34:43
    trying to build right um getting that
  • 00:34:45
    information is amazing but we have
  • 00:34:47
    noticed how our customers want to go one
  • 00:34:50
    step uh one step ahead where along with
  • 00:34:53
    being able to retrieve information they
  • 00:34:55
    want the agents to execute some tasks
  • 00:34:58
    right take some actions and you can do
  • 00:35:01
    that super easily with the agent
  • 00:35:04
    experience in copilot Studio where you
  • 00:35:06
    can add actions you you have like 1,500
  • 00:35:10
    plus connectors that are available to
  • 00:35:12
    you that you can connect to and act on
  • 00:35:14
    them uh Additionally you can use power
  • 00:35:17
    automate Cloud flows to build more
  • 00:35:19
    advanced Logic for for your business uh
  • 00:35:22
    you can add a custom connector uh you
  • 00:35:25
    can add directly connected to a rest API
  • 00:35:28
    so all of this is possible just through
  • 00:35:30
    actions within copilot studio so let's
  • 00:35:34
    look at in in my scenario that I was
  • 00:35:36
    talking about the furniture retail
  • 00:35:37
    assistant right we already have this
  • 00:35:40
    retrieval agent we built it with like
  • 00:35:42
    you know how it can fetch data from four
  • 00:35:44
    different products now let's see if my
  • 00:35:47
    agent can further take some actions
  • 00:35:49
    using the information that it has
  • 00:35:51
    already gathered what I'm going to do
  • 00:35:54
    here is I want my agent to help not just
  • 00:35:57
    retrieve this information but also
  • 00:35:59
    initiate return request for me so I
  • 00:36:01
    should be able to just tell hey initiate
  • 00:36:03
    return requests and the agent should
  • 00:36:05
    know uh what it needs to do in order to
  • 00:36:08
    go return that request
  • 00:36:10
    so it can at least give me the steps
  • 00:36:13
    needed so I don't have to go figure it
  • 00:36:15
    out manually always uh you know like uh
  • 00:36:18
    pav comes and asks me like hey I want to
  • 00:36:20
    return this product I don't have to go
  • 00:36:22
    in manually and look for oh what was the
  • 00:36:24
    order ID what was the product that was
  • 00:36:26
    purchased is in the return or refund
  • 00:36:29
    period right so there's a lot of manual
  • 00:36:31
    effort that I need to do I don't want to
  • 00:36:33
    do that I want to delegate that to my
  • 00:36:35
    agent and here's where I'm adding an
  • 00:36:37
    action to call that I uh in this case I
  • 00:36:41
    have pre-created this initiator return
  • 00:36:44
    which is a power automate Cloud flow all
  • 00:36:46
    I'm going to do is uh this initiate a
  • 00:36:49
    return takes in a couple of inputs it
  • 00:36:51
    takes what is the order ID what is the
  • 00:36:53
    order date who is the customer and then
  • 00:36:56
    it gives me back with a confirmation
  • 00:36:58
    number saying like hey here is your
  • 00:37:00
    return confirmation number along with a
  • 00:37:03
    date by which the customer needs to
  • 00:37:06
    return the product so what I'm going to
  • 00:37:08
    do here is I'm going to add this uh
  • 00:37:10
    action to my agent but I also want to
  • 00:37:12
    make sure I'm giving correct
  • 00:37:15
    instructions to this agent so my agent
  • 00:37:19
    knows when to go fetch this or use this
  • 00:37:21
    action as it is orchestrating a plan
  • 00:37:23
    based on all its available knowledge and
  • 00:37:27
    action so let's go ahead and just add
  • 00:37:29
    this action here it might take a couple
  • 00:37:32
    of seconds but you see how it's using
  • 00:37:34
    that action to understand how it needs
  • 00:37:36
    to reason over and leverage that action
  • 00:37:39
    um and switching back to my overview to
  • 00:37:42
    look at my instructions I haven't
  • 00:37:44
    updated anything with respect to my
  • 00:37:46
    instructions here this is exactly how it
  • 00:37:49
    was when we started building that
  • 00:37:51
    retrieval agent added service now added
  • 00:37:53
    the data ver Knowledge from customer and
