Roche acquires Path AI, Agilent acquires Proscia. Tempus acquires PAIGE. In a carefully guided Chat GPT 6 essay, we try to figure out what it means and where we're going.
This is an AI generated report. It should not be taken as ground truth ,but as an example of currrent AI abilities to grasp, plan, and write multi-part essays.
##
Access this white paper as a 15-page PDF here.
Two Presentations, Two Acquisitions, and the Future of Clinical Digital Pathology
Dr. Maria Wilson and the meetings described below are fictional. The demonstrations are a teaching device, not accounts of actual vendor presentations or comparative product testing. Company and product descriptions draw on the public sources linked throughout.
At 1:45 p.m., Dr. Maria Wilson, chair of a large academic pathology department, looks over her afternoon schedule.
The department is preparing to go digital. Scanners have been selected. Storage and networking are being arranged. But the software decision remains unsettled: what will the pathologists actually use, and how will it change their clinical work?
Representatives from PathAI and Proscia will each have thirty minutes. At five, Maria will discuss the proposals with five departmental leaders and two representatives of the Dean’s office.
She sends both vendors the same instruction:
“Please walk us through a prostate biopsy, a breast biopsy with immunostains, and a colon resection with multiple blocks and lymph nodes. Begin when the scanned slides become available. Finish when the pathologist signs the report.”
She adds one more request:
“Show us an ordinary working day, including the things that go wrong.”
The First Presentation: PathAI
The PathAI representative begins with the names.
“PathAI is the company. AISight is our software platform. AISight Dx is our FDA-cleared image-management product for primary diagnosis.”
Maria appreciates the distinction. She had initially thought PathAI might chiefly be an algorithm that finds cancer. Instead, the proposal concerns the environment in which a pathologist could review an entire digital caseload. PathAI’s public materials describe case and image management alongside analytical applications. AISight Dx.
For the prostate biopsy, Maria asks the representative to begin with the worklist.
“How do I know all the slides are here? How do I move among the cores? Where do I see the specimen information? What happens when another level arrives after I have begun reviewing the case?”
Then she turns to analytical assistance.
“Suppose an application identifies suspicious tissue. Show me how I examine that finding, decide whether I agree, and incorporate anything useful into my report.”
She is interested in the whole sequence. A colored overlay might help, but she wants to know what it saves her from doing—and what additional checking it creates.
For the breast case, she asks how a quantitative result would be associated with the correct stain and specimen. For the colon resection, she asks how the system keeps numerous slides organized and helps the physician avoid overlooking an incomplete part of the case.
The representative describes the platform and available analytical offerings. Maria insists on separating what can be demonstrated now from what would require another product, an interface project, local development, or future software.
“Please don’t let a promising feature become an assumed capability in our purchasing decision.”
PathAI’s analytical expertise interests her. Its broader business includes developing tissue-analysis tools and diagnostic applications. But she redirects any discussion of pharmaceutical studies.
“That may matter to another committee. Today our question is whether our clinical faculty can complete patient cases more efficiently and reliably.”
With ten minutes remaining, she asks for a regulatory map.
“For each proposed function, identify the product, its intended use, the supported equipment, and its status. We need to distinguish a cleared diagnostic viewer from an authorized analytical application—and both from a research demonstration.”
Finally, she asks about report generation.
“Does the software prepare a diagnostic draft from the case? Does it populate selected findings? Or does it mainly help me review images before I dictate the report myself?”
Those are three different levels of assistance. Maria does not want the attractive phrase “AI-enabled workflow” to conceal which one she is buying.
Her first take-home lesson is that PathAI offers a clinical workplace as well as analytical capabilities. The department must evaluate how those pieces function together in the actual proposed configuration.
The Second Presentation: Proscia
At three, the Proscia representative begins.
“Proscia is the company. Concentriq is our software platform.”
Maria writes it down. PathAI makes AISight; Proscia makes Concentriq.
She presents exactly the same cases and asks exactly the same questions.
