For decades, digital transformation in healthcare has been something typically done to clinicians, not driven by them.

Today, the clinician-led revolution isn’t just a story about technology - it’s a response to mounting systemic pressure. Across specialties, early adopters are no longer waiting for enterprise rollouts: they’re adopting AI tools themselves, often outside the walls of IT approval. What started as freemium-enabled experiments to lessen administrative burden through AI note-taking or documentation support is fast becoming standard of care.[1]

The rise of bottom-up adoption highlights a deeper reality - the supply-demand equation in healthcare is breaking. Funding pressures are rising just as shifting populations drive unprecedented demand for services. Clinicians rarely get more resources - only more work, more documentation, more pressure. Many are burning out, not because they’ve lost purpose, but because the system gives them no leverage to do what they came to do: help people.

Physician AI usage nearly doubled between 2023 and 2024, a pace the AMA described as unusually fast for healthcare technology adoption.[2] But much of this is happening ahead of governance: a Wolters Kluwer survey of over 500 healthcare workers found that 40% have encountered unauthorized AI tools in their organizations, with clinicians citing faster workflows and better functionality as primary drivers.[3] This is the demand signal that a selection of healthtech companies have built their go-to-market around: channelling clinician-led adoption into purpose-built tools, lowering the barrier from shadow AI experimentation with an eye toward sanctioned, integrated workflow support.

Clinicians and care teams are turning to tools that help them work faster, with less friction, and without compromising care. This is clinician-led growth - and it isn't a rebellion against the system. It's a clinician's way to keep the system functioning.

Write, Read, Reason

The path from clerical to clinical

There are three distinct phases that align with the broader progression from clerical support to clinical support, corresponding to the shift from minimal system integration to full integration. Each changes how clinicians work, how vendors grow, and how Systems of Record (SoR) defend their centrality. Understanding these transitions is key to seeing where value and control will move next.

The three phases are write-only automation, reading the record, and reasoning. Today (2025 at the time of writing), we are somewhere between leaving the first phase and entering the second; across both AI native and legacy companies - think clinical knowledge tools like Open Evidence or Up To Date, and documentation tools like Abridge or Heidi.

Write-only automation: Easing the documentation load

Clinicians test and adopt tools that take work away without significant change - neither in their workflow, nor in the system’s. Think ambient scribes, with different levels of automation from full transcript to structured documentation - with clinician oversight over the final acceptance into the SoR, levelling up from manual copy-paste through to automated push integration.

SoRs are happy to oblige writing this data into the record, as it reinforces the data gravity and their centralized role - on their terms.

Vendors can leverage the product-led growth (PLG) motion into expanding enterprise deals - driving usage and revenue.

While clinician autonomy is expanding through lightweight, edge-deployed tools, procurement and budgetary control remain centralised. Leveraging a product-led removal of the purchasing decision creates momentum to move towards a large-scale deployment - with all of the formal enterprise purchasing cycles, integration reviews, and contractual compliance. Securing user buy-in before undertaking large-scale deployment is a significant driver of success, as many such initiatives struggle with adoption and engagement, limiting the realization of expected benefits.

The move from writing into the record to reading from it introduces a fundamentally different challenge. Write-only tools operate at the edge - they capture a conversation, generate a note, and push it into the SoR without needing deep system access. Reading requires the opposite: pulling structured and unstructured data from across fragmented systems, each with its own formats, access controls, and integration constraints. Despite growing mandates around interoperability and open APIs, real-world data exchange remains uneven. This is the wall that separates phase one from phase two.

Reading the record: Organizing the chaos

As documentation workflows become increasingly handled, the natural next opportunity emerges: managing the parts of a workflow powered by finding and interpreting the right information. But this shift also demands more - more product capability, more trust from SoRs willing to open their data, and more tolerance for risk as AI moves closer to clinical interpretation.

Clinicians need to access the relevant patient information, for the situation at hand, when it is needed - but they work in terms of patient problems, not database schemas. Translated: they need to tame the chaos of multiple systems and workflows to access the structured and unstructured data they need. Think on demand chart or note summarization: levelling up from generic summaries, to specialist-specific summaries, to patient-context-specific summaries.

Vendors look to expand their role and user base, from a single point solution to extended functionality that now requires understanding the patient record from within the SoR. Integrations with specific SoRs based on specialty or marketshare, as well as partnership opportunities to deploy these tools within their ecosystem will be key.

SoRs will now need to consider what is their strategic role - do they want to own the System of Record role, or the Engagement and Reasoning role? What could or should they own on top of their SoR vs allowing others to build on them? If a clinician was to never use their interface, can they remain in control of their future? Also: interoperability across systems, data governance, and monetization.

This is also where the tension between AI-native platforms and legacy incumbents becomes tangible. Incumbent SoRs - especially EMRs - are adapting AI capabilities, while AI-native platforms are built from the ground up around reasoning and automation. But legacy systems expand outwards, taking back control with a single update. Incumbents do not need to be first or best; they simply need to be "good enough" and integrated.

