Healthcare has always been a tricky ground for technology. Unlike retail or advertising, where speed matters more than certainty, in medicine, accuracy and safety are non-negotiable. Being cautious about such overreach in the MedTech ecosystem is indispensable for companies building automation at scale, particularly with the rise of agentic AI.
The industry must view AI adoption through three burdens: regulations, determinism, and privacy, according to Abhishek Shankar, CEO Emids, a firm providing AI-driven engineering and digital transformation solutions for the healthcare and life sciences industries.
In an exclusive interaction with AIM, Shankar said, “We have been doing this for, say, various levels of automation leading to agents for over 36 years. And, we take extra caution to handle some of these so-called burdens which are needed to create good healthcare.”
The Three Burdens
Automation is not a new concept at Emids. The company has built agents for revenue cycle management, payment integrity, and member engagement for years. The shift from Robotic Process Automation (RPA) to intelligent agents to truly agentic AI is evolutionary, not abrupt, at the firm.
But unlike consumer AI tools, healthcare demands deterministic outcomes. “It needs to repeat and give the same answer every time. That is non-negotiable, the accuracy and determinism of information,” Shankar stressed.
Regulatory compliance is a challenge. Frameworks such as HIPAA in the US or ISO standards in medical devices demand clear traceability. Shankar pointed out how AI is already assisting here, and codifying regulatory checks into models, ensuring audit trails are intact.
In global healthcare supply chains, AI can help trace product genealogy swiftly during recalls, something human teams would struggle with at scale.
Privacy is another pillar. Generative AI tools often risk exposing personal health information (PHI) while handling logs or generating code.
“You have to really, really take care of privacy,” Shankar said, highlighting why a human-in-loop model is still essential.
Learnings and the Road Ahead
Shankar admitted that not all use cases pan out smoothly. Early attempts at revenue cycle automation, for instance, encountered hurdles due to the fragmented nature of healthcare data. Human oversight was necessary to stitch the context together.
Yet, the efficiencies gained are undeniable: “Five years back, 70 to 80% of prior authorisations were manual. Today, only 20 to 30% need human validation.”
On the software side, however, caution remains high. Code generation tools show promise but often hallucinate or leak sensitive information.
The company needs to shift its focus from vibe coding to addressing the core problems it aims to solve. Significant progress has been made in business processes like revenue cycle management and in authoring clinical trial documents, leading to increased efficiency, observes Shankar, alongside improvements in software engineering.
To address these gaps, Emids has built Pacca AI, its proprietary stack named after the nimble alpaca. The system embeds regulatory guardrails, continuous training, and domain expertise into AI workflows. Every engineer, even the CEO, undergoes rigorous certification to ensure a shared understanding of healthcare processes.
Shankar recalled: “When I joined as CEO, even I had to go through a rigorous training program to certify myself, so that I understood that a bit more deeply.”
Emids claims that with 36 years of experience, they offer training platforms that their clients utilise for various purposes, including revenue lifecycle management and payment integrity.
Not Just Responsible, Thoughtful AI
Looking ahead, Shankar sees healthcare’s GenAI adoption following a measured curve, unlike telecom or advertising. In his view, within five years, 60 to 70% of code generation will be AI-assisted, while 80 to 90% of information flows will still retain a human in the loop.
He highlighted that history tends to repeat itself in technology cycles, as seen with web 1.0, 2.0, mobile, and cloud. Similarly, AI will eventually integrate into healthcare, despite a potentially slower adoption rate compared to advertising.
To that, he said, “And I’m glad it does, so that we know what we are getting is of the most potent quality.”
The company is also betting on a hybrid model it calls Forward Industry Deployed Engineering (FIDE), where engineers with over a decade of healthcare experience are embedded directly into problem-solving.
This blend of industry depth and engineering skills helps avoid overdependence on commoditised transformer models from large vendors.
Ultimately, Shankar believes the real shift will not be visible in the backend systems but in how empathy drives design. “I think the EQ of the industry will start driving it, rather than the IQ of the industry,” he said. For healthcare, that may be the most important guardrail of all.
The post ‘Thoughtful AI Matters More Than Fast AI in Healthcare’ appeared first on Analytics India Magazine.