Enterprise software buyers want more than impressive demos. They want reliable systems. Workday has responded with a new research team focused on enterprise AI.
Workday AI Research will study trust, control, and efficiency in AI agents. These systems can remember context, make decisions, and act for employees. That creates fresh challenges for HR, finance, IT, and communications teams.
For VoIP engineers and IT leaders, the issue feels familiar. Automation only scales when monitoring, governance, and recovery work well. AI agents need the same discipline before businesses trust them deeply.
One major focus is agent memory. Workday researchers built a more selective memory method. It keeps useful information and removes weak, duplicated, or outdated details. In tests, it raised precision by 12 percent. It also improved overall memory quality by about 8 percent.
The system retained 97 percent of important memories. It also ran around 31 percent faster than the AI comparison model. That speed matters when agents support live business processes.
However, memory also creates risk. Workday tested whether an agent can truly forget information. It found deleted information remained recoverable from older summaries about one in five times. Full removal required deleting those summaries too.
That finding should interest compliance and security teams. Enterprise AI cannot simply “forget” because a user requests it. Vendors need stronger controls over stored context, summaries, and audit trails.
Workday also tested multi-agent collaboration. This means several specialized agents share a complex task. The company found this approach improved accuracy by 5.8 percent. Every final answer also met its defined constraints during the study.
This design could help enterprise workflows. One agent might check policy. Another could review data quality. A third could prepare the final action. Still, more agents also mean more places for errors to appear.
Gartner has warned that many agentic AI projects may struggle. It predicts more than 40 percent could face cancellation by late 2027. Rising costs, unclear value, and weak risk controls drive that outlook.
Megan Barker, Senior Manager, Talent Acquisition at Workday, said: “In areas like HR and finance, enterprise AI needs total trust and precision, but off-the-shelf models simply aren’t built to solve challenges like privacy, auditability, efficiency, and accuracy.”
She added: “That’s why we launched Workday AI Research, tackling the toughest technical hurdles, from persistent agent memory to multi-agent collaboration and explainability, all backed by peer-reviewed science.”
Workday will also launch a PhD fellowship program. It will offer $50,000 in annual research funding. The program includes mentorship and possible collaboration with Workday researchers.
The wider message is clear. Enterprise AI now faces its production test. Better models alone will not guarantee adoption. Businesses need agents that remember correctly, forget safely, and explain decisions clearly.
For technology teams, this marks an important shift. AI success will depend on trust engineering, not hype. Workday’s move shows how vendors may compete in the next phase.
