The following is a guest article by Rishi Nayyar, Co-Founder and CEO at PocketHealth
Radiology has a habit of getting there first. It digitized earlier than most departments, built the modern imaging stack (PACS, RIS, modality worklists), and over the last decade became healthcare’s most experienced “AI buyer,” piloting everything from CAD-adjacent tools to triage and workflow optimization. Now the next wave is arriving: agentic AI—software that can interpret context, make decisions, and take actions across systems even when workflows are messy and exception-heavy.
If you want to know where agentic AI will first become an operational layer inside hospitals, don’t look to the enterprise. Look to radiology. Not because it’s trendy, but because radiology sits at a rare intersection of acute operational pain and high technology readiness. That combination makes it the natural proving ground for non-clinical agentic workflows—and what works here won’t stay confined to imaging.
Radiology’s Administrative Burden is Unusually High–and Measurable
Radiology’s operational footprint looks less like a department and more like a logistics business running at hospital scale. Orders arrive from everywhere—ED, inpatient units, specialty clinics, screening programs, outside referrers—and each order can trigger a chain of non-clinical work: eligibility checks, protocoling, prior auth, labs, contrast screening, scheduling constraints, prep instructions, and follow-up routing. In practice, the “edge cases” are the workflow.
One consistent fault line is inside versus outside. A director at a large academic medical center told me their internal order process is straightforward, but outside orders quickly become cumbersome—especially once prior authorizations enter the picture. That’s why point solutions keep failing: the work isn’t a single transaction. It’s navigating exceptions across organizations, payers, and systems.
Last month, an access operations director at a Midwest health network shared a number that should make any COO pause: roughly 400,000 inbound calls a year for imaging scheduling alone. That isn’t customer service. It’s a system using phone calls as an integration layer because workflows can’t coordinate themselves.
In competitive markets, this becomes a race. An operations executive at a large Southwestern system told me the push to automate booking is partly financial, but mostly about speed. Whoever reaches the patient and gets them scheduled first wins. Access is revenue. Access is retention.
Enterprise Fixes Often Miss Radiology’s Nuance
Radiology is frequently poorly served by enterprise initiatives that lack imaging-specific expertise. Centralized scheduling and general call centers may work for templated appointments. Imaging does not work that way. Contraindications, contrast questions, sedation requirements, lab prerequisites, protocol changes—these details matter.
One radiology operations leader at a major West Coast academic system described a familiar tradeoff: manual scheduling persists in interventional radiology, but centralized scheduling is “the contract we’ve got.” When departments lose confidence in scheduling quality, they compensate with manual checks and workarounds. Those workarounds quietly become infrastructure.
Radiology is Exceptionally AI-Ready
While other departments debate definitions, radiology has spent over a decade evaluating AI with a pragmatic lens: governance, validation, workflow fit, and skepticism toward anything that adds clicks. That experience matters. Agentic AI shouldn’t be judged by model metrics. It should be judged by how reliably it performs in messy reality: incomplete orders, contradictory notes, patients who don’t respond, protocols that change, and humans who override.
Agentic AI Fits Radiology Better than Traditional Automation
Rule-based automation is brittle. “If X, then Y” collapses under the weight of exceptions—and radiology is all exceptions.
A nuclear medicine supervisor at a major academic system described a recurring failure mode: patients schedule without the department ever seeing the actual script. Days later, the team discovers the scheduled study doesn’t match the order. In PET imaging, that mismatch turns into real waste: the patient arrives, a dose is prepared, and the script indicates a different study.
Communication breaks in predictable ways too. A nuclear medicine clinical director described spending significant time in operations meetings on one issue: getting correct instructions to patients who don’t use the portal or are outside the system. A frontline coordinator described the daily burden of communicating with Spanish-speaking patients—calling a translation line, translating, leaving a message, then repeating the process when the patient can’t answer.
Then there’s the “too many hands” problem. One imaging ops leader told me six people may touch an exam before a technologist ever sees it, with 15–20 team members doing work that could, in principle, be automated.
Agentic AI is designed for this terrain. Not a single-task bot, but a workflow actor that can interpret unstructured inputs (faxed orders, notes, call transcripts), reason about next steps (what’s missing, what study is intended, what constraints apply), take actions across systems (create tasks, request missing items, propose slots, send instructions in-language), and adapt when the plan breaks (retry, escalate, route with context).
Radiology is the Wedge into the Broader Health System
Radiology isn’t just an early adopter. It’s the launchpad. Imaging is high volume, cross-system, exception-driven, and operationally central. If agentic AI can coordinate non-clinical radiology workflows reliably, it becomes a blueprint for operational intelligence across service lines.
Radiology is where agentic AI stops being a demo and becomes infrastructure—and once it’s infrastructure, it spreads.
Rishi Nayyar is the Co-Founder and CEO at PocketHealth, the agentic AI workflow automation platform transforming health system operations. Trusted by more than 900 hospitals and imaging centers across North America since 2016, PocketHealth uses agentic AI and intelligent workflow orchestration to automate non-clinical work across the patient journey — from referral intake and scheduling to image exchange and patient communications — helping healthcare teams reduce manual work, redeploy staff, and deliver safer, more efficient care.
The following is a guest article by