This analysis synthesizes 8 sources published the week ending Oct 20, 2026. Editorial analysis by the PhysEmp Editorial Team.
On Oct. 13, Suki announced a $90 million Series E and said where the money goes: AI agents that handle the work after the visit ends, such as referrals, prior authorizations for new medications, order entry and follow-up scheduling. The same day, Northwestern Medicine put OpenEvidence inside its Epic record and Johns Hopkins Medicine signed on with the same company to rethink how physicians are trained. Read together, the week says less about which vendor is winning than about which parts of a physician’s day investors expect software to absorb next.
That is the employment question inside PhysEmp’s AI in Physician Employment & Clinical Practice coverage. Funding rounds are forecasts. They tell you which tasks companies believe health systems will pay to automate, and by extension which tasks will be left on the physician’s plate and which new ones will be added to it.
The money has moved past the note
Ambient scribing was the first wave. Healthcare Dive notes that Ambience Healthcare raised $243 million and Abridge $300 million last year. The rounds of the past month are aimed somewhere else. Heidi, which raised $340 million at a $900 million valuation on Sept. 22, began as a scribe and now says it is moving from documentation into “supervised action”: administrative automation plus support for clinical decisions and reasoning. Its CEO, Thomas Kelly, described the shift as going “from documenting care to helping clinicians act on it.” Heidi’s annual recurring revenue went from $1 million to $50 million in two years, and Beth Israel Lahey Health has extended its scribe to all of its physicians.
Suki’s pitch follows the same arc. According to MarketScale’s breakdown of the round, the company works with more than 250 healthcare organizations, saw patient encounters on its tools grow more than 190% year over year in 2026, and says its AI should extend into clinical reasoning, operational tasks and revenue-cycle work. The same report puts EliseAI’s September round at $350 million and a $4 billion valuation, for a product that automates patient calls, referrals, scheduling, insurance verification, chart prep and follow-up for specialty physician groups.
Add OpenEvidence’s $250 million round, and four rounds reported between Sept. 22 and Oct. 13 total more than $1 billion. Nearly all of it is aimed at the hours between visits: the inbox, the referral, the prior auth, the chart prep. None of those tasks is a billable visit, which is why the question of who absorbs them, or who keeps the time once software does, matters so much to employed physicians.
When a vendor automates a referral, the time saved does not automatically return to the physician. Unless a contract or compensation plan says otherwise, the employer decides whether those hours become shorter days, more visit slots, or a larger panel.
Evidence lookup is becoming plumbing
The second pattern is quieter. OpenEvidence, now valued at $15 billion after a round led by Andreessen Horowitz and Byers Capital, says more than half of U.S. physicians use it regularly. It is already embedded in Epic at Mount Sinai and Sutter Health. At Northwestern Medicine, nurses, pharmacists and other clinical team members have signed in through their existing EHR credentials since Oct. 8, using it to check dosing, screen drug interactions and review guidelines. In September, the company made three of its production models free for verified clinicians.
Suki is moving the same way from the other direction. Its Ask feature, launched Oct. 8, lets clinicians pull a patient’s latest A1C, medications and allergies alongside UpToDate answers without leaving the documentation app. In an evaluation reported by MarketScale, clinician reviewers preferred answers that included patient context 83% of the time, though that figure measures preference, not outcomes or error rates.
When reference lookup is free, embedded and open to the whole care team, knowing the answer stops being what distinguishes the physician. Judgment about whether the answer fits this patient, and the authority to act on it, becomes the scarce part. Employers will price that differently, and the first place it shows up is likely to be in how team-based models divide work between physicians and everyone else with the same login.
Who answers for the agent’s referral
Every one of these companies describes its product as supervised. That word carries the employment consequence. In a KevinMD essay published Oct. 16, gastroenterologist Brian Hudes, who discloses a commercial interest in AI governance, argues that responsibility should follow control: clinicians answer for clinical decisions, institutions for deployment and oversight, developers for what they build and change. He asks hospitals to trace every AI-assisted action to a named human who has a real path to override it. He also notes that the Joint Commission’s voluntary AI certification, introduced in June, certifies an organization’s governance rather than any individual product.
Apply that to a referral drafted by an agent and filed under a physician’s name. If the physician has no practical way to see why the agent chose a specialist, or no time allotted to check, the accountability is nominal while the liability is real. This is where the funding wave meets the job description: more tasks completed in the physician’s name, fewer of them performed by the physician.
A useful question for candidates in an interview: which AI tools act under my signature here, who chose them, and how much time does the schedule assume I spend reviewing what they produce?
Training for the half of the job that stays
The academic side is already redrawing the profession around this split. Johns Hopkins and OpenEvidence plan a training model built on pathophysiology, first-principles reasoning, diagnostic skill and interpersonal skill, on the premise that the clinician’s role shifts toward what cannot be automated. They disclosed no timeline or financial terms. A day later, a JAMA Viewpoint by Jesse Burk-Rafel and Robert Wachter warned of “never-skilling,” where learners who start with AI skip the struggle that builds judgment, and called for an organizing principle for revising training.
We have written less about the companies building these tools than about the liability and governance questions around them, and the size of this month’s rounds suggests the products themselves deserve more attention. For recruiters, the practical point is already visible: the AI stack a system has chosen, and how it splits the saved time, is becoming part of the offer whether or not anyone puts it in the job posting.
The number to watch is Suki’s 190% encounter growth set against physician headcount at the systems deploying these agents. If encounters keep climbing while staffing plans stay flat, the productivity gain is being booked as volume. If systems start citing agent adoption to justify smaller recruiting targets or larger panels in 2027 budgets, the funding rounds of this fall will have been the first draft of next year’s job descriptions.
Sources
Suki raises $90M in Series E round to create AI agents – Healthcare Dive
Suki raises $90M to take its AI scribe beyond notes – MarketScale
Heidi lands $340M to build AI ‘care partner’ for clinicians – Fierce Healthcare
OpenEvidence raises $250M at $15B valuation – Becker’s Hospital Review
Northwestern Medicine integrates OpenEvidence into Epic – Healthcare IT News
Johns Hopkins, OpenEvidence to develop new medical education model with AI in mind – Becker’s Hospital Review
Physician Training for the AI Era: Rethinking a Century-Old Wager – JAMA
Clinical AI accountability does not come with your EHR – KevinMD.com