Workday Sana AI agents show where business software is heading: permissioned systems that find answers, take action, and automate work.
Article text
Most business software AI still behaves like a polite intern.
It can answer a question, summarize a policy, draft a paragraph, or point you toward the next step. Useful, sure. But the person still has to click through the workflow, update the record, chase the approval, and finish the job.
That is why Workday’s Sana matters.
In its March 17 announcement, Workday introduced Sana as a new AI layer for Workday that can find answers, take action, build outputs, and automate workflows across HR, finance, and connected enterprise systems. The details matter: 300-plus self-service skills, connectors to tools like Salesforce, Slack, Jira, ServiceNow, Google Workspace, and Microsoft apps, plus no-code workflow automation that runs inside existing permissions.
That is a stronger signal than another chatbot launch. It is evidence that the market is moving from suggestion-based AI to execution-based AI.
If you are trying to figure out whether AI in your business should stay a helper or start becoming an operator, we do free workflow calls to help teams separate real workflow gains from expensive feature theater.
Workday is pushing past the copilot phase
The easiest way to understand Sana is to look at the four verbs Workday chose.
Find
: pull answers from Workday data and connected knowledge sources
Act
: complete tasks across systems using the employee’s existing permissions
Build
: generate dashboards, summaries, and documents from live business context
Automate
: run background workflows without code across multiple steps and tools
That framing is smart because it exposes the real gap in most business AI products.
Most products stop at find. Some make it to build. Very few can reliably act. Almost none automate meaningful work unless somebody grinds through a brittle setup or hands the process to IT.
Workday is trying to compress all four into one governed interface inside the system where the work already lives.
That is the important shift.
The question is no longer, "Can AI help me answer this faster?"
The question is, "Can AI move the work forward without breaking trust, permissions, approvals, or reporting?"
Why this matters more than a flashy demo
I think a lot of business owners are getting numb to AI product launches because the demos all look the same.
Ask a question. Get a polished answer. Watch a pretty summary appear. Maybe generate a report.
But the pain inside most companies is not a lack of summaries.
The pain is that HR requests bounce between inboxes. Manager approvals stall. Payroll and benefits updates involve handoffs nobody owns. Finance questions require digging through multiple systems. Policy enforcement lives in tribal knowledge. People know what needs to happen, but the workflow still eats time.
That is why Workday’s announcement stands out.
Sana is not being pitched as a smarter search box. It is being pitched as a permissioned execution layer inside a system of record.
That distinction is everything.
When an AI system works inside the same security model, role permissions, and audit boundaries as the business software itself, adoption gets easier. So does accountability. Instead of adding another tool that employees have to trust on faith, the AI inherits the structure the company already depends on.
We have seen the same pattern in other launches this month. In our breakdown of Oracle’s agentic applications , the big story was not simply more AI. It was that the AI lived inside the transactional system. And in our post on ADP’s marketplace for HR AI agents , the key takeaway was that governed AI adoption gets much easier when it shows up inside software companies already trust.
Workday is making the same bet, and it is probably the right one.
The useful AI layer is the one that actually moves work
There is a simple test I keep coming back to.
When a vendor says their AI helps your team, ask: what step disappears?
Not what screen gets prettier. Not what answer appears faster. What step disappears?
Workday gave a few examples that point in the right direction:
an employee asking to update their home address and immediately seeing downstream tax and benefits effects
a manager generating a dashboard from recruiting data without manually assembling it
a monthly workflow that reviews receipts against policy, prepares a report, and routes it for approval
Those examples matter because they cross the line from assistance to execution.
That line is where ROI starts getting real.
A chatbot that answers policy questions may reduce some interruptions. Good. An agent that resolves the request, updates the system, routes the approval, and leaves an audit trail changes labor cost, speed, and error rates in a completely different way.
That is also why many SMB AI projects disappoint. Teams buy a tool that looks intelligent, but the underlying workflow barely changes. They get a new interface, not a better operation.
What SMB owners should take from a Workday launch they may never buy
Most small and mid-size companies are not Workday customers.
That does not make this irrelevant. It makes it useful as a benchmark.
Here are the three signals I would pull from Sana if I were evaluating AI in any business software category.
1. Answers are table stakes now
If a vendor is still selling AI mainly as chat, summaries, and search, they are already behind where the market is going.
Those features are becoming expected. Helpful, yes. Differentiated, not really.
The next layer of value comes from action.
Can the system create the record, update the field, route the approval, trigger the follow-up, or complete the next step without forcing the human back into five separate screens?
If not, the tool may still be useful, but do not confuse it with workflow transformation.
2. Permissions and governance are not boring details
They are the product.
A lot of AI launches still treat governance like the legal section at the bottom of the page. Workday is doing the opposite. It is making permissions central to the value proposition.
That is exactly right for HR and finance.
Nobody sane wants an autonomous system touching payroll, employee records, time-off balances, approvals, or financial workflows unless it is anchored to clear permissions and traceable actions.
For SMBs, the translation is simple: do not just ask what the AI can do. Ask what it is allowed to do, who can approve it, where the audit trail lives, and what happens when it is wrong.
3. No-code automation inside business software is getting more important
One of the quietest but most important parts of the Sana launch is the automation layer.
This is where a lot of ROI will come from over the next year.
Not from one-shot prompts, but from repeatable background workflows tied to real business rules.
If your HR, finance, CRM, or operations platform starts offering no-code agent workflows inside the product, pay attention. That may be a much better path than jumping straight into a custom build.
We already make this point in our post on AI chatbot vs. AI agent : the useful distinction is not whether the interface feels magical. It is whether the system can complete meaningful work within guardrails.
Where I would stay skeptical
This is a strong launch, but it is still a vendor launch.
There are at least four things I would want proved in practice before getting carried away.
Can it handle messy edge cases?
Enterprise workflows sound clean in demos and get ugly fast in production. Exceptions, broken data, missing approvals, contradictory policies, and human workarounds show up everywhere.
The real test is not whether Sana handles the happy path. It is whether it stays useful when the process gets messy.
Does action really reduce work, or just relocate it?
Some AI systems "take action" in a way that simply creates more review work for humans downstream.
If every action still needs heavy checking, then the labor savings may be weaker than the pitch suggests.
How broad is the execution layer across connected tools?
The connector list is impressive. The harder question is how deep the action model goes across each one. Reading across systems is easy. Writing back, handling approvals, and respecting business logic across systems is much harder.
Will employees actually trust it for sensitive work?
Workday cited early customer stories about adoption and reduced ticket volume, which is promising. But trust in HR and finance automation takes time. A lot of teams will want proof that the system is accurate before they hand it anything important.
The bigger lesson for business software buyers
I think Workday’s Sana is the clearest sign yet that AI in business software has to do the work, not just answer questions.
That does not mean every business should rush to buy action-taking agents tomorrow.
It does mean your evaluation criteria should get stricter.