Jérôme Vetillard

Interesting paper alert - a commentary in Nature Medicine highlights a pivotal tension in healthcare today: while #AIagents promise transformative impacts — from autonomous triage to multi-modal clinical decision support — regulatory frameworks remain built for a different era (a recurring scheme in the world of Health AI which I do not see going away anytime soon: technical progress in AI will always outpace regulatory actions). Here are the take aways from the paper: 🔹 Current state of regulation of AI agents in health and medicine: Most regulatory pathways, like the EU Medical Device Regulation and the U.S. FDA’s 510(k), are optimized for narrow, static, non-adaptive tools. Approval of AI-driven devices with broad scope and high autonomy remains rare and challenging. 🔹 The problem (why existing regulatory frameworks fall short of capturing agentic AI applications): AI agents orchestrate multiple tools, access sensitive data, and make goal-directed decisions autonomously. Their adaptive nature, coupled with non-deterministic outputs and complex multi-component architectures, pushes them outside the assumptions underpinning today’s medical device regulation. 🔹 Proposed solutions (how can we explore and experiment safely with AI agents in clinical settings in the absence of comprehensive regulatory guidance): The paper explores several emerging regulatory strategies, including: Voluntary alternative pathways → tailored approvals for high-risk AI agents Adaptive oversight → dynamic, real-world performance monitoring Regulatory sandboxes → controlled environments to experiment safely Outcome-driven evaluation → focusing on patient-centered results rather than static technical compliance 💡 The takeaway (so where does this leave us): While regulation is still playing catch-up with the risks and complexities of AI agents in healthcare, it is possible to safely experiment and sandbox their use — provided there are clear guardrails, strong oversight, and rigorous evaluation frameworks. This balanced, sandbox-first approach could enable innovation without compromising patient safety. For first hand insights on how this is all playing out in the real world right now, have a read of these two recent papers from Eric Topol, MD (https://lnkd.in/gEycp__S) and Harsha Nori (https://lnkd.in/gKNxpxsA) John Lim Christopher Bain ZongYuan 宗元 Ge 戈 Deval Mehta #AI #Healthcare #Regulation #AIagents #DigitalHealth #NatureMedicine

136

13 commentaires

🚀 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗻𝗴 𝗚𝗲𝗻𝗔𝗜 𝗶𝗻𝘁𝗼 𝗥𝗪𝗘 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀 The Future of Evidence Generation is here. Last week, we have launched three GenAI products to augment the end to end #RWE workflows within #Novartis. AI is reshaping the tempo of Real World Evidence generation workflows. At #Novartis, three pioneering AI tools are enabling teams to move faster, smarter, and without compromising the quality. 💬 𝗥𝗪𝗘 𝗱𝗮𝘁𝗮 𝗲𝘅𝗽𝗹𝗼𝗿𝗮𝘁𝗶𝗼𝗻 𝗔𝗴𝗲𝗻𝘁: Built on Novartis cutting-edge AI platform #Data42, this AI agent empowers researchers across Research, Development and Commercial organizations to explore RWD using natural language—no code, no friction. It enables end to end data exploration including rapid feasibility assessments or patient journey mapping. 📝 𝗥𝗪𝗘 𝗽𝗿𝗼𝘁𝗼𝗰𝗼𝗹 𝘄𝗿𝗶𝘁𝗶𝗻𝗴 𝗔𝗴𝗲𝗻𝘁 is our GenAI-powered authoring solution, is transforming the way we create all types of scientific documents for Real-World Evidence (RWE) studies at Novartis. With AI-generated drafts aligned to Novartis templates and standards, teams are moving from concept to study start in days—not weeks. ⚡ 𝗥𝗮𝗽𝗶𝗱 𝗔𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀: Data Engineering pipelines is the turbocharger of faster RWE study execution. With this rapid engineering pipelines, it delivers low to medium complex studies in weeks—not months. 🔄 𝗔 𝗦𝗲𝗮𝗺𝗹𝗲𝘀𝘀 𝗔𝗜-𝗔𝘂𝗴𝗺𝗲𝗻𝘁𝗲𝗱 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄: These tools don’t just coexist. Together, they form an end-to-end AI-augmented workflow that’s already in production and scaling across Novartis. 🌟 𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗠𝗮𝘁𝘁𝗲𝗿𝘀: This is more than operational efficiency. It’s about empowering associates across Novartis with time to focus on what matters most—driving innovation and delivering better evidence for patients. We’re not just imagining the future of evidence generation! We’re building it—today. Thanks to everyone who made this possible! Our next AI augmented workflow would be starting from Integrated Evidence Planning to Publication writing. Thrilled to keep advancing the journey toward AI-augmented evidence generation workflows! #RWE #RWD #AI #GenAI #GMA #evidencegeneration #agents Adrian Cassidy Kristijan Kahler David Wormser Rachel Studer Maxwell Lawson Olga Greco Sathish Kaveripakam, MS,MBA Birgit Schoeberl Sepideh (Sepi) Farivar Varon Swetha R Palli Jamie Geier Guillaume Wendt, PhD Toshifumi Sugitani Dimitris Papanicolaou Valérie Lehnert