  • 00:37:56
    Order without giving any further
  • 00:37:58
    instructions let me actually just type
  • 00:38:01
    this message to my agent saying that hey
  • 00:38:03
    paven would like to return a recently
  • 00:38:05
    purchased accent
  • 00:38:07
    share um it has worked well for me
  • 00:38:10
    sometimes it has not worked well so I'm
  • 00:38:11
    really going to experiment here uh you
  • 00:38:14
    see
  • 00:38:16
    how very quickly even before I could get
  • 00:38:19
    to what it's doing it has an outcome for
  • 00:38:21
    me here so I just said that pavn would
  • 00:38:24
    like to return a recently purchased
  • 00:38:26
    accent shair and you see how this is
  • 00:38:28
    still running in the context it has the
  • 00:38:30
    context of the customer name the order
  • 00:38:33
    it fetched and everything so without
  • 00:38:36
    much uh next steps what it was able to
  • 00:38:39
    do is it initiated this Cloud flow uh to
  • 00:38:43
    return my return the product so it
  • 00:38:45
    fetched the order ID which was already
  • 00:38:47
    there it provided uh it got the order
  • 00:38:49
    date it got the customer name initiated
  • 00:38:52
    that flow and the flow got me back the
  • 00:38:54
    response that hey here's the number and
  • 00:38:56
    this order needs to be returned by April
  • 00:38:58
    5th so you see how my agent was able to
  • 00:39:02
    keep leverage the context that it was
  • 00:39:04
    already provided passed it on used that
  • 00:39:07
    for slot filling and filled it in this
  • 00:39:09
    initiate a flow return request uh and
  • 00:39:12
    you can also clearly see how across
  • 00:39:15
    these knowledge sources it's able to
  • 00:39:17
    fetch the data for uh where we got the
  • 00:39:19
    data from the customer table then we got
  • 00:39:22
    the data from order it was able to
  • 00:39:24
    stitch all of these together it or
  • 00:39:27
    created a plan that it needs to first
  • 00:39:29
    get the data from customer then the
  • 00:39:31
    order and then go initiate the return
  • 00:39:33
    and all of this was without me providing
  • 00:39:36
    any instruction saying like hey when
  • 00:39:39
    somebody asks you to initiate a return
  • 00:39:42
    go fetch this information and then
  • 00:39:44
    initiate that return request right uh of
  • 00:39:47
    course if I do want to make sure my
  • 00:39:50
    agent is working very well I can go in
  • 00:39:53
    and then update the instructions
  • 00:39:55
    specifically saying like hey when asked
  • 00:39:57
    to initiate a return go do these steps
  • 00:40:00
    and if the price is under $500 you know
  • 00:40:02
    then uh go ahead and confirm the return
  • 00:40:05
    here so you see how easily I was able to
  • 00:40:09
    not just fetch knowledge but also create
  • 00:40:13
    agents instantly here
  • 00:40:17
    um before I move on I'm just thinking am
  • 00:40:20
    I missing any important steps here pav
  • 00:40:23
    is there anything else that you would
  • 00:40:24
    like to add from a task agent
  • 00:40:26
    perspective before we move on to our
  • 00:40:29
    takeaways and autonomous agents no just
  • 00:40:32
    um M just I I wanted to kind of uh
  • 00:40:35
    illustrate what you've done already
  • 00:40:36
    really well is the fact that you know
  • 00:40:39
    creating and then based on the questions
  • 00:40:41
    that people are asking uh if you click
  • 00:40:43
    on actions if you just want to quickly
  • 00:40:45
    click on that um any of those 1500 data
  • 00:40:49
    connectors that that are already
  • 00:40:51
    available can be used to bring in an
  • 00:40:55
    action conversationally and contextually
  • 00:40:57
    in copilot studio all you need to do is
  • 00:41:00
    use those magic words to make sure they
  • 00:41:03
    are they get executed correctly and the
  • 00:41:06
    slots that they are expecting get filled
  • 00:41:08
    conversationally in the past what you
  • 00:41:09
    needed to do and many platforms still
  • 00:41:12
    require to go do that is to create and
  • 00:41:14
