Hey everyone, in this video I want to

introduce what work IQ is

and why we would actually use it.

So if I think about what we're probably

most used to today.

What we really focus on a lot of the

times is the idea of our M365 solutions.

So if I consider hey great

M365

we used to thinking about exchange,

SharePoint, Teams

and all that data

is accessible through the Microsoft

graph. And also there's data with for

things like Entra, Intune and more which

we have long been able to interact with

programmatically.

So what we're looking at is the various

types of data like our email messages, I

can think about our documents in our

OneDrive, our SharePoint. We have our

spreadsheets of data, we have meeting

transcripts,

we have chats in Teams.

We have all these different types of

data. We also have data that's

integrated through various types of

connectors. That's supposed to be a

puzzle piece.

Um

but you have all those Copilot

connectors.

And also what it's going to do is

because of integration with things like

Entra, it understands certain business

relationships, hi who my manager is,

what is the org chart.

And then for all of this data we protect

it. There's permissioning, we apply

sensitivity labels, we have information

protection.

And then we had this availability and

widespread use of generative AI.

And what that meant is in addition just

interacting with the graph API to grab a

document and maybe do sort of keyword

searches on things. We have to use

natural language because in natural

language what is complicated, one word

can mean many things, many words can

mean the same thing.

And so traditional lexical searches when

you're just looking for that keyword, it

doesn't meet the need.

Instead we want to search based on the

meaning of what we're looking for and

the meaning of the data we had.

And so this is where

the semantic

index

came in.

So these are high-dimensional vectors

that represent the meaning of data. So

that's embedding it's called is created

for all of the different chunks of data

stored in a semantic index and then when

I'm looking for something it takes what

I'm looking for

and it converts that to an embedding

that high-dimensional vector and it

looks for data that really is the

closest neighbor to it so it has the

closest meaning.

And that's how we find data using

natural language that relates to what

we're looking for.

And that semantic index is widely used

today by M365 Copilot for finding the

relevant data to ground requests on. And

it's still checking for the permissions,

it's checking the sensitivity labels, it

enforces the same permissions and

labeling if it goes and creates things.

But just being able to find the closest

data

really isn't enough.

Because I want to understand how data

relates to other data, how people relate

to data, how people relate to other

people, the rhythm of business.

That's a completely different set of

requirements.

And if you think about all the things I

just said, it's what we have in our

brains.

We have that mapping of okay well these

this PowerPoint and this Word doc and

that spreadsheet I know it's about this

project which relates to that meeting. I

just had a chat with Bob about that. So

we have this rich graph of relationships

and understand the full context. So if I

want now agentic agents to be able to

maybe take over certain parts of tasks

or entire workstreams or help me

I need AI to be able to have that same

type of context. And that is work IQ.

That is the whole goal around it.

So

how does this come into play?

Well firstly when I think about the

vision for what is work IQ and a caveat,

not everything I talk about today at

time of recording is available but it's

on its way.

So when I think about the data that is

in scope of this, absolutely yes the

M365 data but I'm also going to think

about things like my Dynamics 365 data.

I'm going to think about Power Apps.

I'm going to think about Power BI.

And kind of those things will grow and

increase over time. Think of all the

Copilot connectors of which there are

hundreds. We have all of these different

things and ways to bring in data.

And I can really think of this whole set

here as our data layer.

while understanding, while respecting

all of the data permissions, those data

sensitivity labels.

And as I mentioned if I create a doc off

of data that has a certain sensitivity

label like highly confidential, the

output will respect and maintain that

labeling.

So this is our data layer, fantastic.

The next thing we now think about is

really where this this whole power and

what's new to what we're used to comes

in

is this idea of

context.

The inferred understanding which again

we do in our brains and now I want for

agents. Now a large part of this is

memory.

And this is going to span all of these

different areas. So we think about this

context.

And a huge part is memory.

And also activity. So the activities

that is performed. So when I think of

memory, that's going to persist across

conversations, across interactions to

enable

agents that are using work IQ to have

very rich context and be able to infer

information. So think about my chat

history. Think about the activities I

perform, what I do in Outlook, tasks I'm

doing, what happened in meetings, who I

work with and about what. That is the

full context. I talked to Jane about

Office of the CTO core business.

We are working on this PowerPoint about

the next field trip.