148

13 commentaires

Beware the moving frontier ! I've seen too many companies or individuals tell me that OpenAI or Gemini isn't good at <x> and this why their solution is better. The problem is that might have been true 6 months ago or whenever they did the eval. But these models are not static. So when I do an evaluation I find better out of the box performance at 80% of their tasks. Right now you need to be continually evaluating. And if you don't have a unique data asset, or a closed loop to generate data you don't have a moat. Everyone can access a frontier model. And for those skeptical of the transformer architecture, I think there is still alot to be gained from these models from data, compute and algorithmic advances around the edges. Check out the Metr benchmark as an example, if the last time you looked was O3, you'll be wrong in your planning. https://lnkd.in/gKqUcD3r

145

14 commentaires

Explore Anthropic's Claude Code powered by Reuven Cohen CLAUDE-FLOW, the emerging AI Operating System for the enterprise. Here are the essentials I found while tracing the codebase: → The CLI is the front door. The main entry point sets up commands and runs them. Simple, clean, and extensible. Classic Command Pattern. → An EventBus as the nervous system. Instead of modules calling each other directly, they publish and subscribe to events. This keeps the system decoupled and easier to test. → A clear hierarchy. The Orchestrator is the CEO, the SwarmCoordinator acts like a project manager, and the AgentManager handles HR and Ops. Each has one job and one job only. → Agents defined in Markdown. New agent types can be created by writing a .md file with YAML frontmatter for capabilities and a system prompt in the body. No TypeScript required. → Agents managed like cloud services. Each agent runs in its own process, with health monitoring, auto scaling, and a registry that can select the best agent for the task. → Flexible memory. Options include in-memory, Markdown, SQLite, or even a hybrid backend that gives you both fast queries and human readable logs for version control. → Quality is baked in. Directories for benchmarks, validation tests, and a 90 percent coverage target show this is engineered for reliability, not a weekend experiment. The bigger idea: CLAUDE-FLOW shows how AI development may shift from writing code line by line to orchestrating autonomous agents that collaborate, scale, and adapt. I am still early in this journey, learning file by file, mistake by mistake. But it is already clear this is a glimpse into how enterprise AI might look like. If you have explored CLAUDE-FLOW or similar multi-agent frameworks, what patterns stood out to you? Thank you Jason La Barbera for pulling me into Reuven’s world.

95

16 commentaires

🚨 GenAI isn’t coming. It’s already reshaping your role. You can either watch the shift — or drive it. The AI landscape is evolving at breakneck speed: 🔹 GPTs are becoming autonomous agents 🔹 Grok Heavy brings deeper real-time reasoning 🔹 ChatGPT-5 promises multimodal mastery and memory ⚡ What does that mean for you? Tasks that once needed a team — research, coding, summarizing, planning, even decision-making — are now handled by personal AI agents, 24/7, tuned to your context. This isn’t a trend. It’s a new operating system for thinking, building, and leading. 💡 I’ve reverse-engineered what separates passive learners from next-gen builders. Spoiler: it’s not talent. It’s learning alignment — knowing which skill, at which time, for which impact. 🎯 That’s why I curated mission-ready learning paths — not for everyone, but for who you are right now: 🧪 The Experimenter — Fast ideas, fast builds Build AI-native tools, test agent workflows, launch fast. Top picks: LangChain playgrounds, Vibe coding, multi-agent orchestration. 📊 The Strategist — Understand, compare, decide Evaluate GenAI stacks, frame strategy, avoid hype. Top picks: Prompting & AI safety, cloud-native labs, tooling reviews. 🔬 The Educator or Analyst — Turn clarity into influence Master concepts, explain with impact, upskill others. Top picks: Conceptual deep dives, LLM explainers, open-source labs. No fluff. No “watch 10 hours then forget”. 📦 Just pick a persona. Run one experiment. 🚀 That’s how transformation starts — with one small, real output you actually ship. AI Agents aren’t a niche anymore — they’re the new normal. The question isn’t if, it’s how fast you adapt. 👇 So — which version of you is about to level up?