This time, she pays particular attention to the connections among images, specimen information, additional slides, analytical outputs, and the laboratory information system.
“Assume the prostate case is missing a slide.”
“Assume a breast immunostain arrives tomorrow.”
“Assume I send the colon case to a colleague at our second hospital. What does that person receive, and where does the consultation return?”
The discussion reveals the importance of work that is less dramatic than identifying cancer: assembling the correct case, preserving context, making the right material available, and preventing staff from repeatedly moving information between systems.
Concentriq’s public descriptions emphasize connecting scanners, laboratory systems, images, applications, and people. That gives Maria a useful framework for the demonstration, but she still needs the vendor to establish what the proposed installation will actually do. Agilent’s description of Proscia.
Then the representative introduces more ambitious AI capabilities.
Maria hears about automatic analyses, case context, and assembling reportable findings. She recognizes movement toward the prepared case she wants: useful work completed before the physician opens the images.
But she repeats the regulatory question.
“Which of those capabilities belong to the diagnostic product we could deploy clinically?”
That question matters. Proscia’s fifth-generation materials describe substantial AI functionality while distinguishing research-use-only capabilities from its separate FDA-cleared Concentriq AP-Dx product. An impressive demonstration does not establish that every demonstrated function is available for clinical diagnosis. Proscia’s product disclosures.
Maria ends with the same reporting question she asked PathAI:
“When I finish examining the case, how much reporting work remains?”
She asks to see how a physician would reject an incorrect suggestion, revise a measurement, and trace a statement back to its supporting tissue image. These become requirements for the evaluation, rather than assumed features.
Her second take-home lesson is that Proscia offers an analytical environment as well as image management. Calling it simply a storage or viewing company would be as misleading as calling PathAI simply a cancer-detection algorithm.
The Five O’Clock Meeting
Maria faces the surgical pathology director, the breast service lead, the quality director, the laboratory manager, and the informatics lead. The Dean’s office has sent representatives responsible for clinical operations and finance.
She begins:
“Both companies are credible candidates for our clinical digital workplace. Neither should win because its presentation used the word ‘AI’ more often.”
She displays a short table.
| What the department needs | What must be established |
|---|---|
| Digital primary diagnosis | The proposed product and equipment configuration support the intended clinical use. |
| Analytical assistance | Each application’s scope, performance, and regulatory status are clear. |
| Efficient reporting | Findings reach the report without burdensome duplication or correction. |
| Reliable daily operations | Missing slides, additional stains, consultations, and downtime are handled safely. |
| Future draft reports | There is a credible development path, without treating future promises as current capabilities. |
The surgical pathology director asks, “What was the practical difference between the two presentations?”
“PathAI’s broader analytical-development expertise was prominent,” Maria answers. “Proscia’s emphasis on connecting the case, users, systems, and applications was prominent. But both companies work across those areas. We must compare the products offered to us, rather than buy a simplified story about the companies.”
The breast service lead asks whether better algorithms should determine the choice.
“Only if we establish that we can use them clinically and that they improve the complete task,” Maria says. “A measurement can be excellent while the process of checking it and transferring it into the report is cumbersome.”
The quality director asks about errors.
“We need to examine both the software’s mistakes and our response to them,” Maria replies. “A plausible suggestion can make us too comfortable. A confusing display can make correct information difficult to verify. Human factors are part of clinical performance.”
The laboratory manager asks who will deal with rescans, incomplete cases, and software exceptions.
“That belongs in the cost assessment. Saving physician time by quietly adding substantial technical work may still be worthwhile, but we need to see the full calculation.”
The finance representative asks the inevitable question.
“What are you recommending?”
- “A matched clinical evaluation of both proposed configurations, followed by a purchase decision against criteria agreed in advance.”
Maria does not recommend an indefinite pilot. She proposes a bounded evaluation using representative cases, routine pathologists, technical staff, and the actual laboratory-system interface. Appropriate governance would determine how the evaluation is conducted before clinical deployment.
