In a new conversation with our partners at Derive Technologies, Aiva founder and CEO Sumeet Bhatia joined Derive's Gemma Lolos and Tara Brown, RN, MSN, to talk about what voice AI actually changes at the bedside.
The throughline: nurses lose their shift to a dozen fragmented systems, each with its own login and screen. Aiva collapses them into one conversational layer. With its new MCP connector and A2A agent, Aiva is the first voice platform to let ChatGPT, Claude, and Gemini take real action across 40-plus hospital systems, from charting in the EHR to filing a work order or pulling up a policy. One nurse's sentence can trigger four tasks at once.
The results back it up: in a pilot using Aiva’s own app as the interface, Cedars-Sinai Medical Center saw documentation delays drop 80 to 90%, overtime fell 50% and Press Ganey scores rose 37%. The takeaway for leaders: start with one high-pain workflow, prove ROI in weeks – and make sure the environment underneath can carry it, which is where Derive comes in.
Watch the full 17-minute conversation.
Hospitals have spent years building out mobile for their nurses. Devices in every hand, Wi-Fi and cellular coverage throughout the building, secure messaging that replaced the pager. The hardware is there. The question we keep hearing from CIOs and CNIOs is what that device should be doing now. The answer came in Gemma, Sumeet and Tara’s discussion. Here are the key themes.
Tara, who has lived this at the bedside, framed it well. Across a single shift a nurse is doing tons of different tasks, like charting in the EHR, filing a work order when something in the room breaks, looking up a policy, messaging the team, ordering a tray and calling an interpreter. Each one is a separate login and a separate screen, and each one pulls the nurse away from the patient.
"None of it feels like documentation while you're in it," Tara said. "It feels like you're spending your shift doing everything except nursing." That fragmentation turns into charting at the end of the shift instead of during it, which is exactly when errors creep in. It turns into overtime. And it is a real reason people leave the profession.
That is the problem Aiva solves: take all those interfaces and turn them into one. A nurse just says what they want, and it's done.
The heart of the conversation was Aiva's new MCP connector and A2A agent.
MCP, the Model Context Protocol, has become the standard way an AI assistant like ChatGPT, Claude or Gemini securely reaches into an external system and takes action, instead of only answering questions. Aiva built an MCP connector and an agent-to-agent agent to interact with Claude for Healthcare, ChatGPT for Healthcare and Gemini Enterprise. So whichever assistant a health system chooses as its front end can now execute real tasks inside the 40-plus nurse-facing systems Aiva already integrates with: chart in the EHR, create a ServiceNow ticket, pull up a policy from Wolters Kluwer or Elsevier, message a colleague through TigerConnect or Voalte, connect to an interpreter.
Aiva is the first voice platform to let those frontier assistants control the hospital systems nurses touch every day. As Sumeet put it, CIOs and chief digital officers increasingly believe these assistants will become the front-end interface for all of their staff-facing systems. Aiva gives them that option now instead of making them wait, and it means IT isn't rebuilding the integration layer every time leadership wants to evaluate a different assistant.
Does it matter to the nurse which model is behind it? Tara's answer: at the bedside, not really. What matters is that it works and that it doesn't make them repeat themselves. Where it matters is one level up, because meeting staff inside a tool they already use is a real adoption advantage.
Tara gave the example that stuck with everyone. A nurse says: "I just turned Mrs. Patterson in 412. She finished about half her dinner, her call light is broken, and I need the fall prevention protocol."
That's one command, but it's four separate tasks. Aiva parses it, documents the turn and the meal intake in the EHR, files a work order for the broken call light, and surfaces the fall prevention protocol, all in parallel, all in an instant. Done manually, that's three logins and far more time. The workflow already exists. Today it lives in the nurse's head until she can get to a computer, and some of it doesn't survive the trip. That, as Tara noted, is a direct patient safety issue.
Both guests were candid that this only works on solid ground. Real-time voice won’t work in dead zones. Identity and access have to map every AI-initiated action back to a real person. The app catalog needs to be rationalized. And governance matters: clinical leadership needs a clear answer on which actions an agent can take on its own and which need a human to confirm.
That's where Derive comes in. They prepare the environment end to end, from infrastructure and device fleet strategy to network and security readiness, identity and mobile device management, and making sure what gets rolled out actually fits how a unit runs. Aiva brings the voice and AI layer and the pre-built integrations. Derive makes sure that layer performs reliably across an entire hospital, not just in a demo.
The outcomes are what make the case.
At Cedars-Sinai, one of Aiva’s flagship deployments spanning three hospitals, 1,200 beds and 2,500 nurses, nearly five million rows have been charted by voice. In the initial pilot, documentation delays dropped 80 to 90%, documentation compliance improved 4%, incidental overtime decreased 50%, and Press Ganey scores rose 37%. One med-surg nurse reported that Aivasaves her 1.5 to 2 hours per shift.
Tara's read on the numbers was sharp. Compliance usually improves only when you make nurses do more. Here it improved by making them do less, which tells you the charting was getting done late or not at all because there wasn't time, not because anyone was careless.
Looking ahead, Sumeet pointed to a simple model. You talk to one assistant on your phone to call, text or get directions, not a different one for each task. Nursing should move the same way: one conversational layer, whichever AI a health system prefers, sitting on top of everything, instead of a separate bolt-on AI for documentation, another for work orders, another for messaging, etc. It's also the healthier path for security, since every standalone app is its own attack surface. And as Tara put it, it's the only version nurses will actually use. Nobody is going to work with a bunch of different assistants.
Their advice for leaders who want to start: pick the one or two workflows causing the most pain – usually documentation or work orders – deploy the connector against those, and prove the ROI in weeks instead of years. You don't need a perfect AI strategy to begin. You need one workflow that gets meaningfully better, fast.
The full Q&A with Aiva and Derive is worth the 17 minutes. You can watch it here: Voice AI at the Point of Care.