Imagine a company with 90,000 employees.
Now the boss announces:
Everyone can have their own AI Agent.
It’s not just for answering:
“What is this email about?”
Instead, you can directly tell it:
“Help me organize this matter, find the necessary information, and prepare the next step.”
The AI can work across different business systems,
decide on the intermediate steps by itself,
and finally hand over parts requiring human judgment.
Sounds like
this should be the ultimate goal of enterprise AI.
But Cisco’s actual experience is:
Giving AI to everyone is just the beginning.
The real challenge is:
Should we keep the entire original workflow at all?
Cisco is rolling out MyAgent to about 90,000 employees
On August 27, Cisco unveiled the latest progress on MyAgent.
This is a personalized AI Agent built on Cisco’s internal AI platform, Circuit.
Cisco says
MyAgent is being deployed to approximately 90,000 employees.
Its biggest difference from typical chat-based AI is that
it goes beyond:
answering questions,
writing summaries,
and generating content.
Cisco positions it as capable of entering “supervised autonomous workflows.”
For example, it can work across
Outlook,
Webex,
Jira,
SharePoint,
and other enterprise systems.
Employees don’t tell the AI every step to take.
They just communicate:
what the goal is,
the context,
and what final result they want.
The AI then determines:
the necessary intermediate steps.
This is where AI Agents start to differ from traditional AI assistants.
From “do this one step” to “get this entire task done”
Suppose an employee has a meeting with a client tomorrow.
The traditional AI usage might be:
Open email yourself.
Find recent emails from the client.
Copy them to the AI.
Ask AI to summarize.
Open meeting notes.
Copy that content to AI.
Then check project progress in Jira.
Organize all data into your own documents.
Only then prepare for tomorrow’s meeting.
At every step,
AI might help.
But the entire workflow
still requires humans to move data around.
MyAgent wants to do it differently.
Employees can simply state the objective:
“Help me prepare for tomorrow’s meeting with this client.”
Then the AI, based on permissions,
gathers the needed information across systems,
organizes context,
identifies outstanding tasks,
and prepares a result for the employee.
The real change is not speeding up one step.
It is:
who coordinates the entire workflow.
But then the company faces an even bigger challenge
Suppose an original task has ten steps.
The most intuitive AI adoption approach is often:
Add AI to step one.
Add AI to step two.
Add AI to step three.
…all the way to step ten.
Each step becomes 20% faster.
Sounds reasonable.
But Cisco’s public enterprise AI experience highlights one key point:
Don’t just improve the original ten steps with AI.
Ask first:
Are all ten steps still necessary today?
This is the core difference between “using AI tools” and an “AI-native workflow.”
A simple example: If three people confirm something, do you still need three people?
Imagine a company processes a batch of customer requests daily.
The original process might be:
Person one receives the email.
Person two categorizes the content.
Person three checks CRM.
Person four finds historical records.
Person five compiles a work order.
Person six assigns the responsible party.
Finally, a manager verifies.
Previously,
this was necessary because information was scattered.
So people had to:
search,
copy,
paste,
forward,
categorize,
and confirm.
Now if AI can:
read emails,
identify customers,
search past records,
organize requests,
and draft work items,
then the question becomes:
Do you still need six people to shuffle information around?
Maybe the new workflow is simply:
AI organizes first.
The responsible person decides.
The manager only intervenes in high-risk situations.
If so,
AI is really eliminating
not just “five minutes in one step,”
but
the entire process that used to exist just to move information.
This is why “everyone having ChatGPT” doesn’t mean the company is AI-transformed
Many companies are in this situation.
Employees have access to:
ChatGPT,
Claude,
Gemini,
Copilot.
And they genuinely use them daily.
Yet the company’s work processes barely change.
Email is forwarded as usual.
Excel data is copied as usual.
Meetings still happen the same way.
The same data is entered repeatedly across different systems.
AI makes employees:
write faster,
search faster,
and organize faster.
But no one questions why the original tasks exist.
This could lead to a strange result:
Everyone becomes more efficient, but the company still faces just as much friction.
Cisco has already passed the stage of “everyone just using AI”
MyAgent is not Cisco’s first day introducing enterprise AI.
In July this year, Cisco shared other internal data.
The company reports that
over 100,000 users have adopted its internal AI platform Circuit,
with approximately 90% employee adoption.
Cisco also reported that
over 21,000 engineers use AI coding tools,
with 80%+ using them weekly.
Cisco estimates engineers save roughly six hours per week,
and overall employees about five hours.
These are Cisco’s internal figures,
not independent external studies.
But more importantly,
it’s not about “saving a few hours weekly.”
It’s about what Cisco does next.
They don’t stop at:
“Great, everyone is using AI.”
Instead, they shift to the question:
Can the work itself be redesigned?
The truly hard question after saving five hours
Assuming AI really saves you five hours weekly.