89

9 commentaires

🤚🏼🛑  Here you have my slides from the API Conference New York As AI becomes increasingly API-driven, we face a critical question: 👉 If APIs are the arteries of AI, what’s flowing through them—just data, or also values? 👈 ⚖️ Why Ethics and Responsibility Can’t Be Optional When flawed decisions travel through widely used APIs, the harm multiplies. From biased credit scoring to opaque hiring filters, the consequences can be costly legally, reputationally, and ethically. With regulations like the EU AI Act on the rise, building responsible APIs is becoming a competitive advantage. 🛠️ The IGT-AI Framework To help structure this challenge, I recommend the IGT (Ikenna Governance Toolkit)-AI framework. It provides a blueprint for embedding ethical guardrails through: 1️⃣ Unified API & Access – Secure, centralized authentication and credentials. 2️⃣ Guardrails – Input/output filtering to stop harmful or non-compliant content. 3️⃣ Observability – Monitoring fairness, drift, and usage patterns. 4️⃣ GitOps for Governance – Making every configuration auditable and version-controlled. 🔑 Key Insights from My Talk ▶️ Transparency APIs should expose model version, confidence scores, and limitations. ▶️ Fairness Deployments should pass bias tests, with fairness metrics exposed in real time. ▶️ Accountability Clear ownership and tamper-evident logs ensure responsibility. ▶️ Privacy & Security Apply data minimization, encryption, and consent tracking at the API layer. ▶️ AI Gateways The central control point for implementing and enforcing all of the above. 💬 I’d love to hear your thoughts: How are you approaching responsibility in AI-powered APIs? What challenges do you see in balancing innovation with governance? I'm reading your comments! ✍ ✍ ✍

94

9 commentaires

🚀 What a year of agentic AI has taught us Just read “One year of agentic AI: Six lessons from the people doing the work” by QuantumBlack, AI by McKinsey — super helpful if you’re thinking about diving into agentic AI or scaling your efforts. https://lnkd.in/gXKQuvZc 🔑 Key takeaways: • It’s not enough to build fancy agents — you need to redesign the workflow around them. • Agents aren’t the answer for everything. Sometimes simpler automation or predictive models are better. • Trust and evaluation are essential — “AI slop” (weak outputs) kills adoption fast. • Reusability matters — build agentic components that can be used across tasks to avoid reinventing the wheel. • #Humans are still #central. Their roles might shift, but oversight, judgment, edge-case handling are things only humans can reliably do right now. If you’re leading or exploring agentic AI programs, these lessons are worth locking in. #AgenticAI #GenerativeAI #Leadership #AITransformation #WorkflowDesign

35

2 commentaires

🚀 𝐑𝐞𝐭𝐫𝐢𝐞𝐯𝐚𝐥 𝐀𝐮𝐠𝐦𝐞𝐧𝐭𝐞𝐝 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧 (𝐑𝐀𝐆): 𝐓𝐡𝐞 𝐁𝐚𝐜𝐤𝐛𝐨𝐧𝐞 𝐨𝐟 𝐂𝐨𝐧𝐭𝐞𝐱𝐭𝐮𝐚𝐥 𝐀𝐈 𝐢𝐧 𝐄𝐝𝐮𝐜𝐚𝐭𝐢𝐨𝐧 📚🤖 In the age of GenAI, we’re seeing how Large Language Models (LLMs) like GPT-4, Claude, and Gemini are transforming learning experiences. But when it comes to domain-specific accuracy — especially in education — LLMs need more than just general training. That’s where RAG (Retrieval Augmented Generation) steps in. 🔍 Why RAG? ➡️ Most LLMs are trained on open web data — which doesn’t include localized curricula or teacher guides. ➡️ RAG integrates custom educational materials (PDFs, lesson plans, policies) into a system that retrieves relevant info in real-time. ➡️ By embedding your own content into a Vector Database, RAG enables the LLM to deliver grounded, personalized, and curriculum-aligned answers. 📈 Use Cases in EdTech: ✅ AI-powered lesson planning (e.g., "Design a gravity lesson plan for Grade 5") ✅ Policy-compliant content generation ✅ Multilingual support using adaptive educational content ✅ Structured pedagogy for various user roles (teachers vs administrators) 🧠 Behind the scenes: Input is transformed into an embedding vector. The system retrieves semantically similar chunks from your content. This retrieved info augments the prompt. The LLM generates an answer based on both the query and context. 🔗 Dive deeper: https://lnkd.in/eA_Vb6nJ 📘 Resource: https://lnkd.in/eV6NPZJP 🌍 RAG is not just a technique — it's a paradigm shift in how we scale quality education with AI. #RetrievalAugmentedGeneration #RAG #LLM #GenAI #AIinEducation #EdTech #PersonalizedLearning #VectorDatabases #Embeddings #Chatbots #OpenAI #GPT4 #CurriculumAI #AIForGood #EducationTransformation #MachineLearning #KnowledgeGraphs #ContextualAI #PromptEngineering #AItools #TeacherSupport #ScienceOfTeaching #AI4Ed

