AWS Podcast Episode #754: Accelerating Healthcare Decisions with Agents

Release Date: April 7, 2026
Host: Gillian Ford
Guests: Gigi Yuen (Chief Data and AI Officer, Cohere Health), Kenji Fujita (Staff AI Platform Engineer, Cohere Health)

Episode Overview

This episode dives deep into the practical aspects of deploying agentic AI systems in healthcare, with experienced leaders Gigi Yuen and Kenji Fujita of Cohere Health. The discussion traverses the high stakes of clinical applications, balancing automation with human oversight, integrating domain-specific data, ensuring security and compliance, and harnessing AWS’s Bedrock and Agent Core to drive rapid, safe innovation. While the stories are healthcare-focused, the insights are broadly applicable across industries grappling with AI adoption in regulated, complex environments.

Key Discussion Points & Insights

1. The Healthcare Problem Space & Cohere Health’s Mission

• Defining the Challenge: U.S. healthcare spends an estimated 20–30% of its budget on administrative tasks, half of which are considered low value—amounting to roughly half a trillion dollars wasted annually.
  • “We want to eliminate the waste. When you do that…patients can get the right care faster, providers can actually focus on what they do best…that’s the problem space Cohere Health is set up to solve.”
    • Gigi Yuen, [01:30]

2. Building Trustworthy AI in Healthcare

• Trust and Transparency: AI systems in healthcare must prioritize reliability, non-hallucination, and strict adherence to clinical standards.
  • “We have to get it right the first time...there’s no tolerance for hallucination, none.”
    – Gigi Yuen, [03:33]
• Human Experts from Day One: Clinicians participate in every development project from inception, ensuring evaluation is grounded in frontline expertise.
  • “We must have the experts in the room on the get go…not just at the end for validation.”
    – Gigi Yuen, [05:54]
• Evaluation-Driven Development: Shift from test-driven to evaluation-driven development, with upfront agreement on metrics and continuous monitoring.
  • “For agentic solutions, we really have to move to the mindset of evaluation-driven development…eval comes first.”
    – Gigi Yuen, [05:54]
• Persona-driven Design: Tailoring agent systems by assessing rollout strategies across different user personas for greater utility and adoption.

3. Domain-Specific Data: Motivation, Use, and Rights

• Why Include Domain Data? Not just for accuracy, but for context, control, and operational alignment with care standards.
  • “Is it because I want the solution to run faster, be more accurate, be more contextual? It’s not a matter of that. You want more code control…understanding the motivation…helps us pick the right data.”
    – Gigi Yuen, [08:58]
• Data Rights & Compliance: Clarify how data can be used (operations, training, education), as nuances are critical for regulated industries.
  • “Put a plug into the notion of data use Rights...it's really, really critical. And it's quite nuanced.”
    – Gigi Yuen, [08:58]

4. Human Oversight vs. Automation: Decision Framework

• Risk-Based Approach: Automate low-risk decisions; always maintain human oversight for high-risk or negative determinations.
  • “We always draw a clean line on when we will not automate is when we have to say no to a patient’s care.”
    – Gigi Yuen, [11:33]
• Real-World Example: For routine imaging, more automation is acceptable; for surgery, human review is mandatory.
  • “For a knee surgery decision, we better be absolutely confident before we automatically…say yes to the case without a human review.”
    – Gigi Yuen, [12:56]

5. Technical Choices: Why AWS Bedrock & Agent Core

• Rapid Innovation & Compliance: Chose Agent Core for quick development, robust tenancy controls, intuitive memory management, and compliance with healthcare regulations out-of-the-box.
  • “The key thing that stood out right away was the speed of innovation here, the ability to build these agents quickly, effectively and safely.”
    – Kenji Fujita, [14:37]
• Developer Experience: Workshops, documentation, and ease of setup significantly accelerated progress.
  • “The workshops…with Jupyter notebooks…had everything that we needed right out of the box.”
    – Kenji Fujita, [16:34]
• Operational Impact: Transitioned from custom memory servers and auth proxies to built-in, configurable capabilities, vastly improving velocity.
  • “With a couple lines of code…you can just pass through the same authentication method that we would typically set up on our own with the configuration enabled by Agent Core Gateway.”
    – Kenji Fujita, [19:01]
• Scalability: Shift from building one or two agents per quarter to scaling out multi-agentic systems.
  • “Looking ahead at Q1 and the rest of 2026, it’s full of agents, it’s full of agentic systems, multi agents…a lot of our focus has been on building out the evaluation framework this quarter so that we can scale.”
    – Kenji Fujita, [20:04]

