Transforming Healthcare with AI Innovations — India AI Impact Summit

A panel at the India AI Impact Summit 2026 on AI-driven advances across diagnostics, medical devices, public health and clinical decision-making, with a focus on real-world deployment in emerging markets. Dr Bakshi was asked how doctors are actually adopting AI in hospitals. What follows is an edited transcript of his answer; his segment begins at 16:44 in the recording above.

I trained as a neurosurgeon a few decades ago. Then I was a neuroscientist doing stem-cell research in the United States, veered into management (McKinsey, in the US and the Middle East), and came back to India, where I was fortunate to run three large hospital chains – Max, then Manipal, then a short stint with IHH's hospitals in India. But my passion has always been technology, and since 2019 I have been that doctor who dabbles in it, building AI companies.

So I have watched this space evolve over 30 years, and one myth that keeps being repeated is that doctors do not like technology. I was told this constantly when I was a CEO. The reality is that doctors are perfectly comfortable with a $20 million piece of radiation equipment. They are perfectly comfortable with a very high-tech MRI, interpreting what a proton spin looks like and what it means for their patient. But when it comes to IT, the story is different: IT has never solved any doctor's problem. Ever. So there is deep scepticism in the doctor's mind – in India, in the US, everywhere I have worked – that you build things, ask us to use them, and give us no benefit in return.

Now a very interesting trend is visible. Doctors, especially in the US (highly regulated, and worried about the medico-legal risk of using AI), are finding that their patients walk into the room very well informed – sometimes better informed than they are.

One thing is clear: doctors using AI for diagnosis or treatment is going to be a long journey, because the burden of proving safety and efficacy is very high, and there is no way around it. The US FDA has a good framework for proving that an AI algorithm is safe and effective, but like a pharma drug it costs a lot of money and carries a lot of risk. What OpenAI and Anthropic have done instead is make their general models very smart on healthcare information – so they do not carry the risk that a doctor would, or that an AI company like mine would, if we claimed to diagnose tuberculosis or cancer better. There are companies doing exactly that, and there is a large body of innovation in faster diagnostics and better therapeutics, but to my mind that is at least three to five years away from full adoption.

What are doctors doing today? Using it informally. I am a neurosurgeon; if I have not seen an eye injury in a long time, I can ask ChatGPT and get a pretty good answer, then call my ophthalmology colleague and do what the patient needs far better than before. The productivity of the doctor does go up.

But if we only think of AI as a chatbot, we are making a big mistake, because of what has happened in just the last two months. Those of you in the AI space will have watched Andrej Karpathy describe going from writing 80% of his own code in the first week of December to 10% by the last week of January – a dramatic shift, driven by the rise of agents.

A real illustration. A non-coding person like me could recently create a personalised app for a friend's autistic daughter. My friend gave me her reports, I fed them to Claude Code, and we built an app around her strengths (hyperlexia, an unusual musicality) and her weakness with mathematics. If I had tried to build an app like that six months ago, it would have cost a few tens of thousands of dollars and it would not have been good at all – a static machine. This one took a couple of hundred dollars and five to seven days.

I use that story to make a larger point: it will soon be possible for seven billion people to each have agents taking care of their health. Imagine a 70-year-old with diabetes, early kidney disease, and knees that make walking difficult. That gentleman can now be well guided by a personalised swarm of agents watching over his own records. To me that is the revolution of the next 12 to 18 months – personalised medicine through personal agents, not the genomics-wala personalised medicine, which never delivered what it promised. Watch that space: your own friendly AI agent, telling you how to live longer.

The moderator closed by asking how many agents are working for me right now. Only agents – I do not have anyone working for me anymore.