    wrap these actions into like let me
  • 00:41:15
    collect these five questions and then do
  • 00:41:18
    data type matching and then then call
  • 00:41:20
    the actions contextually and then what
  • 00:41:22
    if the slot changes right like what is
  • 00:41:24
    say sorry I I didn't mean yesterday I
  • 00:41:26
    mean two days from now that that just is
  • 00:41:29
    automatically handled so that's the
  • 00:41:32
    power of having a intelligent platform
  • 00:41:34
    where you're building these agents and
  • 00:41:36
    we're going to continue pushing the
  • 00:41:39
    technology to make this even more
  • 00:41:40
    simpler even this more intuitive so yeah
  • 00:41:43
    I think we can head over to the
  • 00:41:45
    takeaways yeah no thank you for adding
  • 00:41:47
    that additional context that's super
  • 00:41:49
    helpful um and additionally on top of it
  • 00:41:52
    like previously we saw how you could
  • 00:41:54
    publish this agent like to teams plus
  • 00:41:56
    M365 but you have the ability to publish
  • 00:41:59
    to any of these available channels so
  • 00:42:03
    your U user can interact with them on a
  • 00:42:06
    telephone chat or like chat on website
  • 00:42:08
    or on a mobile application right so you
  • 00:42:10
    have these channels available for you to
  • 00:42:12
    go publish your agent with that quickly
  • 00:42:16
    as we wrap up the task agents uh
  • 00:42:18
    task-based agents let's look at the key
  • 00:42:20
    takeway sorry I think this was
  • 00:42:23
    still yes so uh we were able to see how
  • 00:42:27
    we can easily expand the agents add get
  • 00:42:31
    those agents to answer Beyond just
  • 00:42:33
    questions they were able to even
  • 00:42:34
    interact with the line of business
  • 00:42:36
    applications so we saw how you can
  • 00:42:38
    publish to any channel using copilot
  • 00:42:40
    Studio as well as we we saw you know the
  • 00:42:44
    the number of actions are we we have
  • 00:42:46
    1,500 plus connectors and beyond that
  • 00:42:49
    there are actions like you know you can
  • 00:42:51
    bring in the cloud flow you can add a AI
  • 00:42:54
    Builder prompt or you can bring in a
  • 00:42:56
    custom connector or custom API all of
  • 00:42:58
    this is possible with your task based
  • 00:43:01
    agents and we have seen customer
  • 00:43:03
    examples like as P showed us before uh
  • 00:43:06
    in some of our customer examples we have
  • 00:43:08
    seen how a lot of our customers are
  • 00:43:11
    Thinking Beyond retrieval agents to get
  • 00:43:13
    the agents to do certain actions for
  • 00:43:17
    them and as we switch before going into
  • 00:43:21
    the autonomous agents I would love to
  • 00:43:25
    get into this next poll all okay so we
  • 00:43:29
    are at the same agent we learned how to
  • 00:43:32
    you know build retrieval based agents we
  • 00:43:34
    learned how to build task based agents
  • 00:43:36
    ignore the name here I I it's it's the
  • 00:43:40
    furniture agent that we're building here
  • 00:43:41
    right so just ignore the name there for
  • 00:43:43
    now but um we we saw how we built the
  • 00:43:48
    task-based agent to think a little bit
  • 00:43:50
    further to do actions and we were also
  • 00:43:53
    talking about uh you know multiple
  • 00:43:55
    customers wanting to do that one such
  • 00:43:57
    customer was clex who actually built the
  • 00:44:00
    initiate return request for their guest
  • 00:44:02
    service folks and it saved tons of time
  • 00:44:05
    for their guest service folks so rather
  • 00:44:07
    than them going to do these initiating
  • 00:44:10
    the return request manually they were
  • 00:44:11
    able to automate that that part of
  • 00:44:14
    initiating the return request and it
  • 00:44:16
    saved a ton of time as I was calling out
  • 00:44:19
    however when when you're building these
  • 00:44:21
    task-based agents we saw how you can
  • 00:44:23
    deploy them to multiple channels and you
  • 00:44:26
    also see sometimes the challenge can be
  • 00:44:28
    that you know your customers don't