91

33 commentaires

𝗔𝘁𝘁𝗲𝗻𝗱𝗶𝗻𝗴 𝘁𝗵𝗲 #testmuconf 𝗖𝗼𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 2025 LambdaTest 𝗛𝗲𝗿𝗲’𝘀 𝗪𝗵𝗮𝘁 𝗜 𝗟𝗲𝗮𝗿𝗻𝗲𝗱 𝗔𝗯𝗼𝘂𝘁 𝗔𝗜’𝘀 𝗘𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 from Siddhant Wadhwani What a thrill to dive into the future of AI at the #testmuconf Conference 2025! The theme Generative to Agentic to Quantum – The Evolution of AI was a masterclass in how AI is reshaping industries—especially software testing and automation. Here are my top takeaways:  2025: 𝗧𝗵𝗲 𝗜𝗻𝗳𝗹𝗲𝗰𝘁𝗶𝗼𝗻 𝗣𝗼𝗶𝗻𝘁 𝗳𝗼𝗿 𝗔𝗜 2025 marks a pivotal shift where AI transitions from outputs (e.g., generating text/code) to outcomes (solving real-world problems). We’re moving from generative AI (helping humans) → agentic AI (doing tasks independently) → quantum AI (discovering new solutions). This means testing and automation will rely on AI that plans, executes, and adapts—not just reacts.    GPT-5 & Beyond: The New AI Frontier The launch of GPT-5 is a game-changer:  -Multi-step reasoning: Solves complex problems with fewer hallucinations.   -Unified model family: Handles text, images, code, and more in one system.  Enterprises are already piloting autonomous developer/QA agents that plan, execute, and validate work with minimal human input. Older models are being phased out—making today’s AI far more reliable for business-critical workflows.  LLMs, Agentic AI, & Quantum: What’s Next. -LLMs: Expect a wave of new releases in 2025 (GPT-4.5, Claude 4, Llama 5, Gemini 3, and more!). These models will be larger, more specialized, and better at domain-specific tasks.  - Agentic AI: Think AI agents that act autonomously setting goals, reasoning, and learning from feedback. The Agentic AI Loop (Set Goal → Plan → Observe → Learn) is revolutionizing how we build intelligent systems.  - Quantum AI: Solves legacy AI pain points (resource intensity, optimization bottlenecks, lack of transparency). Tools like Google’s Willow chip and Microsoft’s Majorana 1 are pushing boundaries—enabling AI to tackle problems classical systems can’t.  Gen AI in Testing & Automation: The PALI Framework For testers and automation engineers, Gen AI is a force multiplier:  - Predictive analytics optimize testing processes.   - AI analyzes test results to predict defects.   - leads to more efficient testing cycles.   - Intelligent insights improve decision-making.    𝗙𝗶𝗻𝗮𝗹 𝗧𝗵𝗼𝘂𝗴𝗵𝘁: The line between human work and AI work is blurring and that’s exciting! As AI evolves from helper to doer to discoverer, our roles in testing and automation will shift toward guiding and validating AI systems. The future isn’t just about what AI can do—it’s about how we collaborate with it.  What’s your take on AI’s evolution? Drop your thoughts below!  #AIevolution #GenerativeAI #AgenticAI #testmuconf #testmuconf2025 #TestMuConf #lambdatest

47

27 commentaires