Offcall Physician Spotlight: Meet Dr. Sahar Hashmi

Have a response for Dr. Dr. Sahar Hashmi? Reply to this post directly — she’ll personally read every message!

1. Sahar, what is something another physician taught you that changed how you think about AI in medicine? My physician colleagues and mentors at Mass General Brigham taught me to see a patient's care as a system and to ask one question before ordering anything: "What problem are we solving for this patient?" and to consider the patient's whole ecosystem: their home, medications, co-morbidities, insurance, hospital room, support system and what they can afford. That systems thinking led me to my MIT Master's and PhD. Today, I ask hospitals the same question before they buy or build or partner with an AI tool. Most failed AI projects have skipped this question.

2. What do you want more physicians to understand about AI? AI isn't an IT problem. It shapes clinical decisions, and most tools are still being designed, tested, and prototyped. Physicians must be involved at every stage: designing, selecting, and implementing tools, setting the guardrails, and monitoring them after go-live. That only works if physicians and clinicians understand AI's benefits, its risks, and the hype around it. If physicians aren't in the room when it's chosen, they'll be in the room when it fails.

Read more in my article “Who monitors the AI guardrails?” here.

3. What's something you've changed your mind about AI in healthcare? I used to compare AI tools and agents like ChatGPT, Claude, and Copilot and assume the best AI model would win. I've tested AI tools from nearly every major developer, including OpenAI, Anthropic, Microsoft, Google, Amazon, and xAI. But after years of applying them in practice, building with them, and evaluating them for clients, I've realized the best-prepared hospital wins, not the best product. What matters is whether an AI agent fits the hospital's workflows, integrates with its infrastructure and can be governed by its AI team, and whether the hospital is ready to adopt it. An AI agent that excels at a Boston hospital can underperform in New York because the patients are different. AI readiness, not the race for the best product, is what determines success.

Read more in my HIMSS newsletter interview on AI implementation challenges.

Download to join the waitlist

Medicine is complex enough.
Referrals shouldn't be.

Send and receive referrals, build wealth, and grow your physician community with Offcall.

Download on the

App Store

GET IT ON

Google Play

4. What's something physicians tolerate about AI that they shouldn't? AI vendors grading their own homework with no hospital or physician oversight. We'd never accept a drug trial run only by the manufacturer, yet we routinely deploy AI based on the vendor's own validation, without independent review by physicians or AI experts. Every hospital needs physician-led AI governance, and healthcare as a whole needs an independent AI safety board that tests tools before deployment and monitors them afterward.

Read my article “Healthcare needs an independent ai safety board” here.

5. What's the dumbest thing we do with AI in American healthcare? We chase AI solutions without properly vetting them, fall for snake oil, or buy the AI solution before we've even defined the problem. Health systems sign multimillion-dollar contracts and only afterward ask whether their data, workflows and staff are ready, or whether they needed the tool at all. When it fails, they often lose trust in AI altogether. In medicine, we'd call that treating before diagnosing.

Read more in my article “The $40 Billion Healthcare AI Failure — And The EMR Divide Sabotaging Progress”

6. What's something patients are right to be upset about in the age of AI? Patients now walk in having already asked an AI agent about their symptoms, but the doctor can't see that conversation. The patient arrives informed, while the physician is still reviewing records to figure out what's going on. I call this the AI divide between patients and physicians. Patients are right to be frustrated that their tools and their doctors' tools don't talk to each other, and that their doctor can't see what they've learned. Integrations like ChatGPT with Epic are a first step, but only if they're built to close that gap rather than widen it.

Read my WBUR interview on doctors using Epic with a feature to integrate ChatGPT.

7. What would make you trust an AI tool with your own family? I'd trust an AI agent the way I trust a colleague: It's been tested on patients like mine, and someone checks its work after go-live. That's why I strongly believe in AI guardrails and governance with continuous monitoring so problems are flagged and corrected quickly as we deploy more AI agents in healthcare. Today, I trust AI for specific jobs like supporting screening reads, scheduling, and workflow, but not yet for diagnosis or treatment decisions on its own. Given how fast AI is moving, that day isn't far off.

Read my article on “Why AI Evals and KPIs Are the New Standard”

8. Where is AI moving, and will it replace physicians? AI won't replace physicians, but physicians who use AI well will replace those who don't. The next phase is agentic AI that handles documentation, summarizes past medical records, manages prior authorizations and coordinates care, giving doctors quality time back with patients and easing their cognitive load. The organizations that win will redesign their workflows around AI, not just add AI agents to old ones.

Listed to my interview on The Health Data Ethics Podcast and read more in my article “Beyond AI Pilots” here.

9. Finally, where can medical students, residents, physicians, fellows or faculty learn more about AI agents in healthcare?

I'll be teaching a live online course, Artificial Intelligence Agents for Health Care, Business, and Biotechnology (BIOT E-214), at the Harvard Division of Continuing Education from January to May 2027, Thursdays from 11 a.m. to 1 p.m. ET. No technical background is required. As a semester capstone, every participant builds an AI agent for their own specialty, work or business. The course is open to learners anywhere, and registration open on November 5th, next month. For questions, reach me at dr.saharhashmi.ai@gmail.com.

Download to join the waitlist

Medicine is complex enough.
Referrals shouldn't be.

Send and receive referrals, build wealth, and grow your physician community with Offcall.

Download on the

App Store

GET IT ON

Google Play

About Dr. Sahar Hashmi:

Sahar Hashmi, MD-PhD, is a physician and AI executive and CEO of Magna GenAI Consulting in Boston. She advises hospitals, businesses, healthcare AI companies and startup executives on AI strategy, readiness, implementation and governance. Dr. Hashmi earned her PhD and a master's in Systems Design and Management at MIT. She teaches Artificial Intelligence Agents for Health Care, Business, and Biotechnology (BIOT E-214) at the Harvard Division of Continuing Education. She has given invited grand rounds at Harvard Medical School–affiliated hospitals, including Mass General Brigham, and leads AI strategy and preparedness workshops for institutions such as Cleveland Clinic. A regular HIMSS speaker, she has written on AI in business and healthcare for Forbes, Entrepreneur magazine, CIO magazine and healthcare business today and authored a Becker's Hospital Review op-ed calling for an independent healthcare AI safety board, and discussed AI in clinical care on WBUR. Learn more about her recent accomplishments related to AI on LinkedIn.