And what it's also going to start to

learn is what are the most talked about

items between certain parties.

Which ones have the highest priority?

What are most meetings about? Well that

must be a high priority project. So all

of these different signals get fed into

this idea of memory. And I can think

about that as an implicit memory that's

being built.

And then I think as time goes on

relevancy changes. So that memory will

move with you over time.

So it will start to say the most recent

things, well they will have more

relevance. So it still maintains older

memories

but in terms of what it's going to focus

on with relevance, it's going to be this

sliding window of the most relevance.

It's going to rewrite relevance

as time goes on. So the things I'm doing

this week

are likely

It's going to get even better about the

things I do over a month period for

example.

I can also do explicit memory.

So I can configure custom instructions

to tune how it responds. I can ask it to

remember certain types of things. So I

can actually say hey remember I like X.

We can see this. So if I jump over for a

second

to here I'm just looking at the work

Copilot chat experience.

If I go and look at my settings and look

at my personalization

we'll firstly notice

I can add custom instructions

for how I want it to respond, how I want

it to act.

But it's also got this idea of saved

memories. These are explicit things I

have called out

about how I want it to act.

So in my case I told it once to remember

I prefer brief factual responses, I

don't want a ton of emojis.

I can delete certain memories, I could

delete all of them.

Notice it's also here as thinking about

hey chat history. We're going to use

this about personalizing our responses

as well. So there's this whole concept

here of memory both explicit and

implicit.

And then it's going to use this as part

of its inferencing because it has a

real-time understanding of work, of

actions, of relationships. It's going to

learn who has specific skills based on

work patterns. What is the right next

action?

It can prompt me

based on the context it has gathered and

predict things like who are the people

I'm most likely going to want to work

with about something. What is most

likely file I'm going to want to use. So

if I jump back over here for a second,

I could say

um regarding

um Mars base, set up a meeting

with and I just do {slash}

and it's saying these three people.

Because these three people

are the people that have had meetings

and worked on documents about a Mars

base. So, it has worked out that

context.

I could also go and look at different

files. So, there's like oh villain

threat briefings and it's understanding

the context of exactly what I'm doing to

help me with the various tasks I want to

do.

And the big focus here is this is

primarily based around

personal memory.

There are contexts of maybe larger

organization, but for the most part this

is all about personal memory.

And then there's also this idea

of business

understanding.

And what this is doing is it's really

building a semantic index

and ontology

on top of my Dynamics data, my Power

Apps Dataverse data, either structured

data.

So, it's going to learn and understand

that structured data as well, which is

really important. So, I think about I

want to ask questions about my systems

of record, my Dynamics 365 sales, my

Dynamics 365 customer service. Hey, help

me evaluate issues related to I don't

know inventory, sales. That will all be

available as well. That's that complete

understanding of kind of how we work and

that state of the business.

And then what it adds

as the next layer

is I can think about this idea

of skills and tools.

So, we'll add in

skills

tools.

For example,

there's a whole layer of fine-tuned

models that power certain agentic sub

processes. Think about um there's a

process optimized for deep search

retrieval.

There's one for better fidelity creating

office documents, creating a meeting,

many more. They're basically there to

help the agentic processes have a super

high fidelity output.

So, the skills help describe what to do.

Now, I'm going to do a little

demonstration. I'm going to show this in

the chat experience, but I could do

exactly the same thing in Outlook, for

example.

So, I could say here just looking at the

chat

um recommend

how to resolve any conflicts in my

calendar tomorrow.

So, this is hooking into Work IQ. It's

looking into the skills

and it's understanding.

So, it's identified various conflicts

and it's got recommended resolution

plan. So, hey move or shorten the prep

time,

prioritize the kryptonite removal from

planet. It's a shorter window.

Maybe I could have partial attendance.

So, it it's giving me

options for how I can handle that

because it understands calendars. It

understands the options it could do.

And then hey, do I want to apply these

changes? It will actually go and make

those changes for me. That's really the

whole key point about these sets of

capabilities.

I could ask it to draft an executive

summary of a certain project, highlight

the purpose, the progress, the

implications.

I can in PowerPoint maybe ask it to

create a slide summary. In fact, let's

go and do that one. So, if I go over

here again, actually this time I'll go

into PowerPoint. Again, I could use the

local app,

but I've opened up

Copilot

in PowerPoint.

