Editors: Raymond D. Aller, MD & Dennis Winsten
Agentic AI vs. RPA: assessing their boundaries and benefits
November 2026—Ask a random group of pathologists if they are familiar with lab automation, regardless of whether they use it, and you’re more than likely to get nods of affirmation all around. But follow that up by asking them if they are aware of the pros and cons and ins and outs of robotic process automation versus agentic artificial intelligence and many of those nodding heads may go still.
For this reason, Bryan Dangott, MD, director of clinical informatics at Mayo Clinic, Jacksonville, Fla., tackled the topic of RPA and agentic AI in a presentation at the United States and Canadian Academy of Pathology’s annual meeting last spring. In a nutshell, he said, RPA has been around for decades and is best suited for well-defined, repetitive tasks, such as verifying insurance and filing structured reports. Agentic AI, on the other hand, is much newer and better suited for more nuanced, goal-oriented tasks, including advanced document searches and multistep prior authorizations.
Because the right system, or combination of systems, can accelerate laboratory documentation, research, analysis, and other functions and labs face challenges that are unique to their specific operations, it is important for pathologists to understand which of the technologies will best meet their needs.
“With highly deterministic pathways, you know all the steps, you know all of the inputs, there’s structured data—that favors RPA,” says Dr. Dangott. “If it’s vague and has a lot of variability, such as workflows with images or handwritten elements, that’s not something that works well for RPA. You need to have additional reasoning and logic.”
The technology has other advantages and limitations as well, Dr. Dangott notes. RPA is fast, handles standardized data quite well, and is relatively easy to implement, but its hardcoding is often unsuited for more challenging tasks, and such systems can hit a wall with complex exceptions.
Agentic AI, on the other hand, is nimble and can better manage difficult tasks and exceptions. The increasingly popular technology can learn from its failures and attempt other approaches until it succeeds. But putting these systems into practice can be complicated, and they need closer human-in-the-loop supervision. While agentic AI may think it deftly handled an exception, a human evaluator might feel differently, Dr. Dangott says.
To get the best of both worlds, some laboratories choose a hybrid approach, incorporating RPA and AI as a single solution. In this scenario, RPA extracts well-scripted data while agentic AI is layered on to manage complex situations in which RPA might fail. Because the AI agent is well-suited to handle variability, it can better handle data that are more haphazardly organized. However, labs shouldn’t exclusively gravitate to the most technologically advanced solution, Dr. Dangott notes.
“RPA still has a place,” he says. “It’s easier to implement and, because it’s deterministic, it’s easier to monitor, and deployments can be more straightforward. Agentic systems might have more capabilities for vague scenarios, but they also have a little bit more variability in the output, so they need to be monitored more closely.”
As a hematopathologist, Dr. Dangott frequently reviews smears. He found that RPA can readily handle structured CBC data. However, free-text notes and clinical history can be gray areas. The RPA may not have access to an EHR or other external system and, therefore, won’t be able to answer questions about a patient’s clinical history, he explains. The RPA can handle the simpler steps before the AI agent takes over to gather the clinical context and complete the report for the clincian’s review.
RPA and agentic AI can be co-deployed as retrieval agents as well. Laboratories routinely pull turnaround time reports from structured query language databases, making RPA well-suited for this task. AI agents are better suited to gather data from documents that are not indexed and that may be stored in folders with hundreds or thousands of files, such as standard operating procedures, protocols, package inserts, and similar information.
“An agent can go through and parse the repository and then sit at a user interface level,” explains Dr. Dangott. “The user can then interact with the interface to query the database to search for specific documents. The agent can retrieve those documents and pull them up without having a formal database structure.”
Agentic AI can also be used to assess the quality of digital slides. Image-analysis algorithms can identify slide defects, such as dust or focal plane issues. The technology eventually may be able to take this a step further, quantifying how much of the slide has been impacted and whether those issues obscure the diagnostic area.
“You need to have a more advanced agent that’s not just looking strictly at a defect, but it’s understanding that this is potentially diagnostic tissue or it’s not diagnostic tissue,” says Dr. Dangott. “That’s an area where it might streamline workflow.”
Critically assessing workflows before taking steps to automate or add AI is essential to determining which type of technology can best address a laboratory’s intrinsic challenges, Dr. Dangott says.
As part of this process, labs should work with their information technology team to understand their organization’s IT infrastructure and potentially take advantage of tools used by other areas of the institution. The organization may already have internal solutions, and the lab simply may need the appropriate permissions to access them.
For laboratories that want to purchase such technology, creating requests for proposal can be time-consuming, albeit a good way to define the lab’s actual needs, Dr. Dangott says. Will the system be accessing patient health information or administrative data? Will it be on the cloud or hosted internally? It’s important to remember that the solutions organizations choose must not only meet the lab’s needs but also align with IT support capabilities and infrastructure.
Laboratories that are less technology focused may want to put their toes in the water before diving into agentic AI, Dr. Dangott notes. For example, they could initially use the technology to produce administrative rather than clinical reports.
An institution’s IT group ideally can help the lab understand the system’s failure modes, set up robust security and permission structures, and create strong guardrails to control how the agent accesses and deploys data, Dr. Dangott says. For example, instead of allowing the AI to automatically implement a process that will change a system or database, laboratories should first test and understand the potential downstream impact of such a change.
Dr. Dangott encourages laboratory decision-makers to familiarize themselves with all aspects of these technologies before deciding whether to implement them. Pathologists may want to play with open-source RPA and agentic systems outside of work, for instance, to better understand a system’s advantages and limitations. They should also consider attending presentations covering these topics, such as those increasingly being offered at informatics and pathology conferences.
