Previously, using Microsoft Copilot involved a workflow like this:
First, open Word.
Then call upon Copilot.
Need to calculate something:
Open Excel.
Want to create a presentation:
Open PowerPoint.
In short:
The app was the entry point.
AI was:
Just a feature within.
Microsoft now wants to invert this order.
First, enter:
Copilot.
Then tell the AI:
What exactly you want to accomplish.
Then it creates:
Word.
Excel.
PowerPoint.
Or even:
Apps.
Workflows.
Long-running agents.
This shift is the key quality of the new Microsoft Copilot.
The New Copilot Divides into Three Work Paths
Announced by Microsoft on September 25, the new architecture centers on:
Home.
Code.
Autopilot.
Don’t think of these as:
Three new chatbots.
They’re more like:
Three tiers of AI-driven work.
Home
You have a task:
Ask AI to help complete it.
Code
You need a tool:
Ask AI to build it for you.
Autopilot
This task will happen continuously:
Have an agent watch over it long-term for you.
Complexity and autonomy increase with each layer.
The Most Practical Starting Point Today Is Office Work Within Home
Because:
Code and Autopilot
are still in newer Frontier/Preview rollout phases.
Not all accounts
have access yet.
But the Word, Excel, and PowerPoint agents
are already a practical layer:
You can understand,
and start using them.
Microsoft's current official approach is:
You can directly select from:
Copilot Chat
Choose:
Word Agent.
Excel Agent.
PowerPoint Agent.
Then describe:
Your requirements.
For Example: Don't Open PowerPoint First
Assuming you need to prepare:
A client proposal.
Before:
You opened PowerPoint.
Created a new presentation.
Planned the structure.
Designed the cover.
Formatted the slides.
Slowly inserted the data.
Now, in Copilot you can say:
“Create an 8-page client proposal including issues, current situation, solutions, schedule, budget, and next steps.”
The PowerPoint Agent
will generate:
A genuine PowerPoint file.
Not just:
Eight text chunks in chat
which you have to copy and paste yourself.
The Same Applies to Word
For example:
You already have:
Meeting notes.
Project data.
Client requirements.
You can ask:
To create:
A two-page Project Brief.
Divided into:
Background.
Issues.
Solutions.
Schedule.
Next steps.
AI will first generate:
An editable Word document.
You then open:
Word
and refine the details yourself.
Excel Is Even More Suitable for “A Workbook I Can Continue Using”
For example:
“Create a monthly income and expense tracker,
with 12 months split,
plus a yearly summary and chart.”
The Excel Agent
produces not just:
Suggested column lists,
but an actual:
Spreadsheet.
This marks a big difference between:
“Answers”
and
“Deliverables.”
Generated Files Can Still Be Edited in Office
This point is crucial.
AI-generated:
Word.
Excel.
PowerPoint
files are not:
confined to chat.
Microsoft’s process is:
Generate a preview first.
Once complete,
save the file to:
OneDrive.
Then you can still:
open it in the full
Word,
Excel,
or PowerPoint
applications.
You can continue to:
edit,
tweak formats,
verify numbers,
adjust charts,
and add branding elements.
So a good way to think about it is:
AI creates the first draft file.
It doesn’t produce:
The final deliverable in one go.
Microsoft Clearly Advises: Review Before Official Sharing
This is common across all generative AI:
Just because
the format looks complete,
doesn’t mean
the data is correct.
Especially for:
Dates,
Names,
Amounts,
Formulas,
Charts,
Conclusions,
References,
or client commitments.
Before sharing externally,
manual verification is essential.
The ideal use case is not:
“AI finishes it for me; I don’t check.”
But rather:
“Instead of starting from a blank file, I start from a first draft.”
Who Notices the Difference Most?
Not:
Someone who opens Word for five minutes a day.
Rather, those who:
frequently produce different outputs
from the same context.
For example:
Project managers.
One meeting’s data
that later evolves into:
A project brief,
budget,
and management presentation.
Previously:
The same context
had to be moved three times.
Now:
Ideally, Copilot can
grasp the task first,
then generate separately:
Word,
Excel,
and PowerPoint outputs.
