Head of Workplace

Meet the AI Assistant for healthcare. The next era of clinical software won't be clicked through. It'll be talked to. Abridge led the shift to ambient AI in clinical documentation, centering the patient conversation and cutting the after-hours charting clinicians used to take home. Yet, the rest of the day is still spent on tabs, clicks, and navigating fields. Clinicians should be able to do what everyone else can now do with consumer chatbots and agents: say what they need and get it. That's why Abridge built Abridge Assistant, the AI assistant for healthcare. Assistant is built from the same AI research and clinician-led evaluation that brought Abridge's ambient notes to more than 300 enterprise health systems. Clinicians can ask questions and delegate tasks to Assistant in their own words. As of its September release, every clinician at every Abridge health system has access to Assistant, whether or not they use Abridge for documentation. Ask Assistant about a patient, and it pulls together the relevant details from across that patient's chart, with each source cited. Clinical questions come back with current guidelines, and calculators open with chart values already filled in. Assistant also personalizes the note and drafts artifacts such as referrals and handoffs for the clinician to review. All of it runs inside the security and governance review framework of the health system. Below, Matt Troup, a PA-C and Abridge Clinical Strategy Principal, explains what clinicians need from an AI assistant. Janie Lee, Abridge VP of Product, explains what it took to build it. What clinicians need from their AI assistant: Matt Troup. In hospital medicine, questions often come at inopportune times: walking out of a room, halfway through a note, overnight when a lab comes back. The tasks pile up right behind them, and each one means more time clicking through the chart before the shift ends. Most of the standalone tools I've used share the same problem, regardless of how much they tout AI. They know a lot of medicine, but they don't know my patient. A response is only as specific as the data preceding it, so I end up doing the prompting myself: retyping comorbidities, current meds, and what Abridge tried last admission, then filtering out whatever doesn't apply. These tools rarely show how they got to a response, leaving me uncertain how they generate information or where it's pulled from. The chart has the opposite problem. It holds every detail collected about the patient, sometimes years of notes, results, and outside records, but I have to piece the story together myself, and it's easy to miss something. Add the documentation that follows, and a lot of my time is wasted hunting for information and clicking through screens to act on it, and there's no guarantee that I will find everything I need in the time I have. Health systems have their own requirements. Some of the tools clinicians reach for have not been reviewed and integrated by their organizations, which creates its own set of problems. And institutional knowledge, like clinical pathways, often lives in a policy library that too few people consult. When Abridge started building Assistant, this was my wish list: * It knows the patient: Responses should reflect the chart and the conversation, not just general knowledge. * It's where the work happens: No new tab, no copy and paste, no starting over. * I stay in charge: It does the digging and the first draft but signing off is up to me. * It shows the work: It cites which parts of the chart it pulled and which sources it checked, so I can follow the reasoning before I act on it. * My health system stands behind it: It has to clear security and governance review and be fully integrated into the technology I'm already using. How Abridge built Abridge Assistant: janie Lee. Abridge built Assistant for the clinicians at Abridge like Matt and clinicians who work at every one of its partner health systems - for those who are stuck between two eras of software. They deserve better than a broken workflow of legacy software built around a thousand clicks, with newer AI tools that fall short layered on top. Assistant is built to close both gaps, working from the patient's chart and arriving via a simple chat or voice interface, backed by robust and evaluated intelligence. Input from clinicians and health systems shaped its key development decisions: * Start with the context. Assistant draws on the chart, the conversation, and the evidence clinicians already trust, so each response is shaped to the patient in front of them. Context includes a health system's own clinical pathways, so responses further reflect how care is delivered in that organization. * Keep it simple, wherever clinicians work. The Assistant interface is