There’s a type of task at work
that everyone has done,
but almost no one enjoys.
For example:
I want to take parental leave,
but where do I start?
Workday?
HR Portal?
Payroll?
Ask my manager?
Or open a Ticket first?
It’s the same with password resets.
What you really want is just:
“Help me recover my account.”
But company software often first requires you to know:
Which system this belongs to.
Workday now wants to reverse this order.
The new approach is:
You first say what you want to accomplish.
The system then figures out:
Which system to use,
Which agent to contact,
Which company rules apply,
And hands it all over to:
Sana.
Sana Becomes the New Workday Home
On October 13, Workday announced
that Sana officially becomes:
The new Workday Home.
So when employees open Workday,
the entry point is no longer just:
Menus.
Dashboards.
Task lists.
Instead, it’s an:
Agentic Workspace.
Sana centralizes:
Tasks.
Decisions.
Meetings.
Agent Activity
all into one place.
Employees can simply state in natural language what they want to complete.
Example: I Want to Take Parental Leave
Workday provided an example scenario.
An employee says:
They’re expecting a child in November,
and wants to know:
How long can they take off?
How is the pay calculated?
What are the next steps?
Previously, this might require:
Finding the company Leave Policy →
Checking personal Time-off balance →
Asking Payroll →
Notifying the manager →
Filling out forms →
Going through approvals.
Sana Concierge is designed to:
First find the applicable policy,
Pull in the employee’s Workday data,
Summarize:
Leave eligibility,
Pay impact,
Next steps,
And coordinate:
HR,
Payroll,
Manager approvals.
The employee doesn’t need to know
which system handles each piece behind the scenes.
This Is Sana’s Biggest Difference from Typical AI Chatbots
Typical enterprise chatbots often answer questions like:
“What is the company’s parental leave policy?”
And respond with:
A policy summary.
Sana goes one step further:
After answering, it actually gets the task done.
Examples Workday lists include:
Expense approvals.
Leave requests.
Password resets.
Onboarding.
Offboarding.
Payroll questions.
Application access.
In other words, it moves from:
Answer
to:
Action.
Workday Calls This Concept “AI is the New UI”
This phrase is worth noting.
Corporate software interfaces used to be:
Menus,
Buttons,
Forms,
Dashboards.
Users had to first know
where to go.
In the future, Workday wants the experience to be:
You simply describe the outcome.
For example:
“I want to take leave.”
“Process this expense for me.”
“What’s pending for new hire onboarding next week?”
“I can’t access this system.”
Sana then determines:
Which agent will handle it,
What information is needed,
Which workflow to trigger,
And when to escalate to a human.
If this model matures,
the biggest change in enterprise software may not be:
Fewer buttons,
but rather:
Users no longer need to know how many systems are behind the scenes.
But Sana Doesn’t Do Everything Itself
This is easy to misunderstand.
Sana functions more like:
An orchestration layer.
It takes the employee’s objective,
then identifies the correct:
Workday agent
to complete each part of the task.
On the same day, Workday announced new Autopilot Agents across:
HR,
Finance,
IT,
including Recruitment,
Payroll,
Accounting,
Compliance.
In other words:
Sana handles:
Understanding what you want to do,
Coordinating tasks,
And recalling you only for parts requiring your action.
Dedicated agents handle:
The actual enterprise processes.
Employees Don’t Need to Know Which Agent to Contact
This may be even more important than:
“Having many AI agents.”
If a company deploys
20 agents,
and employees still have to decide:
“Should I ask the HR agent, Payroll agent, or IT agent?”
That just replaces the app menu
with an agent menu.
The problem isn’t truly solved.
Workday’s current approach is:
Employees only interact with Sana.
Sana then decides
who to call next.
This is closer to a true:
single work entry point.
How Are Permissions Handled?
One of the biggest concerns with enterprise agents is:
What data and functions AI can touch.
Workday says agents inherit the company’s existing:
Policies,
Permissions,
Approval rules.
For instance, if you currently don’t have permission to approve your own expenses,
Sana won’t suddenly grant you new finance permissions just because you used natural language.
Processes requiring manager approval
still route through approval workflows.
Workday calls this design:
Lawful Agents.
“Lawful” Is Not a Legal Free Pass
This boundary is important.
“Lawful Agent”
describes Workday’s product architecture.
It means agents are designed to work according to company
policy,
permissions,
and approvals.
It does NOT mean:
Every agent action is legally guaranteed by courts or regulators.
If the company’s
permissions are misconfigured,
policies are outdated,
data is incorrect,
or approval workflows flawed,
AI could still efficiently carry out incorrect actions
based on erroneous rules.
So when deploying these agents,
the real priority remains:
Making sure original company rules are clean and accurate.