  • 00:44:30
    always interact with you through just
  • 00:44:33
    one form uh it it could be a phone call
  • 00:44:36
    it could be an email it could be like a
  • 00:44:38
    uh you know a chatboard on a web app
  • 00:44:40
    right so there are multiple ways that
  • 00:44:42
    your customers interact and it's harder
  • 00:44:44
    for you to uh kind of address and uh
  • 00:44:48
    manage responses across all those
  • 00:44:50
    channels that is where you can think
  • 00:44:53
    about using autonomous agents to
  • 00:44:55
    actually build uh sorry that is where
  • 00:44:58
    you can actually think about building
  • 00:45:00
    these autonomous agents where you can
  • 00:45:02
    use your autonomous agents to handle
  • 00:45:04
    some of those processes for you so the
  • 00:45:06
    way these autonomous agents are going to
  • 00:45:08
    work is just with a simple trigger right
  • 00:45:11
    uh in my scenario in in the case of the
  • 00:45:14
    furniture retail store that I was
  • 00:45:15
    talking about I was chatting with a
  • 00:45:18
    customer and then use that information
  • 00:45:21
    to go initiate the return request but
  • 00:45:24
    what if you know the customer is
  • 00:45:26
    struggling to get on a call and do all
  • 00:45:28
    of this the customer can just send out
  • 00:45:30
    an email uh into the support inbox
  • 00:45:33
    saying hey I bought this product but I'm
  • 00:45:36
    not happy with it and I would just love
  • 00:45:38
    to go return that product right so even
  • 00:45:40
    that could be a trigger for us to
  • 00:45:42
    initiate the return request and you
  • 00:45:45
    don't have to wait all the way until you
  • 00:45:47
    do a call or any other uh channel to
  • 00:45:50
    respond to your customer so let's look
  • 00:45:52
    at how we can build make this existing
  • 00:45:55
    agent that we've already created into an
  • 00:45:58
    autonomous agent so anytime a new email
  • 00:46:01
    arrives even just the email will trigger
  • 00:46:03
    off the whole process that we are
  • 00:46:04
    talking about so switching back to the
  • 00:46:07
    demo here uh we are in the same demo
  • 00:46:09
    that we already created where we added
  • 00:46:11
    the Last Action initiate a return
  • 00:46:13
    whenever uh you know uh customer asks
  • 00:46:16
    for a return so next what I'm going to
  • 00:46:19
    do is in autonomous agent this feature
  • 00:46:22
    is still in preview and you see how uh
  • 00:46:25
    along with adding knowledge and actions
  • 00:46:27
    you can now add a trigger for your agent
  • 00:46:30
    and the trigger that I'm going to add
  • 00:46:32
    here is when a new email arrives in this
  • 00:46:36
    particular inbox let me just uh fix the
  • 00:46:39
    name here I don't want to use the V3 but
  • 00:46:42
    I I can leave my uh trigger assis when a
  • 00:46:46
    new email arrives in inbox it you see
  • 00:46:48
    how it has uh signed into my Outlook and
  • 00:46:51
    I'm going to ask it to look explicitly
  • 00:46:54
    for emails coming into my inbox
  • 00:46:57
    and as easy as that I can create this
  • 00:47:00
    trigger so the the one thing that I
  • 00:47:03
    wanted to just include is like yeah
  • 00:47:05
    again those 1500 data connectors and the
  • 00:47:07
    triggers that already are available are
  • 00:47:11
    ones that you can automatically build
  • 00:47:13
    these autonomous agents around these
  • 00:47:16
    triggers
  • 00:47:17
    contextually with the necessary
  • 00:47:19
    information can trigger these agents so
  • 00:47:21
    that the agents can carry on with their
  • 00:47:23
    next things uh we also added a if you
  • 00:47:25
    click on the add trigger again we also
  • 00:47:27
    added a few others that are uh more
  • 00:47:30
    recurrent time based so if you wanted
  • 00:47:33
    something to automatically run every
  • 00:47:35
    morning or every month um you know you
  • 00:47:39
    that's possible too I just wanted to
  • 00:47:40
    kind of show that these things can just
  • 00:47:42
    run
  • 00:47:43
