Priya Vijayarajendran left the CEO job by introducing herself again. After four and a half years at ASAPP, including two years leading the company, she wrote that she was moving forward and closer to the work she loves. Then came the sentence that made the announcement feel less like a corporate departure and more like a declaration of intent: “I started as a builder, an engineer. And that’s still who I am.”
The line contains a useful tension. Her career has carried her steadily away from the conventional engineer’s desk. She spent nearly 17 years at SAP, eventually leading innovation work in Silicon Valley. At IBM she became CTO and vice president of Applied AI. At Microsoft she led the worldwide Data and AI business, a portfolio that spanned Azure data products, machine learning, the Internet of Things and SQL Server. At ASAPP she arrived as chief technology officer, became president of technology, stepped in as interim CEO, and took the permanent job in 2025.
But the work she chooses to emphasize is still the work beneath the titles: deciding what should be built, translating a difficult business problem into a system, and making that system reliable enough to use. Her next organization has not been announced. Her next problem has. She wants to keep building enterprise AI at scale.
Before AI could talk, enterprise software had to remember
Vijayarajendran’s career began in the less cinematic world of enterprise applications. At SAP, she worked through application development, enterprise architecture, strategic customer engagements and innovation. This was software built around business objects, transactions, rules and exceptions. A 2014 patent names her among the inventors of a consistent interface for flags and tags in a business object model. It sounds narrow. It is also a neat artifact from a career spent making complicated systems behave consistently.
That background matters now. Modern AI can produce a graceful answer in seconds. An enterprise still has to know whether the answer is grounded in the right account, policy, reservation or claim. It has to decide which action is permitted, where an approval is required, and what happens when confidence drops. The model may be new. The organizational plumbing remains stubbornly specific.
At SAP’s Silicon Valley Innovation Center, Vijayarajendran moved closer to customers and emerging products. She later described those years as the foundation in ERP, enterprise search, SAP HANA and next-generation enterprise software. The sequence taught a lesson that would follow her: innovation is not a laboratory trick when the customer runs a large business. It must enter the system of record and work inside the business process.
“I am a technologist at heart and passionately believe in artificial intelligence that is designed to augment human capabilities.”Priya Vijayarajendran, on joining ASAPP in 2022
IBM taught the model to work. Microsoft taught the business to scale.
Her move to IBM in 2017 put artificial intelligence in the center of the job. As CTO and vice president of Applied AI, she was responsible for end-to-end solutions for enterprise customers. The phrase “applied AI” does a lot of work. It distinguishes a promising model from an application that people can use, and an application from a system that an organization can adopt. Her team built first-of-a-kind cognitive products while she cultivated a broader community of AI experts.
Two years later she joined Microsoft as vice president of Data and AI. The job expanded from building systems to moving a worldwide business. She oversaw product leadership and end-to-end sales, partner and go-to-market strategy across Microsoft’s commercial organization. Public talks from that period covered Azure Synapse, Azure Arc, modern data warehouses and the culture required for data-led transformation.
The logos make the journey look tidy. The useful thread is a shift in altitude. SAP exposed the operational core. IBM focused intelligence on real customer problems. Microsoft added global distribution. Each stage answered a different question: Can the system represent the business? Can intelligence change the process? Can the whole thing travel across industries, partners and geographies?
Customer service is an unusually honest test for AI
Vijayarajendran left Microsoft for ASAPP in March 2022 because the problem pulled her back toward engineering. Contact centers produce relentless, messy evidence of whether software understands a person. A traveler calls during a weather disruption. A bank customer asks about a sensitive transaction. A telecom customer explains a problem over several turns. The answer depends on history, policy, timing and tone, and the customer has little patience for repeating any of it.
A traditional chatbot can classify an intent or retrieve a prepared response. The larger system has to preserve context, select data, reason through a workflow, call an API, protect sensitive information, monitor quality and bring in a person at the right moment. It must do this across thousands of simultaneous conversations. “The hardest challenge,” Vijayarajendran said while at ASAPP, was maintaining context, safety and compliance across all of them.
Her definition of empathy is revealing because it avoids theater. Empathy in AI, she has argued, is about understanding context and consequence. A system does not need to imitate a person for the sake of it. It needs to recognize why someone is calling, choose a useful action, and avoid making a consequential situation worse. Accuracy and resolution become part of empathy. So does restraint.
The production test
Can the system retain context, reach verified data, act through a workflow, respect a safety boundary, and escalate to a person when confidence runs out?
This is why Vijayarajendran resists treating AI only as a way to compress costs. Her preferred frame is capacity expansion. A useful system can absorb spikes in demand, shorten resolution cycles and make new service models possible. Cost matters, but a narrow headcount calculation misses the larger value and encourages brittle automation. The better scorecard includes resolution, customer satisfaction, lifetime value and the ability to serve more people without lowering trust.
Her CEO method was proximity, not distance
Running a lean AI company changed how Vijayarajendran led. In a large organization, she said, leaders operate through systems. In a smaller company they can operate through proximity. At ASAPP that meant listening to calls, talking with customers, debating model architecture and reviewing demos. She developed a preference for fewer, faster decisions anchored in data. The logic was technical as much as managerial: ambiguity compounds.
During her departure, she named three decisions that required conviction. ASAPP consolidated separate products into a Customer Experience Platform and retired things it had already built. It elevated explainability, security and compliance as product differentiators. And it continued investing in research and intellectual property when easier routes were available. She also highlighted the pivot from software that assists human agents toward a business built around autonomous agents.
Those choices form a coherent operating thesis. A collection of clever features can impress buyers and still fail to create a system. A platform forces the harder questions: how components share context, how performance is observed, how models are swapped, how failures are contained, and who remains accountable. The shift toward autonomy raises the stakes. If software can take an action, governance must sit inside the architecture rather than around the slide deck.
“In a lean company, you lead through proximity.”Priya Vijayarajendran on her operating style
A departure with a technical specification
Vijayarajendran’s farewell was generous to the people around her. She thanked ASAPP’s board and advisers, investors, customers, and teams in New York, San Francisco, Buenos Aires and India. The gratitude did not blur the lessons. She left with an unusually specific inventory of what it takes to run enterprise AI: data foundations, governance, operational discipline, compliance, organizational change, infrastructure and constant model innovation.
That list is less romantic than a prediction about artificial general intelligence. It is also closer to where value gets made. The enterprise AI race will not be decided only by access to a model. Models change quickly and are increasingly orchestrated together. The durable work sits in the control layer: context, evaluation, integration, permissions, monitoring and the feedback loop between human judgment and machine action.
Her public ambition is now simple enough to fit in one sentence and broad enough to occupy another career: continue as a technology leader building enterprise AI at scale. There is no announced employer attached to it. There does not need to be one yet. The direction is consistent with every stage that came before.
The arc from engineer to CEO is usually told as an ascent. Vijayarajendran’s version is more circular. The executive job supplied a wider view of customers, capital, product choices and organizational change. Now she is carrying that view back toward the build. She is not discarding leadership. She is placing it closer to the work, where a difficult system either earns trust or does not.