When Do Humans Still Need to Step In?
Workday’s design is not to hand over all HR/Finance
work entirely to agents.
They clearly state companies can decide:
Which tasks run automatically,
Which require review,
How far agents can go,
And when a final human decision is required.
For example:
Password resets
may be well suited for full automation.
But:
Payroll exceptions,
major compensation changes,
personnel decisions,
financial anomalies,
may still require
human judgment.
A truly mature process is not:
Humans completely out of the loop.
It’s about clearly distinguishing between
“Places that don’t need humans”
and
“Places that absolutely require humans.”
Sana Is Also Expanding to Claude, Gemini, Copilot, Teams, and Slack
Another notable direction is:
Workday doesn’t want employees to only find Sana from within Workday.
The company also announced Sana will be available in:
Claude,
Gemini Enterprise,
Microsoft Copilot,
Microsoft Teams,
and Slack.
This means you could be working in Teams,
and say directly,
“I want to take leave next Friday.”
Behind the scenes, Workday will:
Confirm policies,
Retrieve data within your permissions,
Run the correct workflow,
Without you needing to leave Teams, open Workday, and hunt for the right function.
Workday and Anthropic’s integration even states
that Workday data and business rules remain safely within Workday,
with Claude only receiving
the necessary context to complete specific steps.
This design is crucial.
Because for enterprise AI to cross apps,
it doesn’t mean each app gets access to all company data.
So The Real Question in the Future May Not Be “Which AI App Is Strongest”
But rather:
Can your work context follow you?
Currently, common enterprise situations are:
Information is in Workday.
Chat is in Teams.
Documents are in Drive.
Analytics in another AI tool.
Every time you switch tools,
the context resets.
If Sana-style agent orchestration matures,
the ideal state becomes:
Wherever you work,
the workflow follows.
But your company’s
permissions,
policies,
and approvals
remain enforced.
This is much closer to what enterprises can realistically accept
than “throw all data into one super AI.”
One Important Availability Difference Today
The new:
Workday Home with Sana
is already available,
but company admins can still decide
when to enable it,
and whether to set it as the default entry.
The more full-featured:
Sana Concierge—the idea of “solve all HR/IT requests first through Sana”—
is still in:
Early Adopter phase.
Workday expects
general availability in Q1 2027.
So it’s not accurate to say today:
“All Workday customers can now automatically handle all HR/IT tickets with Sana.”
Not yet.
Can Anyone Try This Now?
This isn’t:
ChatGPT,
Claude,
or Gemini—
consumer tools.
It is essentially:
part of a company’s Workday environment.
Whether you can use it depends on:
Whether your company is a Workday customer,
Which products your company has purchased,
Whether admins have enabled it,
And your role and permissions.
This article isn’t about:
“Go download Sana now.”
It’s meant for companies to watch how
work software entry points are evolving.
The Most Valuable Lesson for Companies Isn’t Workday’s Product Itself
Even if your company doesn’t use Workday at all,
Sana’s design embodies an important concept:
Don’t start from:
“How many AI agents can we build?”
Start from:
“How many different systems does an employee have to check to complete one task?”
For example:
New hire onboarding,
Taking leave,
Expense reporting,
Password resets,
Checking pay,
Requesting software access.
Identify those processes that are:
Most common,
Have the clearest rules,
Have fewest exceptions,
And are easy to verify completion.
Those are probably the best candidates
to hand over to agents first.
For Example, Password Resets Are a Better Starting Point than Compensation Decisions
Both are HR/IT tasks,
but they carry very different risks.
Password Reset
The rules are relatively clear.
Identity verified.
Conditions met.
Execute.
The outcome is easy to check.
Compensation Decision
Involves:
Performance,
Market pay,
Fairness,
Manager judgment,
Budget,
Employee relations.
Even if AI can gather all information,
it doesn’t mean AI should decide alone.
This shows why:
Breaking down tasks is more important
than
“How powerful the model is.”
The Most Important Aspect of Workday Sana Is Not Just the Added Chat Window
If you think today’s announcement means simply:
“Workday has made an AI chatbot,”
you’re underestimating it.
The real change is that enterprise software workflows may be shifting.
Before it was:
I want to complete a task →
Find the system →
Find the function →
Fill out data →
Open a ticket →
Wait for someone to process it.
Now it aims to be:
I want to complete a task →
**Just say what you want →
AI finds the workflow →
AI handles what it can →
Brings a human in only when needed.**
If this approach succeeds,
the most important interface in the next generation of enterprise software
may no longer be:
beautiful dashboards,
but rather:
You just say what you want to accomplish.
The complex system knowledge users used to need
gets gradually hidden behind agents.