A lot can be gained from listening to conference presenters “who have already done something close to what you’re doing,” says Dr. Dangott. “And networking is another way to get started.”
—Josh Baxt
Agilent to purchase Proscia
Agilent Technologies has entered into a definitive agreement to acquire the digital pathology software provider Proscia.
Agilent, a provider of pathology instrumentation, reagents, and services, will acquire Proscia’s vendor-agnostic Concentriq platform, among other solutions, and established customer base of more than 4,000 laboratories, hospitals, and clinics.
“The acquisition establishes an Agilent platform for scientific intelligence in tissue-based precision medicine, connecting data generation with AI-based interpretation in a workflow that supports biomarker research through routine clinical use,” said Agilent president and CEO Padraig McDonnell, in a company press statement.
The announcement builds on Agilent’s 2022 multi-year distribution partnership with Proscia, which combined Agilent’s staining and reagent solutions with Proscia’s Concentriq Dx enterprise digital pathology platform.
Upon completion of the merger, Proscia will become part of the Agilent Life Sciences and Diagnostics Markets Group.
Agilent Technologies, 800-227-9770
Clinisys adds document intelligence to lab system
Clinisys has integrated document-intelligence capabilities into its Clinisys Laboratory Solution to help laboratories convert information contained within documents into structured data that can be connected directly to laboratory workflows. These capabilities include optical character recognition, intelligent classification, metadata extraction, and automated indexing.
The new functionality is intended to help laboratories capture information from requisitions, reports, scanned documents, images, and other sources; classify and organize information automatically; and make content searchable and discoverable.
Clinisys, 520-570-2000
Lifepoint Informatics partners with Interpret My Results
Interpret My Results has formed a partnership with Lifepoint Informatics that combines the former’s artificial-intelligence analysis capabilities with the latter’s clinical data-exchange infrastructure, giving Lifepoint Informatics clients easy access to Interpret My Results’ platform.
The patent-pending Interpret My Results solution supplements laboratory reports with patient-focused explanations of test results and related findings. It integrates with laboratory systems via HL7, Fast Healthcare Interoperability Resources, and application programming interfaces.
Interpret My Results is a wholly owned subsidiary of Computer Service and Support.
Computer Service and Support, 609-653-6444
Fujifilm releases new version of pathology system and AI foundation model
Fujifilm Healthcare Americas Corp. has introduced version 3.0 of its Synapse Pathology picture archiving and communication system, as well as Synapse Pathology AI Foundry, a large-scale pathology-specific artificial-intelligence foundation model trained using millions of pathology images.
With Synapse Pathology version 3.0, “we listened closely to our customers and translated their feedback into meaningful enhancements across navigation, measurement, classification, and organization, while simultaneously creating new opportunities to bring AI directly into the pathology workflow,” said Mark Lloyd, PhD, vice president of digital pathology for Fujifilm Healthcare Americas, in a press statement.
Introduced in this latest version are enhancements to the company’s proprietary algorithms, including TissueSync and TissueDirect, and new capabilities for simplifying access to third-party AI applications, including Ibex’s AI-powered solutions for detecting malignancy in prostate, breast, and gastric specimens.
The new release incorporates the Synapse AI Orchestrator, Fujifilm’s automated hub that streamlines the exchange of information between Synapse Pathology and compatible third-party AI applications, creating an infrastructure in which laboratories can incorporate specialized algorithms into their workflows.
Synapse Pathology AI Foundry, introduced at Pathology Visions 2026 last month, can support multiple downstream applications and will serve as Synapse Pathology’s foundation for various current and future pathology AI capabilities.
Touted by Fujifilm as “the world’s first foundation model for screening in pathology,” the solution is designed to identify and distinguish normal, inflammatory, atypical, and malignant conditions in gastrointestinal specimens. It is available for research use only.
Leica expands Aperio AI store with Tempus products
Leica Biosystems has added two Paige digital pathology products from Tempus AI to its Aperio AI store, a unified interoperable digital pathology platform that provides multiple artificial intelligence-powered applications in a single interface.
New to the AI store are Paige PanCancer Detect, a research-only application for identifying suspicious areas across more than 40 types of cancer, and Paige Prostate, a suite of AI-powered applications for assessing prostate cancer features in H&E-stained whole slide images.
“The Aperio AI store is designed to simplify and scale AI adoption by bringing partner algorithms directly into the Aperio Halo AP software workflow,” reported Leica. “Instead of navigating separate software interfaces for each algorithm or vendor, users can access a broad and growing selection of curated AI applications through one consistent experience.”
The AI store is available exclusively through the Aperio Halo AP anatomic pathology image-management system. Aperio Halo AP, developed in partnership with Indica Labs, is available for research use only in the United States.
Leica Biosystems, 312-565-6737
Aiforia products available on PathAI marketplace
Aiforia Technologies has announced a commercial agreement with PathAI under which Aiforia’s artificial-intelligence-driven pathology solutions will be available through PathAI’s AISight marketplace, a curated digital ecosystem hosted within the AISight Dx digital pathology image-management system.
The agreement pairs Aiforia’s deep-learning technology and position as a provider of CE-IVD–marked pathology AI solutions in Europe with PathAI’s installed base and market presence in the United States, thereby extending the reach of Aiforia’s AI models across both regions of the world.
AISight Dx is CE-IVD marked for primary diagnosis in the European Economic Area, Switzerland, and the United Kingdom.
PathAI, 617-500-8457
Dr. Aller practices clinical informatics in Southern California. He can be reached at rayaller@gmail.com. Dennis Winsten is founder of Dennis Winsten & Associates, Healthcare Systems Consultants. He can be reached at dennis.winsten@gmail.com.