The next question is:
Where does those five hours go?
If you just handle 20 tasks a day before,
now handle 25,
that’s a productivity gain.
But a company can also ask:
Why were there 20 tasks to begin with?
Are five of those just:
looking up data,
re-entering data,
forwarding information,
waiting for confirmation,
or duplicating documents?
If you can eliminate these tasks entirely,
the value is greater than just “doing things faster.”
So when enterprise AI truly enters the second stage,
the question shifts from:
“What can AI help employees do?”
to:
“Which work shouldn’t be done the old way at all?”
The first tasks AI Agents are likely to replace are those “nobody really wants to do”
For many people,
most of their day isn’t spent on professional work.
Instead, it’s fragmented by many small tasks like:
finding yesterday’s files,
confirming the latest version,
turning meeting notes into action items,
copying data from system A to system B,
checking who hasn’t replied,
tracking progress,
organizing weekly reports,
preparing materials before meetings.
Usually these aren’t why
the company actually hired you,
yet they consume a lot of time.
This is where AI Agents can create value first.
Not by making the final expert decisions,
but by removing the work of:
searching,
organizing,
coordinating,
and moving information.
But that doesn’t mean AI should take over the entire workflow
This is another point Cisco emphasizes.
MyAgent is not designed for:
AI doing whatever it wants.
Cisco’s phrase is:
“From human-in-the-loop,
to human-in-control.”
Meaning:
AI can handle more execution,
but humans still:
set goals,
make key judgments,
and take responsibility for results.
Because once AI moves from “answering questions” to “taking real actions,”
the risks increase.
If AI mis-summarizes an email,
humans can correct it.
But if AI
changes project status,
sends messages,
updates systems,
or adjusts work assignments,
the stakes are completely different.
So a mature AI Agent
doesn’t mean removing humans completely.
It means finding:
which steps AI can automate and which must pause for human judgment.
This isn’t just about Cisco’s 90,000 employees
You might think:
This is only a concern for a company like Cisco.
Our company only has:
3 people,
5 people,
or 10 people,
how can we possibly redesign workflows?
Actually, small companies may see this more clearly.
For example, a five-person company gets daily customer inquiries.
Someone copies the content into Excel.
Someone checks inventory.
Someone looks up previous quotes.
The boss then confirms prices.
The real question isn’t:
“Should we buy an AI Agent?”
It’s first about charting the workflow:
What do we receive?
What happens next?
Where do we find data?
Which steps repeat daily?
Which steps are just copying and pasting?
Which steps require the boss’s decision?
Once you map it out,
you’ll often see:
Not all the work can be handed over to AI.
But certain steps
are perfect to delegate first.
The tasks most worth automating aren’t the most complex
Many people first want to use AI
for the most complex tasks.
Hoping AI will:
help decide strategy,
negotiate with clients,
or run the entire business automatically.
But the easiest first successes
often come from very boring tasks.
Tasks that happen daily,
have clear rules,
where errors are easy to spot,
and problems can be stopped quickly.
Such as:
sorting attachments,
categorizing inquiries,
identifying missing info,
summarizing meetings,
comparing versions,
drafting documents,
and reminding about to-dos.
These tasks don’t sound glamorous.
But skipping them even once a day
adds up over months
and truly changes work.
The first step in AI-driven workflows isn’t picking AI tools
This might be the most important lesson from Cisco’s story for most people.
AI tools change daily.
Today it’s one model.
Next month, a new Agent appears.
If you start each time by asking:
“What can this new AI do?”
you’ll always be chasing tools.
A steadier approach is to start with your own work.
Find out:
What repeats daily?
Where is the data?
Which tasks just move data around?
Which require judgment?
Which mistakes are serious?
Which results can be easily verified by humans?
Then ask:
“Is there a step here that can be handed to AI?”
This way, AI comes into your work
instead of your work constantly chasing AI.
The real value in Cisco’s 90,000 AI Agents is not the number
Cisco’s rollout of MyAgent to about 90,000 employees
is certainly a massive enterprise AI experiment.
But the story worth focusing on isn’t:
“Cisco has 90,000 AI Agents.”
It’s that Cisco is moving enterprise AI from
answering questions
to
actually completing work.
When AI truly starts finishing tasks,
companies must begin to ask:
Which steps should AI handle?
Which are for humans?
Where must the process pause?
Who owns the final outcome?
Out of ten original steps,
how many are really needed?
These questions will determine whether enterprise AI delivers:
a faster chatbot,
or
a genuinely different way of working.
Future AI-native companies
might not be those with the most AI users,
but those that, one day looking back,
find many daily tasks they used to do
no longer need to do.
Today, grow alongside AI.
Learn one AI skill every day.
Save a bit of time daily.
Improve your abilities a little every day.
SasaDaily, growing with you.