They will measure total physician time, technical work, clinically important discrepancies, reporting effort, and the frequency of exceptions. They will examine the experience of an ordinary faculty member, not only the department’s most enthusiastic digital pathologist.
“We should give substantial weight to dependable clinical use today,” she says. “Future capability matters, but it needs a separate score.”
The informatics lead asks about data portability, interfaces, and dependence on particular scanners or algorithms.
“Those belong in the contract,” Maria answers. “We need to understand the cost and difficulty of changing an application, adding equipment, exporting our information, or eventually changing platforms.”
A Dean’s representative asks whether the acquisitions settle the question.
“Roche backing PathAI and Agilent backing Proscia matter,” Maria says. “They may strengthen development, regulatory, and support resources. But neither corporate name proves that our installation will work well. We need product-level evidence and contractual commitments.”
Finally, the surgical pathology director asks:
“When do we get the system that has already reviewed the case and written a useful first draft?”
“That is the destination I want us positioned to reach,” Maria says. “But we must distinguish buying today’s digital workplace from buying that future reporting capability.”
Her recommendation is firm: select the system that best performs the department’s clinical work under a fair comparison, while preserving a credible route to more advanced assistance.
“The deciding demonstration,” she concludes, “is what happens when one of our pathologists opens an ordinary case on an ordinary Monday.”
The Alternate Universes: What If the Buyers Had Been Reversed?
The acquisitions can be understood more clearly by imagining the opposite pairings.
Roche agreed to acquire PathAI; Agilent agreed to acquire Proscia. Both transactions build on earlier commercial relationships. Roche and PathAI had established a particularly close collaboration in AI-enabled companion diagnostics, while Agilent had distributed Proscia’s software since 2022. Roche–PathAI; Agilent–Proscia.
But neither pairing was technologically inevitable.
In an Agilent–PathAI universe, Agilent would acquire AISight as a potential clinical workplace and PathAI’s capabilities for developing analytical and diagnostic applications. It could seek to connect its tissue-diagnostics products with software that interprets resulting images and ultimately helps prepare reports.
For Maria, that might look quite similar to the PathAI presentation she just attended. The ownership change would principally affect the resources, development priorities, commercial arrangements, and integration choices behind the product.
This hypothetical transaction would also have to address PathAI’s existing Roche agreements. Acquisition would not automatically remove contractual obligations. The 2024 collaboration explicitly described exclusivity in companion-diagnostic algorithm development within stated terms and boundaries. Collaboration announcement.
In a Roche–Proscia universe, Roche would acquire Concentriq’s clinical workflow infrastructure, relationships, and expanding analytical capabilities. Roche could seek to combine those with its staining systems, scanners, diagnostic assays, and internally developed or partner-supplied algorithms.
Again, Maria could encounter a recognizable Concentriq proposal. The difficult work would concern product integration, support, clinical evidence, and deciding which functions to develop next.
The hypothetical swap exposes an important point: both targets bring a digital workplace, and both have analytical ambitions. PathAI’s analytical-development business and Proscia’s emphasis on connecting the digital workflow are useful distinctions, but they do not divide the companies into mutually exclusive categories.
Roche’s own investor materials emphasize AISight image management as part of the PathAI acquisition. Proscia’s materials describe increasingly sophisticated models and automated work within Concentriq. Neither transaction is adequately described as buying only algorithms or only infrastructure.
For a clinical department, the ultimate question survives the swap unchanged:
Which combination will deliver a supported clinical system that performs useful work before the physician arrives—and makes completing the case demonstrably easier?
That leads to the larger question behind both acquisitions.
##
##
SIDEBAR
##
Sidebar:
Bruce’s Prime Question—Can Pathology Escape the RUO Death Spiral?
Bruce’s premise is that software capable of preparing diagnostic reports for efficient pathologist review could unlock explosive growth in digital pathology—and that the industry underestimates this opportunity. The decisive product would combine reliable interpretation with superb human interaction: findings, supporting images, and a proposed report presented so clearly that the physician can verify, correct, and finalize the case substantially faster.
