JUST NOW POSSIBLE with Teresa Torres
Episode: Building Agent Studio – How Medable Is Using Agentic AI to Accelerate Clinical Trials
Recorded: March 2026

Episode Overview

This episode dives deep into how Medable, a clinical trials technology company, is building and deploying an agentic AI platform—Agent Studio—to solve significant workflow and data challenges across clinical trials. Host Teresa Torres is joined by the Medable team: Luke Bates (Product Leader, Agent Studio), Jen (Product Management), Matt Schofield (Product Designer), and Fikra Matthews (Principal Architect). Together, they cover the journey from spotting mammoth customer pain points to designing, testing, and iterating on a flexible agent platform that accelerates clinical trials globally—discussing everything from workflow orchestration to data layers, retrieval strategies, evaluation methods, regulatory compliance, and product vision.

Key Discussion Points & Insights

1. What Medable Does and the Clinical Trial Challenge

• [01:34] Jen frames Medable's mission: “Our mission and vision is to bring effective therapies to patients faster.”
• Current clinical trial processes take over a decade and generate a staggering volume of documentation.
• Medable focuses on reducing these timeframes, especially with tools for remote patient engagement (e-concents, electronic assessments) and, more recently, agentic AI solutions that automate and optimize clinical trial workflows.

“There are single studies that produce tens of thousands of documents per month. It's a lot of documentation.”
— Fikra Matthews [14:25]

2. Evolution Toward Agent Studio and Platform Thinking

• Medable tackles two major pain points: immense cognitive load on staff and complex, fragmented data sources.
• [15:19] Luke: “It was a complete natural fit… this high cognitive load... that type of problem is really well suited to be able to bear RAG knowledge for a protocol and provide easy answers versus having to depend on a CRA.”
• The Agent Studio platform allows both internal teams and customers to build, deploy, and customize AI agents tailored to their own environments, leveraging pluggable models, workflow automations, and connectors (MCPs).

“We take a platform approach to every solution… so that the next solution that comes around will have that capability already baked in.”
— Luke Bates [05:21]

3. The Flagship Agentic Applications: ETMF and CRA Agents

Electronic Trial Master File (ETMF) Agent

• [06:37] Jen: The ETMF system manages the tens of thousands of trial documents required for regulatory review.
• Users upload documents and assign metadata manually—a process ripe for automation given over 350 classification categories.
• The ETMF agent automates document classification and metadata assignment, starting with human-in-the-loop to ensure trust and accuracy.

Clinical Research Associate (CRA) Agent

• CRAs monitor data across 13+ disparate systems —time-consuming and error-prone.
• The CRA agent aggregates these sources, surfaces insights, and can recommend—and sometimes even take—actions, all with human oversight.

“Legacy systems today can surface a potential signal... but the AI can actually go one step further and take the action for you on behalf of the human. Again, with that human in the loop.”
— Jen [09:53]

4. Inside the Agent Studio Platform

• Enables both engineers and non-engineers to configure agents using any model (“bring your own model”), connect to multiple data sources, and orchestrate workflows via skills (Anthropic protocol), triggers, and MCP connectors.
• “Our platform allows users, not just engineering users, to be able to configure their own agents using agnostic models.” — Luke [03:50]
• Broad deployment patterns: user-facing apps, internal augmentations, and custom builds for clients.

User Experience: Customization Spectrum

• Medable supports a range of customer technical capabilities:
  • Some clients want "out of the box" solutions (pre-built agents).
  • More advanced users assemble their own multi-agent systems with platform Legos.
  • Middle ground: easy configuration of parameters without coding.

“You can just use it in other ways. You're saying, here's the Lego blocks, make whatever you need... and then maybe there's even this middle ground, like, [you] can pick the color, you can pick how big the thing is... offering that full range. Is that a fair analogy?”
— Teresa Torres [28:08]
“Yeah, exactly.” — Matt [28:40]

5. Data Retrieval, Ontologies & Context Window Management

• Massive, heterogeneous data is a consistent obstacle.
• The team invests in an AI-powered data layer, aligning all sources to a “common ontology”—making sure synonyms across systems (e.g., 'participant') map correctly for agents.

“We use the phrase RAG a lot, but RAG is just so... people tend to align it with vector DBs and embeddings... we allow customers to create simple, identifiable sort of little vector pools, but we also have more complex systems where you’re talking about data hierarchies and layers of summarization.”
— Fikra Matthews [33:16]

“We need that common layer that allows us to understand… when we talk about the 13 different systems that a clinical research associate uses… you need that common layer.”
— Jen [36:25]

• Platform enables multiple retrieval strategies (embeddings, keyword search, file traversal) and aims to automate smart selection of these as the field matures.