So, I might say create a single slide

summary

of my Word

document

related to a hero base on Mars.

So, now is it's working out

okay, how to find the right Word

document, i.e. the the full context and

inference. It's checking here and it's

put the most likely one at the top. So,

I'm just going to confirm that is the

right document.

Then it might give me some other options

maybe about style I want to do and other

stuff. And then it will go ahead and

create the content.

I'm just going to skip this. I don't

care about the format.

So, now it's going to review the Word

document.

It's going to inference based on what it

believes is the most important points

from it.

And then it can use its skills

to create that very high fidelity new

slide in this case and add it to my

deck.

Okay, so it's finished and well,

a nice one-slide summary

of the top five considerations for my

base on Mars for my superheroes. It

doesn't seem

um

totally convinced it's a good idea, but

hey, whatever.

I can ask it other things. Hey, I've got

to sync with some person, prep me using

our last two one-on-one meeting

transcripts, the latest version of this

product overview. Give me a crisp

agenda, three specific things I should

update them on and two tough questions I

should ask to ensure we stay on track.

So, it's all about really

massive amounts of context and

capabilities to hook into many different

things to help me have the best

experience. And then for tooling, I'm

hooking into MCP server tools, APIs,

plugins, agent flows, Power Automate.

So, tools do the work.

And I can bring my own tools. These are

going to evolve over time.

And then I think about this and I've

tried to sort of demonstrate this. All

of these

are then getting used by the various

experiences.

So, obviously we think about M365,

the Copilots there, Dynamics 365.

Your own agents though. So, the whole

point here is we have this idea of the

Work IQ API.

So, everything I'm doing here around

everything,

I can create my own agents.

Copilot Studio, Microsoft Foundry and I

can have GitHub Copilot helping me do it

but other platforms as well.

Now, I already did a separate video on

using Work IQ with GitHub Copilot CLI,

so you could check that out in detail,

but just to give you a super quick idea

of just directly hooking in via the Work

IQ API.

So, I'm going to fire up Copilot.

It's going to automatically load in the

Work IQ

plugin and skill because I've already

configured that for the environment. And

I'm just going to ask it to do

something.

So,

schedule a meeting with Clark

next Tuesday

to discuss Mars

base.

And it will work out that okay, I need

to use the Work IQ skill. You can see

it's doing that already because it's

about calendar appointments.

Then it's asking it hey, I want to

schedule a meeting with Clark. It's

worked out the date.

And then it's going to go ahead and

actually schedule that meeting with Work

IQ. So, this is just using the Work IQ

API from outside the standard kind of

M365 experiences or the chat.

And hey, I could then carry on.

Let me know if you need any adjustments

and it will go and do all of that.

So, that's just another example of how

we can

leverage it.

And then on top of this,

>> [laughter]

>> the whole big

set of capabilities is going to grow

over time,

but we have multiple models. So, all of

these different things it is

multimodal.

So, yes, we obviously think about the

Open AI models.

We think about the Anthropic

models. There's There's others as well.

And over time, hey,

the

the best models to use for a certain

type of hey, long reasoning cycle or

really quick interaction is going to

change. Today, the time of recording,

Copilot co-work is made possible by late

2025. There are now very complex

reasoning models that can reason for a

really long time, i.e. multiple days

before going off the rails, but they're

100% going to change over time and 100%

it's not just one right model.

Different models are good for different

sets of circumstances.

And really the the key point here is

that as I obviously I'm going to create

agents.

That's the whole goal of all of this.

We have agents

that are going to

work autonomously

or even provide assistance

without context

they're really limited to what they can

do. So Work IQ supplies that full

context of how we work, our data. When I

build in Copilot Studio, I just get all

this automatically.

And yes, this is part of the Microsoft

experiences, but now you can bring it to

your own agents.

So how I work is available to any of the

agentic capabilities you choose to do.

Now, as I mentioned again, not

everything is here at the time of

recording, but hey, it it's coming in

the next couple of months if it's not

there today.

So go ahead, go and play.

Uh go and light up your agents to

maximize what they can do

by using Work IQ for that full how we do

business, how we work, and then

obviously Foundry IQ

for that exposure and usage of

institutional knowledge, Fabric IQ, hey,

the state of your business. So all those

things together is that complete 360 of

how my organization functions. I hope

that was useful. Till next video, take

care.