The question is whether pathology’s accustomed way of developing and adopting software can deliver that product to ordinary hospitals.
“Bob Will Validate It”
Imagine a department with a Dr. Bob, a capable pathologist who likes computers. He obtains a modest program that counts mitotic figures. He troubleshoots it, checks its performance, and spends several months making it useful in the laboratory. Everyone appreciates the result. “Bob will validate it” becomes a familiar answer to software adoption.
Then a vendor presents a system that interprets slides, considers additional studies, drafts the microscopic description and diagnosis, and organizes the entire case for sign-out.
Everyone turns to Bob.
The difficulty is the expectation his earlier success established. Debugging and validating a 300-line counting program gives little indication of the resources required to evaluate a two-million-line diagnostic system. Those numbers are illustrative; code length itself is no measure of clinical risk. The important change is the breadth of the system’s judgments, its possible failures, and the consequences of relying on it.
A sensible beginning, an unsustainable destination
A historian of innovation might call this path dependence leading to institutional lock-in. Reasonable decisions made for small applications establish habits that persist as the technology changes. The vendor supplies capabilities; the hospital makes them clinically usable. Each incremental step seems manageable, until the accumulated responsibility exceeds what most hospitals can carry.
The frog in the metaphorical pot never expected to face the task of validating a comprehensive diagnostic reporting system. The responsibility grew gradually. The kind of responsibility changed fundamentally.
From a go-kart to a Boeing 777.
A talented workshop can build ("validate") an excellent go-kart.
Developing and supporting a passenger airliner requires an organization capable of coordinating engineering, testing, manufacturing controls, training, maintenance, and continuing accountability.
The workshop’s achievement can be admirable without establishing that its methods will scale to the airliner.
Likewise, an exceptional academic center might build a sophisticated diagnostic system around its own personnel, equipment, and practices. Its success could even obscure the adoption problem. When an ordinary hospital asks to purchase the system, must it also assemble a development team, reconstruct the evidence, maintain custom interfaces, and investigate every consequential update?
A technology can work brilliantly at its birthplace and still lack a workable route to a broad market.
How the RUO Death Spiral Develops
Difficult adoption limits the customer base. A small customer base discourages investment in standardized clinical products. The shortage of those products leaves hospitals dependent on local projects, keeping adoption difficult.
This is the proposed RUO death spiral.
Research-use-only products and laboratory-developed tests are distinct categories; an RUO label does not itself provide a clinical-use pathway. The concern is the broader allocation of development, validation, and support responsibilities to individual institutions. Local innovation has produced valuable advances. The question is how much institutional reinvention the next generation of software can require before its market becomes untenably narrow.
The MRI sales pitch nobody would accept!
Consider another thought experiment. An executive is selling a medical center a top-of-the-line MRI scanner. During the presentation, the executive explains that several essential diagnostic software functions are research-use-only and that the hospital must spend months establishing their clinical performance.
The purchasing committee might wonder what, exactly, it was buying.
This is an invented scenario. Imaging departments have substantial installation, verification, and quality responsibilities. Nevertheless, there is a consequential difference between implementing a supported clinical product and assuming responsibility for turning its essential components into one.
Dr. Wilson’s whiteboard reveals the same problem. In the main essay, she needs to map the products, functions, permitted uses, and responsibilities before she can understand the proposed purchase. Her diligence is appropriate. The amount of disentangling required is itself part of the adoption burden.
The Billion-Dollar Application
The familiar digital workflow puts a scanned slide in front of the pathologist, who examines it, perhaps opens analytical applications, and begins composing a report. The proposed “super app” performs substantial work before the physician arrives: it evaluates the supported case material, identifies findings, assembles relevant images, and prepares a draft interpretation.
