Last fall, Dr. Melania Poonacha took on a side hustle: tutoring AI.

In her regular job, Poonacha, a 46-year-old internist and pediatrician who lives in San Francisco, works 12- to 16-hour night shifts at California Pacific Medical Center and the Chinese Hospital.

But Poonacha has found that AI tutoring provides breathing room from the hospital grind. She signed up with Mercor, a San Francisco company that contracts with professionals across various industries to train AI in their trades. For physicians, Mercor’s platform requires her to write prompts about medical situations in her specialty, then grade the AI’s responses according to accuracy and relevancy.

Even though she makes a healthy salary in hospital work — around $375,000 a year, she said — the extra income has helped her financially and in developing a better work-life balance. Her doctorate at the University of Kentucky and masters in public health at the University of Pittsburgh left her with $400,000 in student loans, and she had wiped out her savings paying them off. “I do enjoy being a doctor, but it’s not sustainable for me to work 50 to 60 hours on nights a week for the rest of my career,” Poonacha said. “These types of nonclinical side gigs are ideal ways to make an extra income without extra stress.”

The idea of doctors who earn nearly $400,000 needing a side hustle might seem like a gag. (Indeed, comic Austin Nasso mocked the idea in a San Jose set last year.) But it’s no joke. Many Bay Area physicians have taken on AI tutoring as a lucrative part-time job, working for Mercor or Handshake, another AI-training company. And Poonacha is happy to spread the word. “I talk to physicians all the time that are like, ‘I hear you’re doing this AI work. Tell me more about it,’” she said. “A lot of us are looking for extra work on the side. … Our salaries are stagnant.” 

Even in San Francisco, where the median salary is around $140,000, a $375,000 income is hardly meager. But the high cost of living hits physicians like it does everyone else. Dozens of Bay Area ZIP Codes rank among the most expensive in the nation. And after adjusting for cost of living, San Francisco is the fifth-lowest metro area for physician compensation nationwide, according to a 2025 report by Doximity (opens in new tab). 

Which explains why many doctors don’t mind moonlighting in AI, even if they might be helping to create the tools that may one day replace them.

• The SF Fed chief has two words she won’t say about the AI boom

• How this surging data cloud company is mandating AI at work

“It’s hard to think of a doctor making $250,000 living paycheck to paycheck, but that’s the reality,” said Dr. Robert Pearl, the Bay Area-based former CEO of The Permanente Medical Group and author of the book “ChatGPT, MD” and the newsletter Monthly Musings (opens in new tab). “The Bay Area has a very high overall cost of living,” he added. “They can’t afford the middle-class lifestyle they desire because of housing prices and student loans.” Around 40% (opens in new tab)of doctors have second jobs, Pearl said.

AI companies are courting doctors because healthcare is hungry for automation. The U.S. spends about $5 trillion on healthcare annually, and much of that goes to highly manual tasks and procedures. McKinsey estimates that AI (opens in new tab) and automation could save $200 billion to $360 billion a year in healthcare spending, from billing and prior authorizations to clinical guidance. But considering medicine’s high legal and clinical stakes, AI labs need board-certified doctors to take the lead in training their models, to ensure they are safe for clinical use with real patients. Radiology, cardiology, and neurology are frequently cited as specialties ripe for AI disruption.

Among doctors, whether they may be training their robot replacements remains a hot topic. “Doctors are worried, but they don’t see it as being imminent,” said Pearl, who estimates it might be 20 years before this becomes a real issue — by which point, he figures, AI replacement is “an inevitability.” Still, Pearl noted that physicians are in short supply in the U.S., with an 86,000-doctor deficit projected by 2036. (opens in new tab) (opens in new tab)“We can do a lot more generative AI to improve health without compromising the job market.”

Wang, a 38-year-old internist from Menlo Park who declined to share his last name, earns around $310,000 a year. He started training AI models in January. 

Have thoughts on this story?

He took the gig to offset unexpected expenses but found the work easy and liked being able to log hours at night. “Working a second job was never part of the plan,” Wang said. Still, the money helps: With 60-hour weeks, his day job pays around $60 an hour after taxes, and he has large student loan payments. “My friends in the Midwest think I’m rich, but they own their own houses.”

But for many doctors, AI work provides more than just extra pocket money. “They don’t go [into AI] for the big payout but to be able to do something a day and a half a week that is different,” said Pearl. That can ward off burnout, he added. “Some people are doing it because they just like being in a technological world.”

That was the case with Dr. Krystal Lin, a Bay Area occupational medicine physician who signed up with Mercor last May after seeing a targeted ad on Instagram. “I was more curious than anything else,” she said. “The pay wasn’t that great.” Mercor trainers in her specialty earn $150 to $200 an hour.

Appropriately, Mercor had her do a screening interview with AI — an experience she described as “surreal.” But she was impressed by the AI agent’s nuanced grasp of her specialty — it knew more than, say, a family medicine doctor might have. 

The work was fairly technical, Lin said, translating clinical assessments into a format that the AI could parse. “What’s tricky is not the response but writing it in a way that would fit into [their] rubric,” she said. “If this was an actual patient, you would explain it to them conversationally, but to an AI, you have to break it down.” That meant bullet points of symptoms and potential treatments, ranked by importance, with primary and secondary criteria separated out.

“Most physicians doing this could earn more picking up extra shifts,” she surmised. “But they want to be part of AI training to have this front-row seat of where AI is headed.” 

Lin found the experience promising enough that she blogged about it on Mozibox (opens in new tab), an “AI career copilot for doctors” side project of her own.

She has been flooded with inquiries from doctors, many of whom wanted to verify she was human before they sent in their applications. She has since signed up with Handshake, a Mercor competitor, to document their process (opens in new tab) for her community. “So far, no red flags,” she said.

Mercor and Handshake declined to provide figures on how many local doctors are working for them, but the demand for medical expertise appears to be growing. Handshake has begun offering a year-round AI fellowship for physicians, a (opens in new tab)nd other companies have started recruiting doctors to train their medical models, including Outlier (a subset of Scale AI) and Redwood City-based Snorkel AI.

And many doctors are less worried about losing their jobs than falling behind the curve of where medicine is headed. By training the AI, doctors are “able to push the boundaries of what’s possible in medicine,” Lin said. “We’re  working hand in hand with the AI [to] find cures to many of the incurable diseases.”