6. Orchestration, MCP Connectors & Tool/Skill Management

• MCP connectors are the main protocol for plugging in both internal and third-party data/services, layered with custom authentication.
• Key lesson: API/connector quality is often dictated by external vendors’ capabilities.
• Sub-agent architectures and manual/automatic tool filtering are used to avoid context window bloat and improve efficiency.

“When you’re building MCPs for some of these external tools, you’re at the mercy of the other vendors’ API mechanisms... we’ve used agent skills to help prime MCPs.”
— Luke [45:22]

“We also have a sub-agent that will filter tools before passing to the main agent to avoid context rot... all these options work well in different circumstances.”
— Fikra [47:59]

7. Evaluation Strategies & Ensuring Reliability

• All agent configurations pass through a rigorous evaluation phase before launch—both at the Agent Studio/platform level and product (ETMF, CRA) level.
• Golden datasets (e.g., 2,000 documents for ETMF) used for pre-launch accuracy assessment and ongoing human-in-the-loop review for continuous improvement.

“We’re able to monitor... which recommendations the human’s agreeing with and which they aren’t. Humans aren’t always 100% correct… there’s a continuous kind of evaluation process we go through.”
— Jen [54:14]

• Human-in-the-loop corrections aren’t always ground truth; the team collaborates with clients to resolve ambiguous cases and improve both agent and human accuracy.

8. Regulatory (GXP) Compliance & Industry Requirements

• Health and drug trial industries are heavily regulated; Medable’s platform is built to meet and document specific intent, design, and test traceability in line with GXP (Good Clinical Practice and similar frameworks).
• Evals and documentation aren’t just QA; they’re foundational to regulatory acceptance.

“We need to be able to also prove… that we’re building these agents to a specific intent… evals become really important in that part.”
— Luke [60:16]

9. Notable Quotes & Memorable Moments

• “Our ambitious, big hairy goal is one year [for clinical trials].”
  — Jen [11:14]

• “There’s no one right way of [data retrieval]. Everything we’ve tried has been good in some way and bad in another.”
  — Fikra [33:16]

• “We want to enable that full, end-to-end clinical trial process that has these agent workers helping along the way. Today there’s 10,000 uncured illnesses; at current pace, it’ll take 200 years to get to market.”
  — Jen [63:50]

• “You can’t even read books about the stuff that’s being advanced right now. You got to stay on top of the latest papers and YouTube videos... It’s a different way to learn than traditional SaaS platform technologies.”
  — Luke [19:30]

• “Why isn’t this doing the exact thing my traditional system should do? ...We shouldn’t be just comparing these agents to these systems. We should be comparing them to humans. Humans make errors, too.”
  — Luke [62:38]

Timestamps for Key Segments

• [01:34] – Medable’s mission and industry challenge
• [05:21] – Agent Studio platform approach overview
• [06:37] – ETMF agent: document/metadata automation problem
• [08:56] – CRA agent: unifying trial data & workflow support
• [13:33] – Clinical trial timeframes and manual processes
• [15:19] – Why a platform over point solutions?
• [18:14] – Team’s AI/ML learning curve and backgrounds
• [22:03] – Timeline: building Agent Studio amid AI/agent revolution
• [23:24] – Modular agent and sub-agent system explained
• [25:39, 28:08] – Customer customization spectrum; LEGO analogy
• [30:23] – Skills protocol & platform evolution over time
• [32:34] – Data layer, ontologies, and rag/retrieval strategies
• [42:41] – MCP server implementation, security, and design
• [45:22] – Context window management strategies
• [50:47] – Latency, cost, and UX tradeoffs in agentic workflows
• [50:57] – Evaluation harnesses: pre- and post-deployment
• [54:14] – Continuous evaluation, handling human disagreement
• [58:06] – GXP regulatory compliance challenges
• [63:48] – The vision: “full self-driving” clinical trials

Final Thoughts & Vision

• Medable’s north star is end-to-end automation of clinical operations—reducing time-to-market for therapies from years to months.
• The Agent Studio approach blends robust platform thinking, cutting-edge AI techniques, regulatory rigor, and deep customer empathy.
• Key theme: Shift from AI being a “black box” to building reliable, transparent, customizable, and compliant agentic solutions for some of humanity’s hardest problems.

Listen If...

• You want practical insights into building and shipping real-world AI/agentic products.
• You’re interested in how cutting-edge AI techniques intersect with highly regulated industries.
• You want to learn from builders at the intersection of product, engineering, design, and compliance.

Memorable closing moment:

“It’s very refreshing to hear people use especially this type of technology for genuinely hard human problems. So keep it up.”
— Teresa Torres [65:32]