# When AI Stops Being a Project: Turning Technology into Real Value for Patients and Providers

## Executive summary

The conversation explores the shift of AI from a mere 'project' to an integral business function in healthcare. Key focus areas include moving beyond simple Q&A chatbots to complex, long-running agentic workflows that can handle tasks previously requiring many hours of human effort. Speakers emphasize that while the technology is rapidly advancing (e.g., GLM 5.2 and advanced LLMs), successful enterprise adoption requires significant architectural changes: establishing robust governance, managing token costs, mitigating vendor lock-in, and fundamentally reimagining existing clinical workflows rather than simply automating point solutions.

## Key takeaways

- Shift to Agentic Workflows: AI is moving past simple Q&A (quick, short, transactional) toward complex, long-running agentic tasks. OpenAI internal data suggests agents are now performing work across finance, recruiting, and legal that can take up to 8 hours of human effort [0:14:25].
- Governance and Architecture are Critical: For large enterprises (like United Health Group), long-form agentic work requires establishing strong governance, guardrails, and security protocols. Simply calling an API a 'super agent' is insufficient; true agency requires reasoning and decision-making capabilities [0:23:15].
- Addressing Vendor Lock-in and Cost: Enterprises must manage the risks of vendor lock-in when restructuring workflows around a single model or API. Concerns include escalating token costs and geopolitical instability, making architectural flexibility paramount [0:47:28].
- Reimagining vs. Automating Silos: The most impactful approach is not to automate existing tasks (silos) but to fundamentally reimagine the entire end-to-end workflow—from patient intake to payment processing—given the new technological capabilities [0:35:12].
- AI as a Business Mandate: Successful adoption requires AI to be elevated from an IT or innovation problem to a CEO-level, cross-functional business mandate. Leadership must obsess over defining the core metrics (the 'what is the metric?') and driving change at scale [0:57:12].

## Technical details

- LLM Capabilities & Performance: Generalized LLMs are showing performance gains compared to specialized clinical AI tools. The trend suggests that providing models with sufficient data and compute power allows them to outperform highly specific, encoded medical expertise [1:29:50].
- Model Architecture & Efficiency: The GLM 5.2 model is noted for being significantly cheaper than frontier models while achieving comparable quality in certain benchmarks (e.g., coding). However, running such large parameter models requires substantial infrastructure [1:43:46].
- Enterprise AI Infrastructure: One major health system utilized a 'harness' or studio containing multiple LLM models (including internal open-weight SLMs) and a token gateway to manage consumption, monitoring, and prevent excessive costs [1:54:30].
- Workflow Design Principle: The focus should be on building an 'experience' or workflow that embeds different models, rather than attempting to train a specialized model from scratch on proprietary data. The architecture must allow for easy model swapping (model agnosticism) [1:35:40].

## Practical implications

- Health systems must shift focus from building point solutions to redesigning entire end-to-end patient and payment workflows.
- Architectural strategies (like 'harnesses' or gateways) are necessary to manage token costs, monitor usage, and prevent vendor lock-in when integrating multiple LLMs.
- AI adoption requires executive leadership buy-in that treats AI as a core business transformation, not merely an IT upgrade.

## Topics

Generative AI, Healthcare Technology, LLM Governance, Agentic Workflows, Digital Transformation, Stanford Online Healthcare AI programs, OpenAI internal data on work changes

Source: https://www.youtube.com/watch?v=ZlLV9ZHauAM