Google Just Announced the Future of EMS Documentation. They Just Don’t Know It Yet.
Robert Grand · Battalion Chief who still runs calls
Yesterday at Google I/O, Sundar Pichai stood on stage and declared the agentic era of AI officially open.
Voice-native document creation. Background agents running 24/7. AI that takes your verbal brain dump and turns it into a structured document while you move on to the next thing. Google called it the future of how people get work done.
He’s right. He’s just not talking about us.
What Google Announced
The headlines from I/O are worth understanding, because they’re not abstract.
Google announced a feature called Docs Live (https://blog.google/innovation-and-ai/sundar-pichai-io-2026/), where you talk out loud and Gemini drafts the document. No prompts, no typing, no structured input required. You say what’s on your mind, the agent does the assembly work. They also announced Gemini Spark, a personal AI agent that runs continuously in the background, completing long-horizon tasks under your direction, integrating with your tools, acting on your behalf.
The combination of those two things, voice input plus agentic task completion, is the architecture that transforms post-call documentation in EMS. Not someday. The architecture exists right now.
Google built it for someone writing a memo. But the workflow is identical to what a medic needs after a cardiac arrest.
The 150 Words Per Minute Problem, With an Upgrade
My first article (https://www.linkedin.com/pulse/why-fire-services-biggest-technology-leap-wont-come-from-robert-grand-k1xlc/) made the case for voice as the primary unlock in EMS documentation. A medic who just ran a call can reconstruct it in three minutes of talking. That same medic navigating drop-downs and free-text fields takes 25 minutes and still misses clinical nuance, because the form didn’t have the right field for what actually happened.
150 words per minute versus 20. That gap hasn’t closed.
But voice alone was always incomplete. Talking fast doesn’t automatically produce a compliant, billable, defensible PCR. Someone still had to do the structured work, populating the required fields, cross-referencing the protocol, flagging the billing codes, checking for completeness against your medical director’s documentation standards.
That’s where agents change the equation.
Voice captures the narrative. The agent does the assembly. You talk through the call, the system extracts the structured data, maps it to your PCR fields, flags what’s missing, and surfaces a draft for your review. You spend four minutes confirming and signing instead of 25 minutes reconstructing from memory.
That is not a future product concept. That is a current capability being deployed for people writing Google Docs.
The Honest Problem With Voice Adoption
Here’s something I don’t hear people say out loud: most of us don’t actually use voice input, even when we know it’s faster.
I write articles about voice-native workflow and I still default to typing. On my phone, on my laptop, drafting texts. My hands go to the keyboard before I’ve made a conscious decision. The argument in my head is immediate: typing feels more controlled, more precise, easier to edit in real time.
That feeling is a habit, not a fact.
Voice input feels slower mentally even when it’s measurably faster physically. The friction isn’t in the technology. It’s in the rewiring. You’ve spent decades routing “I need to communicate something” straight to your fingers. Switching that default takes intentional repetition, not just awareness that the better option exists.
The medic I watched pacing the parking lot outside the ED, dictating his SOAP narrative into his iPad, didn’t discover a philosophy. He discovered a habit that worked. He looked a little ridiculous doing it, and he didn’t care, because he was done charting before his partner finished pulling the stretcher.
That’s the bar. Not elegant. Not seamless. Just done, and done well, before the next call drops.
If you’re reading this and thinking “I should try voice input more,” the answer is yes, and the way to do it is to start somewhere low-stakes and private, voice texts, quick notes, short emails, until your brain stops treating it as a novelty and starts treating it as the default. Two weeks of intentional reps. That’s the cost of the transition.
Your crews will face the same resistance when this technology reaches your agency. The vendors who understand that will build onboarding that accounts for it. The ones who don’t will ship a voice feature nobody uses.
What “Agency-Trained” Actually Means
Here’s where Google’s announcement stops being relevant to us, and starts being instructive.
Gemini Spark knows everything. It can write your memo, manage your calendar, draft your emails, browse the web on your behalf. It is a general intelligence layer built for a general user.
Your agency’s documentation agent needs to know your everything.
Your protocols. Your formulary. Your medical director’s specific documentation preferences. Your geographic response zones and the mutual aid agreements that affect how you chart transfers. The shorthand your medics actually use versus what the drop-down menu says. The difference between how your agency documents a working arrest versus a VSA, and why that distinction matters for billing, for outcome tracking, and for the after-incident review when something goes wrong.
A general AI writing EMS reports is like handing a blank form to someone who’s never been on a call. The output is technically a document. It is not a defensible PCR.
The agencies that win in the next five years won’t be the ones who adopted AI fastest. They’ll be the ones who adopted AI that was trained on their operations, their language, their outcomes. That data stays with them. It improves their tools. It encodes their institutional knowledge instead of exporting it to a vendor’s training pipeline.
Google proved yesterday that the technology is ready. The only remaining question is who builds it with the operational context to make it actually work in the field.
The Call Is Coming
The agentic era isn’t coming to EMS. It’s already here. The infrastructure exists. The voice models are accurate. The agent frameworks are production-ready. Google just deployed them to hundreds of millions of people who write memos for a living.
Your medics are still charting in the parking lot.
That gap is not a technology problem. It hasn’t been for a while. It’s a domain problem, and domain problems don’t get solved by the biggest company in the room. They get solved by someone who has been in the room where the call actually happens.
The fire service deserves tools built by people who’ve lived it. That standard hasn’t changed. The urgency just went up.
Robert Grand is a Battalion Chief at Eugene Springfield Fire with 24 years in the fire service. He writes the Frontline Intelligence newsletter.
If this landed, share it with a chief who needs to hear it.
Subscribe
From the floor, not the vendor booth. Two times a week.