Its value depends critically on human interaction. Can the physician move directly from a proposed finding to its image evidence? Correct an error without fighting the interface? Recognize uncertainty and missing information? Reject the draft without having to undo a cascade of dependent entries?
An impressive report that requires laborious checking could increase physician work. A system that reliably reduces work by approximately half on eligible cases could transform purchasing decisions. That remains a performance hypothesis, not an established result for comprehensive pathology reporting. But even across a limited, commercially meaningful range of cases, such savings could support a business case through capacity, turnaround time, and staffing flexibility.
Preparing the case before the physician arrives
The opportunity could extend to preparing the workup itself. In a hypothetical case, software identifies a pattern warranting a special stain. It might recommend the stain, prepare an order for approval, or—within a specifically established scope—initiate the order before physician review. These are three different allocations of responsibility.
At the most ambitious level, the pathologist could encounter the additional stain and preliminary interpretation together, avoiding an entire cycle of opening the case, requesting more work, setting it aside, and returning later. The gain would include elapsed time as well as physician minutes.
Who could establish that this sequence works reliably? Evaluation would need to address missed findings, unsupported statements, inappropriate orders, tissue consumption, erroneous combinations of information, and physician responses to persuasive but incorrect drafts. It would also need to measure the quality and efficiency of the completed case across supported laboratory conditions.
A manufacturer could fund that shared evidence and accept responsibility for a defined, maintained clinical product. Hospitals would retain local implementation, validation, and quality obligations; CAP’s guidance emphasizes those responsibilities. The economic opportunity is to avoid repeatedly recreating the underlying development effort.
Which Pairing Could Break the Cycle?
This brings the argument back to Roche–PathAI and Agilent–Proscia.
Roche–PathAI: building the diagnostic application
Roche–PathAI offers a plausible route through diagnostic development: combine PathAI’s analytical capabilities and AISight workspace with Roche’s tissue-diagnostics experience, regulatory resources, and distribution. A strategy could begin with a defined specimen type and reporting task, establish the complete physician-reviewed workflow, and expand as evidence accumulates. Its strength would be translating analytical expertise into supported diagnostic products.
Agilent–Proscia: organizing the complete clinical workflow
Agilent–Proscia offers a plausible route through the clinical workspace: use Concentriq to assemble cases, connect laboratory systems, coordinate analyses, and present prepared work. Agilent could combine that infrastructure with its diagnostics capabilities and algorithms developed internally or by partners. Its strength would be making multiple capabilities function as one usable clinical system.
These are differences of emphasis. Both platforms encompass workflow and analytical functions. Neither combination should be reduced to “algorithms” on one side and “image storage” on the other.
Which is better positioned to break the cycle? Roche–PathAI presents a strong starting hypothesis for building and establishing diagnostic applications. Agilent–Proscia presents a strong starting hypothesis for organizing their delivery within everyday work. But the proposed breakthrough requires both. The eventual advantage may depend less on the acquired asset than on the purchaser’s willingness to fund—and accept responsibility for—the entire product.
An acquisition alone does not establish that commitment. A compelling plan would specify the initial clinical scope, physician-time target, evidence program, regulatory pathway, integration requirements, support model, and milestones for expansion. Neither announcement establishes an authorized comprehensive draft-report system.
Investing for the Market That Could...Should... Exist
Imagine, finally, a sponsor committing an illustrative $100 million to this undertaking. Skeptics point to slow adoption and limited reimbursement. The sponsor’s competing hypothesis is that demand has been suppressed by incomplete products and the work required to adopt them. A coherent clinical system could create a much larger market than today’s individual applications reveal.
Success would require financing the difficult parts between an impressive demonstration and dependable deployment at the hundredth hospital. Superb human interaction would be central to that investment.
“Built to fail”—if unchanged. That describes the danger if the existing allocation of responsibility persists. Like Scrooge’s encounter with the future, the warning matters because a different course remains possible.
The software may be built to succeed. The system for adopting it may be built to fail. The opportunity for either purchaser is to redesign both together.