6. Model Selection & Evaluation

• Metrics-Led Choice: Set clear business metrics (accuracy, cost, latency, reliability) and use a leaderboard to compare models in real-time.
  • “It’s always the question, build versus buy…setting up a leaderboard so we have ability behind the scene...to keep monitoring and tracking which models are winning.”
    – Gigi Yuen, [21:04]
• Architecture for Flexibility: Invest in platform and gateway layers to allow composability and swapping models as needed per use case.
  • “It’s not always one size fits all…your architecture must allow composability…otherwise adoption cost becomes unbearable.”
    – Gigi Yuen, [24:45]

7. Measured Impact & ROI

• Automation Rates: Achieved 85% automation in prior authorization decisions, and 30–40% increase in productivity for human-reviewed cases.
  • “85% of decisions are made within minutes…for those 15%...we’ve seen about 30 to 40% improvement in productivity.”
    – Gigi Yuen, [26:29]
• User Satisfaction: Clinical staff report greater job satisfaction—AI handles information retrieval; humans focus on expertise.
  • “…better retention…they can really just focus on the area expertise.”
    – Gigi Yuen, [26:29]

8. Practical Advice for AI Agent Development

• Three Essentials for Scaling from POC to Production:
  1. Evaluation-Driven Development: Build rigorous, relevant evaluation frameworks with expert labeling tied to business metrics. [28:58]
  2. Clarity on Success: Align stakeholders on what “success” looks like at scale, considering people, process, and integration—not just functional AI metrics.
  3. Data Readiness: Ensure data ingestion and integration challenges are addressed early.
  • “We must be very, very clear on what are the important criteria for it to be successful…if you don’t change your processes but you get a new tech, you’re not going to see the benefits.”
    – Gigi Yuen, [28:58]
• Team Composition: Success depends on a mix of tech-savvy domain experts, core platform developers, and data scientists working tightly together. [34:04]

9. Centralized vs. Federated Agent Teams

• Unresolved Experiment: Whether a centralized or distributed agent development model works best is still an open question for Cohere Health.
  • “…is it better to have a small team that focus on agent development in a larger org or…to plug the agent development into every single development team?...The jury is still out.”
    – Gigi Yuen, [32:28]

Notable Quotes & Memorable Moments

• “We have to get it right the first time…There’s no tolerance for hallucination, none.”
  – Gigi Yuen, [03:33]

• “For agentic solutions, we really have to move to the mindset of evaluation-driven development…eval comes first.”
  – Gigi Yuen, [05:54]

• “The key thing that stood out right away was… the ability to build these agents quickly, effectively and safely.”
  – Kenji Fujita, [14:37]

• “We always draw a clean line on when we will not automate is when we have to say no to a patient’s care.”
  – Gigi Yuen, [11:33]

• “Don’t let FOMO get in the way...really always start with the why. You’ll be surprised, there’s still a class of problems that may not be agentic.”
  – Gigi Yuen, [34:04]

• “Ask the questions early and often to the service teams…and it’s helped, you know, uncover the solutions that we would have spent more time trying to figure out ourselves.”
  – Kenji Fujita, [33:39]

Timestamps for Key Segments

• [01:30] – The core administrative challenge in healthcare & Cohere Health’s mission.
• [03:33] – Importance of trust, transparency, and non-hallucination in AI for healthcare.
• [05:54] – Evaluation-driven development: process and rationale.
• [08:58] – Incorporating domain-specific data: motivations, challenges, and rights.
• [11:33] – Framework for human oversight vs. automation.
• [14:37] – Deciding on AWS Bedrock and Agent Core—technical and compliance considerations.
• [20:04] – Acceleration in agent deployment thanks to AWS tooling.
• [21:04] – Model evaluation, business value, and adopting a leaderboard approach.
• [26:29] – Quantitative impact: automation metrics and clinician feedback.
• [28:58] – Three essentials for scaling agentic systems from POC to production.
• [32:28] – Centralized vs. federated agent team models—open question.
• [33:39] – Advice for newcomers to AI agents: start hands-on, leverage AWS resources, ask early.
• [34:04] – Importance of purpose-driven adoption and team composition.

Actionable Takeaways for Listeners

• Start early with evaluation-driven development and bring domain experts in from day one.
• Define clear business metrics and set up processes for continuous evaluation and model comparison.
• Use platforms and frameworks (like AWS Bedrock Agent Core) to speed development, ensure compliance, and enable future scalability.
• Automate where risk is low; retain human oversight for high-risk or high-stakes decisions.
• Never lose sight of data rights and user privacy, especially in regulated industries.
• Invest in change management—AI’s value is realized only when people, process, and data are ready, not just the tech.
• Don’t blindly follow hype—always align agentic adoption with real business need.

Final Thoughts

This episode serves as a masterclass in responsible, effective AI agent deployment in high-stakes environments, fusing practical technical detail with hard-earned leadership wisdom. While the context is healthcare, the lessons around evaluation, team composition, scalability, and ethical deployment apply to any domain ready to embrace agentic intelligence on the cloud.