    contextually at any given point in time
  • 00:47:46
    based on any external event based on any
  • 00:47:48
    external event yeah so back to you masi
  • 00:47:53
    perfect no thanks thanks P for adding
  • 00:47:55
    that additional context on more
  • 00:47:56
    available triggers here um and as Perman
  • 00:47:59
    was sharing that we see how our trigger
  • 00:48:01
    was already added uh I haven't had any
  • 00:48:05
    emails received in the inbox since I
  • 00:48:07
    created this trigger so you see how this
  • 00:48:09
    is uh still empty but what I'm going to
  • 00:48:11
    do now is uh along with adding this
  • 00:48:14
    trigger I do want to instruct my agent
  • 00:48:17
    explicitly saying hey not just when
  • 00:48:19
    asked to initiate a return uh I want to
  • 00:48:21
    call out explicitly saying he when asked
  • 00:48:23
    to initiate a return or when an email
  • 00:48:26
    arrives asking to return the product
  • 00:48:29
    follow these steps and I want to add one
  • 00:48:32
    additional Step at the end of it along
  • 00:48:35
    with uh you know confirming with the
  • 00:48:38
    customer send an email confirmation to
  • 00:48:40
    the customer including the return
  • 00:48:42
    confirmation and the date by which the
  • 00:48:43
    customer needs to return the product so
  • 00:48:46
    I'm just updating the instruction so my
  • 00:48:47
    agent knows when to do what uh quickly
  • 00:48:50
    go ahead and save the instructions that
  • 00:48:52
    I've added um I have added this
  • 00:48:56
    instructor instruction to send a
  • 00:48:57
    confirmation email so let's go ahead and
  • 00:49:00
    quickly add an action to do that as well
  • 00:49:02
    uh we saw how we can add multiple
  • 00:49:04
    actions here and one of them is sending
  • 00:49:07
    an email
  • 00:49:09
    so I'm searching for send an email
  • 00:49:12
    action uh I find this in the Office 365
  • 00:49:16
    Outlook and again as always think about
  • 00:49:19
    what is the description that you're
  • 00:49:21
    going to provide for your action because
  • 00:49:23
    it will really help your agent to know
  • 00:49:26
    when it needs to come use this action
  • 00:49:29
    right if you don't provide a detailed
  • 00:49:30
    description it it may be confused uh
  • 00:49:34
    email
  • 00:49:35
    confirmation to the customer so I've
  • 00:49:38
    updated the description saying like hey
  • 00:49:40
    use this operation to send a
  • 00:49:41
    confirmation email to the customer who
  • 00:49:43
    initiated the return request and in this
  • 00:49:46
    case I'm going to use the copilot author
  • 00:49:47
    authentication instead of end user
  • 00:49:49
    authentication here uh and that is all
  • 00:49:52
    right I'm not going to go provide any
  • 00:49:54
    specific inputs here I'm not going to
  • 00:49:57
    say whom do you explicitly need to send
  • 00:49:59
    the message what should be the subject
  • 00:50:01
    what should be the body nothing I'm just
  • 00:50:03
    creating this action and adding it um
  • 00:50:07
    and now we have our agent ready with the
  • 00:50:10
    trigger added and the action is getting
  • 00:50:12
    added so let's see if we can uh you know
  • 00:50:16
    send an email into the inbox into my
  • 00:50:19
    inbox here and see if this agent gets
  • 00:50:22
    triggered so I have
  • 00:50:24
    another email that I'm going to used to
  • 00:50:27
    send this uh email to my test account in
  • 00:50:30
    the demo tenant so I'm just saying hey
  • 00:50:32
    I'm Emily Brown I recently purchased a
  • 00:50:34
    product and I want to go return that so
  • 00:50:37
    the email is sent from my Emily Brown
  • 00:50:41
    account let's just give a second so the
  • 00:50:44
    email shows up
  • 00:50:46
    here perfect we see that email in the
  • 00:50:50
    vasavi test account right so I've
  • 00:50:53
    received this email from my customer and
  • 00:50:57
    who is wanting to return that product I
  • 00:50:59
    think this was yes this is the screen
  • 00:51:01
    that we were still at let's see if it
  • 00:51:04