consistent inside the EHR and on the clinician's phone, and it stays simple as capabilities grow. The clinician orchestrates, and Assistant carries the context from one step to the next. Ask what changed since a patient's last admission, for example, and the response can become a draft handoff in the same thread. * Show the work. A request like "compare treatments for this patient" sets off a chain of work behind the scenes, from pulling the right context to checking that the response is complete. Assistant shows those steps as they happen, so the clinician can see the reasoning and which sources are referenced as a response is compiled. * Evaluate continuously. Abridge evaluate Assistant with human and LLM judges, using clinician-built rubrics for clinical quality and safety, before and after a capability ships. Testing also includes a fresh review of existing PHI protections, so patient data stays under the same controls across Abridge. Assistant's simplicity hides most of the effort behind it. A lot of this work was in progress before Assistant existed: Abridge was already connected to multiple EHRs and embedded in the clinical workflow, with security, compliance, and AI governance built in from day one. Few products start with that foundation, and Assistant was built on it. Because clinicians can open Assistant inside Abridge, under approvals their health systems have already granted, it's been adopted at scale already. How clinicians use Assistant in practice: Matt Troup. Take a typical morning in the hospital. Before rounds, Assistant shows me what has changed for each patient on my list, so I'm not starting from scratch. For a patient I'm meeting for the first time, using only my voice, I ask what happened at their last admission. The first response from Assistant surfaces key details, which means my follow-up question is more precise. I might ask how the patient's kidney function affects an antibiotic choice. The response brings the evidence together with this patient's history, with guidelines cited and the health system's own pathway included. Assistant can then open a calculator with chart values already filled in for me to check. When care moves to another team, I can ask Assistant to turn what Abridge has just worked through into a draft handoff, then review and edit it. That's what Abridge mean by orchestrating the workflow instead of clicking through it. I decide what I need from the past, combine it with what's happening now, and choose the next action. Assistant does the searching and the first draft. I make the clinical judgments. Unlike the standalone apps I described before, I can follow the clinical question through to the work it creates, rather than rebuilding the context each time I move from sorting out what matters from the chart, to considering a decision, and, finally, to communicating a plan. The setting may change - a hospital ward, a clinic, an emergency department - but the need is familiar. Abridge need to understand the patient, decide what comes next, and make sure the plan reaches the people who depend on it. Better-informed decisions flow into the documentation and into how Abridge talk with the patient. This is why the ambition has to be bigger than fewer clicks. Abridge is building toward a more connected way of working, where the effort that goes into understanding a patient carries forward into the decisions, documentation, and conversations that follow. The best clinicians I've worked with weren't simply the ones who knew the most facts. They were the ones who knew how to apply what they knew to the person in front of them. Abridge Assistant is designed to support that work: less effort piecing the story together, more attention to the person living it. Where Abridge is taking assistant and what this means for the future of care: janie Lee. Near term, Abridge is focused on taking on more of the clinical and clerical work clinicians have asked Abridge to offload. Abridge partner closely with clinicians at its 300+ health systems and are always working with them on improving Assistant's capabilities. Longer term, Abridge want Assistant to stay with clinicians across devices, and to work across systems. A clinician's work can span multiple EHRs and the systems that payers and life sciences organizations manage. Assistant will help clinicians across those boundaries. Its goal is for clinicians to spend less time on tasks and more on what they're trying to accomplish for the patient. Getting a patient the right outcome, whether that's a medication or a spot in a clinical trial, takes front office and back office work from clinicians, nurses, payers, and life sciences. Assistant is how Abridge plan to move that coordination into the background, while the experience stays simple for clinicians.