Small Companies Also Benefit From This Approach
For example, a
three-person event studio.
Today they receive:
Client needs,
event budget,
timeline,
venue info.
Previously, one person made the proposal,
another managed Excel calculations,
and finally, someone reorganized it into PowerPoint.
The new Copilot approach is to let these three differently formatted deliverables
all start from the same AI workspace.
This is
Office in Copilot’s
real innovation beyond just “an extra Copilot button inside Word.”
Microsoft 365 Copilot License Affects Context Access
Microsoft officially states:
Word, Excel, and PowerPoint agents
aren’t reserved only for
enterprise Copilot license holders.
Microsoft 365 users,
including:
Personal,
Family,
and Premium plans,
can access these agents under supported conditions.
However,
the amount of context data differs.
Accounts with Microsoft Copilot License
Agents can use the files,
emails,
chats,
meetings,
and other enterprise context
you have permission to access
to generate content,
all while respecting
existing company access policies.
Without a Copilot License
You can still
work via web,
and explicitly provide
files
for the agent.
Personal accounts can use
their OneDrive files,
notes,
and personal data
as context.
So, it is incorrect to say:
“If both have Word Agent, they have the same abilities.”
The real difference is often:
How much contextual work data the agent can access.
Traditional Chinese Is Already Supported in Copilot Languages
This is reassuring for users in Taiwan.
Microsoft officially lists
Traditional Chinese
among
Microsoft Copilot’s
supported languages.
This includes
prompt understanding,
grounding,
output,
and UI localization.
However,
it’s important to distinguish between:
Language support
and
Feature rollout.
Traditional Chinese support does not mean
all new features announced yesterday
have appeared on every Taiwan account today.
Microsoft also reminds:
New features
may not launch simultaneously
across all regions and languages.
If You Don’t See Home/Code/Autopilot in Copilot, Your Setup Isn’t Broken
The new Copilot
is still
rolling out gradually.
Especially:
Code
and Autopilot
are advanced features that also require governance and usage controls.
So the most practical approach today is
not to ask:
“Why don’t I have Autopilot yet?”
But instead, check if your Copilot includes:
Word,
Excel,
and PowerPoint
Agent/editing functions.
If yes,
test them first
with documents you regularly create.
Don’t Start With a Large Report for Your First Test
The best first task is:
Something you already know the answer to,
and
is easy to verify.
For example:
Turning meeting notes
into a two-page project brief.
Or:
Turning an expense record
into a simple Excel summary.
Or:
Turning a confirmed plan
into a six-page presentation.
This way you can quickly see:
Where AI saves time,
and where it errs.
Don’t Give a Vague Task on Day One
For example,
“Make me an impressive presentation.”
AI still has to guess:
Who is the audience?
How many slides?
What problem does it solve?
Do you want numbers?
What style?
A good prompt should include:
Type of deliverable.
Target audience.
Essential content.
Approximate length.
Data sources.
For example:
“Using this client interview notes,
create a 6-page PowerPoint for my manager.
Page 1 shows the problem,
Pages 2-4 summarize the three main findings,
Page 5 lists risks,
and the last page shows next steps.
Don’t add any numbers that aren’t in the data.”
Such clear instructions
are more likely to produce
a usable first draft.
Home’s Real Goal: Stop Making People Decide Which App to Open First
Previously:
To write:
You opened Word first.
To calculate:
You opened Excel first.
To make a presentation:
You opened PowerPoint first.
The new Copilot
starts by asking:
“What do you want to accomplish?”
For example:
“Explain this project to my manager tomorrow.”
Copilot can then
organize data,
conduct analysis,
produce documents,
create presentations
all from a single work entry point.
This shifts the work style from
App-centric
to
Task-centric.
Code Takes It One Step Further: Build an App When None Exists
The second path:
Code.
Microsoft positions this to let more knowledge workers
build apps, dashboards, automations,
and workflows using natural language.
It uses behind-the-scenes technology related to
GitHub Copilot.
The interesting part isn’t:
“Every office worker needs to become a developer.”
It’s that many tasks
get stuck because:
“This needs a small tool, but it isn’t important enough for the IT roadmap.”