    got this trigger automatically so today
  • 00:51:06
    is March 5th and at 108 a.m. we have
  • 00:51:10
    received this email and here's the
  • 00:51:12
    trigger and here's the live moment uh
  • 00:51:15
    you know praying again to the demo Gods
  • 00:51:17
    hoping this works as
  • 00:51:19
    expected okay we see that the trigger is
  • 00:51:22
    detected and as a first step it is
  • 00:51:25
    looking through the knowledge sources
  • 00:51:27
    based on the instructions that we
  • 00:51:29
    provided again the same sequence of
  • 00:51:31
    steps right from the email it got the
  • 00:51:34
    customer name that is Emily Brown and it
  • 00:51:38
    knew that it had to go look into the
  • 00:51:40
    knowledge sources to go figure out who
  • 00:51:42
    is that customer and what is the product
  • 00:51:44
    that the customer uh ordered so I'm
  • 00:51:48
    showing that here again so first it
  • 00:51:49
    looks into the available data wasas
  • 00:51:52
    knowledge product order customer
  • 00:51:55
    identifies the product or based on the
  • 00:51:57
    customer name and the customer ID and
  • 00:51:59
    you see how it initiated the return flow
  • 00:52:02
    in this example you see Emily was the
  • 00:52:05
    one who sent the email not paven so it
  • 00:52:07
    was able to fetch from the data that
  • 00:52:10
    Emily uh you know ordered so and so
  • 00:52:12
    product on February 13th and here's the
  • 00:52:16
    order ID so it initiated the return
  • 00:52:18
    request got all the confirmation details
  • 00:52:20
    and finally sending the confirmation
  • 00:52:23
    email to the customer asking you know
  • 00:52:26
    you you need to uh return the product by
  • 00:52:28
    so and so date so you see how the agent
  • 00:52:31
    was able to kind of orchestrate through
  • 00:52:33
    all these actions without a lot of uh
  • 00:52:36
    you know just through PR instructions it
  • 00:52:38
    was able to get all of these uh and you
  • 00:52:41
    uh quickly I know we we spent a good
  • 00:52:43
    amount of time on the demos and at 10:00
  • 00:52:46
    uh 10:10 this time you see how the
  • 00:52:48
    customer also received this email uh
  • 00:52:50
    asking to return by the specified date
  • 00:52:54
    so with auton autonomous agents you can
  • 00:52:56
    think about like how you can automate
  • 00:52:58
    your repetitive task even like simple
  • 00:53:00
    initiate return request you could
  • 00:53:02
    automate that and save a lot of time for
  • 00:53:05
    yourself um with that pav over to you if
  • 00:53:08
    you want to add something more here and
  • 00:53:11
    we can get into the next parts of the
  • 00:53:13
    session yeah yeah just key takeaways I
  • 00:53:15
    think uh V we we we just basically you
  • 00:53:18
    know agents can initiate these tasks
  • 00:53:19
    autonomously based on any external
  • 00:53:21
    trigger you have full visibility on what
  • 00:53:23
    the agent is doing it's actioning the
  • 00:53:26
    Deep reasoning that it's going to be
  • 00:53:27
    employing uh and then you know you could
  • 00:53:29
    you can automate parts of the
  • 00:53:32
    experiences you needed right so you
  • 00:53:33
    could have a very healthy you know um uh
  • 00:53:37
    handoff handback between agents and
  • 00:53:39
    humans uh for your business processes so
  • 00:53:42
    hopefully you were able to see how this
  • 00:53:45
    platform is evolving so that those use
  • 00:53:48
    cases that we kind of you know uh
  • 00:53:50
    flirted at the start of the session gets
  • 00:53:53
    you an idea of how to go build them in
  • 00:53:55
    the tools so uh it was it was amazing
  • 00:53:57
    set of uh demos to kind of show the full
  • 00:54:00
    spectrum and hopefully you guys are now
  • 00:54:02
    energized to go in and and and really
  • 00:54:04
    think about how you can kind of start
  • 00:54:06
    building these things
  • 00:54:11
    [Music]
タグ
  • AI
  • Microsoft
  • Copilot Studio
  • Automation
  • Business Transformation
  • Agents
  • Webinar
  • Productivity
  • Data Integration
  • Customer Experience