Now in Practice: September 2026. Welcome back to Now in Practice. One throughline in this month's release is the timing and detail behind how complex care makes it into the record. Diagnoses organized and refined while the clinician is still in the note. Related conditions grouped together so combination codes can surface before the chart closes. Cues that capture what was discussed during an inpatient stay while the story is still fresh. So much of what often comes back weeks later as a retrospective query was known in the moment, and each of these updates is a version of the same goal: settling that detail at the point of care so the record reflects the full complexity of the care delivered and holds up on review. Alongside them, two updates give administrators more direct control over their Abridge implementation, and two specialty model enhancements now capture the details that define allergy and palliative care. Closing CDI gaps before the chart reaches coding. Three capabilities in this release are focused on closing the gap between the care delivered and the care that is documented: the diagnosis that's accurate but not specific enough, the combination code split across a flat list, and the observation made at the bedside that never reaches the note. This builds on Abridge's Query Engine, introduced in July, which surfaces AAPC-compliant CDI queries inline in the note when a documentation gap is present. This month extends the same idea to the diagnosis list itself and to the smaller details of an inpatient stay, where a formal query would be heavier than the detail warrants. Visit Dx integration enhancements. Its Visit Dx integration now brings problem headers, diagnosis search, and specificity options together in a single place, within the documentation workflow. Diagnoses can be organized under the right headers, found or replaced inline, and refined to a more specific option when the modifier detail is available, all without stepping out of the note. The goal is an accurate diagnosis list that's settled during documentation, rather than days or weeks later when a review or query flags what's missing. Problem grouping. Abridge now groups related diagnoses on the problem list into logical clusters, organized the way clinicians reason about a patient rather than as a flat list. With related conditions side by side, combination codes are easier to spot for review while the clinician is still in the note. A combination code split across a flat list is exactly the kind of gap retrospective review exists to reconstruct; catching it at the source means the case mix reflects the complexity of the patients actually treated. Query Engine: inpatient cues. While clinicians chart an inpatient stay, Abridge now surfaces non-blocking cues that point to care already discussed or assessed: whether wounds were present when skin integrity was checked on admission, or the key decisions from a goals-of-care conversation. Each cue carries its supporting evidence and can be acted on or dismissed in one step. Because the prompt arrives while the context is still fresh, the full complexity of the care makes it into the chart, and clinicians can have more confidence that nothing was left out. Cued, not interrupted. The instinct with any documentation prompt is to make it impossible to ignore. But a prompt that blocks signature teaches clinicians to click through it, and once that habit forms, the prompts that matter get the same treatment as the ones that don't. Cues are dismissible because the clinician is the one who knows whether a detail belongs in the record, which is also what makes every acted-on cue meaningful. More control for the teams running Abridge. As implementations scale across care settings, it becomes more important than ever for the teams administering Abridge to be able to manage access on their own timeline and verify the work the platform does on their behalf. Two updates this month give project and CDI teams more direct control over their Abridge implementation. Self-Serve user management. Partners can now add, edit, and remove user access directly from the Enterprise Portal, with users across surfaces visible in one place. For Abridge Inside, an optional approval queue routes clinician access requests to admins to approve or deny, replacing manual updates to the Epic Ambient Allowlist for organizations looking to centralize Abridge access. Every hire, role change, and departure can be handled as it happens. It's a direct response to partner feedback: the time between a staffing change and a clinician's working account shouldn't depend on someone else's queue. QA Viewer for Query Engine logs. CDI teams can now open any Query Engine query in the QA Viewer and see its full record alongside the clinical note: the documentation gap it surfaced, the evidence behind it, the compliance rationale, and how the clinician responded. Instead of re-checking every query by hand, reviewers can spot-check the engine's judgment case by case and build confidence in the routine layer of query work, so their time goes to the complex cases that need real judgment. Deeper detail for specialty care. Every specialty has terminology and details that a general note structure may not be tuned to surface. Two recent note enhancements capture what matters most in allergy/immunology and palliative care. Allergy/Immunology model enhancement. Allergy/Immunology notes now record allergens, triggers, and reaction timelines exactly as discussed, so a peanut allergy stays a peanut allergy. The HPI captures medications by name, dose, and device along with prior skin prick and IgE results, and the plan carries immunotherapy schedules and dosing. In allergy care, the specific allergen and the timing of a reaction are the clinical picture, and the note should carry them that way. Palliative model enhancement. The palliative note now generates a dedicated Goals of Care subsection in the HPI, with a corresponding summary of action steps in the Assessment & Plan. The note also improves capture of medication dosages and changes, including MEDD and pain pump settings, along with palliative-specific scores like the Palliative Performance Scale (PPS). Goals-of-care conversations are often the most consequential part of a palliative visit, and they deserve a place in the note that matches their weight. This month's release is about the record keeping pace with the care. A diagnosis is refined before the note is signed, a combination code surfaces while the clinician can still act on it, and what was observed at the bedside reaches the chart while the story is fresh. Behind the note, admins manage access on their own timeline and CDI teams can see exactly why each query was raised. That's what Abridge shipped in September, and Abridge'd love to have you see it in practice.