For Example: An Internal Company Tool
“I need to combine three CSVs daily into one report.”
“I want a dashboard to track event registrations.”
“I need a simple quote calculator.”
“Five people need to update project status together.”
Before, you might look for:
Excel,
Notion,
Airtable,
Power Apps,
or ask engineers.
Code’s approach is to:
Describe what tool you want,
and have AI build it.
But Code Isn’t Ready for Everyone to Fully Rely On Yet
Microsoft clearly groups
Code,
Copilot Managed Runtime,
FinOps,
and usage-based billing
within the same enterprise framework.
The reason is simple:
When you create an app,
a workflow,
or automation,
companies will ask:
Who made this?
What data does it use?
Where does it run?
Who can access it?
How much does it cost?
How to turn it off?
So once AI coding enters enterprises,
it must be accompanied by:
governance.
Autopilot Goes Even Further Than Code
The third path:
Autopilot.
This is a persistent agent,
formerly called:
Scout.
The big difference is:
It doesn’t wait for you to prompt it.
It can have its own:
identity,
memory,
workspace,
and permissions,
continuously working
in the background.
For example:
Checking important meetings every morning,
preparing materials,
spotting risks,
or coordinating schedules.
You may not have Copilot open,
but it can still be working for you.
This Is Why Autopilot Cannot Be Viewed as Just a “More Powerful Chat”
Chat:
One question at a time.
Autopilot:
For a whole day,
or even long-term.
It doesn’t need:
longer prompts,
but clear:
goals,
permissions,
boundaries,
budgets,
audits,
and owners.
This is why Microsoft keeps these capabilities
in a more controlled frontier/preview environment for now.
For General Users, It’s Best Not to Chase All Three Layers At Once
The new Copilot offers many things:
Home,
Cowork,
Office,
Code,
Autopilot,
Managed Runtime,
Microsoft IQ,
Work IQ.
Don’t try to learn everything at once.
Segment by difficulty:
First layer
Create deliverables.
Word,
Excel,
PowerPoint.
Second layer
Create tools.
Code.
Third layer
Continual automated work.
Autopilot.
If you haven’t yet found
a truly time-saving first layer task,
there’s no need to rush
to always-on agent automation.
The Most Worthwhile Test Today
Take a task you already need to do today.
Don’t open a
blank Word, Excel, or PowerPoint document.
Instead, enter
Copilot,
select the corresponding agent,
then provide the real context,
and request a first draft.
Once it’s generated, test just three things:
First
How much time do you save
going from blank
to first draft?
Second
What content
ends up kept?
Third
How long do manual corrections take?
If
you originally spend 45 minutes on a presentation,
AI generates it in 20 seconds,
but you redo it for 40 minutes,
you haven’t saved 44 minutes and 40 seconds.
The real time saved is only:
5 minutes.
Do Not Use “AI Generation Speed” as ROI
The most important metric is:
Total time from need to deliverable.
This applies to all agent workflows.
Fast AI
does not equal faster work.
If
there are many errors,
the context is wrong,
and formatting has to be completely redone,
the total time
does not decrease,
no matter how shiny the tool is.
What’s Most Worth Noticing About New Microsoft Copilot
It’s not:
“There are three new names.”
It’s that Microsoft
is redesigning the product so that:
AI becomes the entry point.
Office apps:
are output formats.
Code:
builds tools as needed.
Autopilot:
delegates ongoing tasks as needed.
This breaks from the old
“A Copilot button in each app”
logic.
Lastly, A Very Practical Rule to Remember
If your work is:
You need a deliverable today.
Focus on:
Home and Office agents.
If your problem is:
The company doesn't have the proper tool at all.
Look into:
Code.
If your problem is:
This task repeats daily or weekly.
Then start thinking about:
Autopilot.
Don’t pursue full automation immediately
just because AI can be more autonomous.
A truly good AI tool
does not:
remove humans completely from the process.
It
helps humans avoid starting from scratch each time.
Today, improve a little with AI.
Learn an AI skill each day.
Save a bit of time daily.
Enhance your abilities incrementally.
SasaDaily, growing with you.