If the transition from Write to Read is an interoperability challenge, the transition from Read to Reason is a category change. Reasoning tools don't just retrieve or organize information - they interpret it, weigh alternatives, and surface recommendations that inform clinical decisions. This moves AI from administrative support into clinical territory, where the stakes, scrutiny, and liability frameworks are fundamentally different. When a clinician can independently review the basis for a recommendation, the tool remains a support layer. As outputs become more sophisticated - and harder to fully evaluate in real time - that line blurs. And with it, the question of who owns the risk becomes unavoidable. Is it the clinician, the vendor, the health system, the model, or a combination? This is the threshold that separates workflow tools from clinical intelligence and is why the reasoning phase will unfold more slowly, and with more friction, than the first two.

Reasoning: From information to intelligence

With organized information, the context of the situation, and goals or outcomes, reasoning becomes the process that converts structured inputs into clinical frameworks to support decision-making.

Clinicians will increasingly expect tools that do more than retrieve data - they want systems that help them make sense of it. The next generation of reasoning tools won’t just predict a diagnosis or flag a risk - they will collaborate in the thought process: clarifying why something matters, what alternatives exist, how different actions might play out - simplifying the administrative burden to make it happen and enabling each individual to focus their effort at the top of the value chain.

Vendors see reasoning as the next step up the value chain - moving from workflow utilities to intelligence layers that streamline, inform, and enhance clinical and operational decisions, empowering clinical teams to act and drive outcomes. Deeper integrations across multiple systems create a fuller picture of the patient context, as well as unlocking capabilities that require realtime data sync - like understanding and interpreting the patient chart and patient cohort matching.

SoRs now face their defining strategic question: do they remain the system of record, or evolve toward the system of reasoning? Owning reasoning means more than storing or exposing data - it means enabling context-aware intelligence within the SoR’s ecosystem. But doing so changes their identity: from neutral repository to active participant in care decisions - and the reality of risk ownership.

If reasoning happens elsewhere, SoRs risk becoming the database everyone relies on but no one actually experiences. If reasoning happens within them, they must operate like platforms - open, composable, and interoperable - yet still be trusted as the robust single source of truth. Avoiding the “black-box” problem common to probabilistic systems becomes essential: data must be traceable to its original source, decisions auditable, and insights explainable in context. Clinically relevant intelligence must be not only accurate, but understandable and defensible within clinical and regulatory frameworks. This represents a significant evolution: from a passive repository of facts to an active steward of reasoning - and a genuine partner to the clinician.

A quiet replatforming of healthcare

Ironically, the same Systems of Record that once centralized data and workflow discipline partially contribute to the appetite for change. Their rigidity and deterministic nature - designed for compliance - push clinicians to seek tools that work for them.

AI tools aren’t just adding features, they’re quietly replatforming healthcare - transforming how data moves, is understood, and acted upon. As automation evolves from simply writing into the record to reading from it - and ultimately reasoning with it - the center of value and control shifts.

This bottom-up adoption wave isn’t a rebellion against the system; it’s the system adapting from its edges. Clinicians are defining the new healthtech roadmap - choosing the tools that fit their workflows, building the bridge from documentation to decision, and accelerating the pathway for AI integration across healthcare.

In Short

Healthcare is riding on the most significant adoption wave of the past few decades - and clinician-led growth has become the surprising entry point. In a world of regulatory approvals, compliance hurdles, and top-down purchasing, there is a shift of power - and it's changing the way innovation enters the system, spreads through clinical networks, and ultimately reshapes how care is delivered.

This shift will follow a clear trajectory: from AI point solutions that write into existing systems, to tools that read and organize data across them, and ultimately to fully integrated systems of engagement and reasoning that sit at the heart of clinical work.

The next generation of health-defining companies will emerge where clinician-driven adoption meets platform ambition. Whether the incumbents evolve - or new entrants replace them - will define the next decade of healthcare technology.

What else?

Enterprise approvals remain essential: Without question, moving from supporting just one part of a workflow at the edge to a clinician-centred system of record, engagement, and reasoning across the full patient context will require deep integration with existing systems - and, by extension, governance, compliance, and approval from central health authorities.

Who owns the risk? The transition to reasoning raises the question - but doesn't answer it, because nobody can yet. Current liability frameworks weren't designed for probabilistic systems. If a reasoning tool surfaces a recommendation that a clinician follows and the outcome is poor, existing tort doctrines struggle to assign fault across clinician, vendor, health system, and model. How this gets resolved (regulation, case law, contractual frameworks, or some combination) will shape how fast and how far the reasoning phase can go.

Evidence-based (AI) care: With the move from silo to platform, and individual to organization, explainability and traceability will be key to prove that system performance meets defined clinical, regulatory, and operational standards. How well can tools and workflows be validated? How do future platform or model changes benchmark to current performance? How will clinicians and organizations ultimately control or trust their outputs?

The patient in the loop: This article focuses on clinicians, vendors, and systems - but let's remember: the patient is the reason all of it matters. As AI tools move from documentation to reasoning, new questions emerge about consent, transparency, and equity. Do patients know when AI is shaping their care? Are these tools validated across diverse populations, or do they risk encoding existing biases at scale? Healthcare's purpose is to help people achieve better health - and that can only succeed if systems maintain trust, ensure privacy, and actively reduce inequity in access and delivery.