Why most AI in healthcare creates more work, not less

The growing narrative is that the AI medical scribe market is oversaturated, but Andreas Cleve, CEO and co-founder of Corti, argues that the real problem isn't too many AI tools—it's that most of them are too generic to deliver value in a hyper-specialised industry. 

Cleve's background is in economics and programming, but he admits, "I got pulled into healthcare. It's the kind of space that once you're in, it's hard to leave — it's so intellectually stimulating and meaningful to help professionals do critical work."

"What struck me was the overwhelming administrative overhead in healthcare.

I started thinking about how to reshape the supply-demand curve and get patients better access to care. That's where the idea for Corti came from: a purpose-built AI that understands medical interactions and can augment care delivery."

AI-powered clinical assistance

Corti has developed an AI-powered clinical assistant that enhances patient-provider interactions by listening to conversations in real time, transcribing them, and offering decision support. It suggests relevant follow-up questions and provides diagnostic insights to ensure no critical information is missed during consultations. 

Corti also assists with medical coding by recommending appropriate ICD-10 codes, streamlining billing processes. Automating documentation and organising clinical notes significantly reduces administrative burdens and improves accuracy. 

However, the company also empowers developers to build and integrate tailored AI functionalities into their own healthcare platforms.

Augmenting, not replacing

According to Cleve, the quality of patient care improves dramatically when you have highly trained professionals. The challenge is we don't have enough of them. 

"Our thesis is that augmenting care delivery—intake, consultations, follow-ups—is where AI can make the biggest impact, both for clinicians and patient outcomes.

Think of the consultation. It's a transaction of information between people, rooted in history going back to Hippocrates. AI should augment that moment, not replace it. And as AI improves medical reasoning, we'll unlock many more use cases across the care journey."

At Tech.eu we get plenty of pitches from companies aiming to bring AI and deeptech experience to healthcare, especially for doctors' clinics. 

For example, they work to streamline workflows, create actionable data, provide real-time insights, and overall reduce the administrative bottlenecks that mean healthcare providers spend more time with paperwork than patients. 

However, Cleve contends that every medical specialty has deep, nuanced workflows, and even small changes to those workflows matter a lot. 

So when AI tools aren't trustworthy or purpose-built, they become an added burden.

Why ambient scribes fall short

Ambient AI scribes, tools that passively listen and document clinical encounters, are growing in outpatient settings, especially in the US. 

The problem is that most of these scribes are built on general models like OpenAI's, which aren't designed for medicine. 

Cleve asserts that "without fine-tuning or alignment to clinical use, they won't get better. Doctors need systems built specifically for their domain."

Corti surveyed thousands of doctors and found that many early adopters spend up to three hours a week reviewing and correcting their AI's notes. 

"That's a new task, not a relief," asserts Cleve.

The problem of "pilot paralysis"

Further, research by Corti earlier this year revealed that while 74 per cent of healthcare professionals in Europe support AI use in practice, 52 per cent say they wouldn't feel confident using current AI solutions in their work. 

This crisis of confidence has contributed to what industry experts call "pilot paralysis" - widespread AI trials that fail to advance beyond testing due to accuracy, cost, and integration challenges.

To accurately interpret patient histories, lab results, and evolving medical research, Corti trains dedicated reasoning agents informed by millions of healthcare cases rather than relying on general, non-specialist inputs.

Additionally, Corti prioritises alignment and interoperability by building interpretability into its models, enabling clinicians to trace AI reasoning and leveraging alignment models to detect inconsistencies in outputs before they lead to harm.

Further, Corti has developed specialised infrastructure and APIs that empower these apps to go further - enabling them to deliver smarter, more reliable solutions that can adapt to clinicians' needs."

A flexible, powerful AI stack

High-performing across multiple languages and medical specialties, Corti's AI complies with strict medical regulations, offering transparency, explainability, and results that augment, not replace, healthcare professionals. Inspired by medical residency, the system adopts an "AI residency" approach: a training process through human supervision that gradually takes on more responsibility.

It has developed three foundational models:

Solo: a fast model with audio reasoning. Builds expert clerk and transcription agents, handling complex medical terminology in over 10 languages while integrating effortlessly with existing systems

Ensemble: a powerful model focused on exceptional documentation. Builds agents that turn consultations into action, transforming medical discussions into structured documentation that is 25 percent more concise and more accurate than general-purpose AI.

Symphony: merging powerful reasoning with speed: Builds agents for real-time clinical support, operating 35x faster than GPT-4 and providing evidence-based insights during patient consultations.

In addition to these core models, customers can integrate up to 20 expert models that function like healthcare specialists, purpose-built to build agents that tackle tasks like medical coding, quality control, and summarisation. 

This flexible architecture ensures seamless integration into existing workflows, helping healthcare systems leverage AI without the disruption or uncertainty often caused by solutions wrapped around general-purpose models.

Ultimately, Corti aims to make high-quality healthcare AI more accessible and affordable. That means lowering the cost and increasing the quality of its models so that others can build with them, by opening more API endpoints, including transcription, summarisation, dictation, patient engagement, letter generation, and more.

Corti also wants developers to easily build new healthcare apps and services, even in niche specialties. 

Cleve shared that, for example, building a scribe app a year ago might have required $3 to 4 million in funding. 

"Now, we've built a demo scribe in seven minutes using our tools. That's the direction we're going: enabling rapid, compliant innovation at scale."

Clinicians as co-creators

According to Cleve, trust is built slowly. This is why Corti  publishes a lot of research, stays transparent, and prioritises compliance.

It's also launching new accessibility features that make its platform easier to adopt and evaluate so clinicians can try it, compare it, and iterate.

"The goal is to put these tools in the hands of people who know their domain. Doctors shouldn't settle for generic workflows. With AI, they can help design purpose-built tools that actually work for them."

As clinicians become part of the tech creation process. Cleve contends that "I don't think the old story of 'doctors don't understand tech' holds anymore. AI is changing that. In the future, clinicians will design their own workflows. Why settle for generic, when you can have something that fits you perfectly?"

We're not there yet when it comes to personalisation

Lately, everyone in healthcare talks about personalisation. I asked Cleve if he sees this reflected in AI solutions today. He shared, 

"Honestly, no. We've done a lot of research into personalisation, and while it's a great concept, especially in medicine, the implementation is limited.

Imagine I'm a doctor working in a specific department, within a provider system, governed by insurance constraints. How much of my AI assistant can — or should — be personalised?

You can't personalise away from value-based documentation, regulatory requirements, or systemic care standards.

So, before we even get to personalisation, we need to help clinicians consistently achieve best practices. Let's first build AI that reliably augments care and then move toward deeper personalisation."

In a market crowded with generic AI tools, Corti stands out by building deeply specialised, clinician-aligned systems that prioritize trust, transparency, and real-world usability.