Many enterprises are hitting a wall with AI. After deploying copilots and RAG-based chatbots, they're finding that the most valuable, complex processes-like flight rebooking or insurance claims-remain untouched, and true ROI is elusive.The solution? Moving from assistive AI to fully autonomous agents. But this requires specializing models to reliably handle complex, real-world actions, a task that frontier models often fail.Anand Nimkar (Partner at Deloitte's Applied AI Practice) and Julien Launay (Co-founder & CEO of Adaptive ML) are on the front lines of this challenge. They partner to deploy sophisticated AI systems for global enterprises, using reinforcement learning (RL) to teach models how to act safely and effectively within legacy IT environments.In this episode, they provide a blueprint for moving beyond endless pilots, using RL to create hyper-adapted models, and tackling the true challenges of enterprise automation - and much more.(00:00) Intro(05:03) The integrator + deep tech model (Deloitte & Adaptive)(10:27) Market status: Moving from POCs to production(14:56) Use Case: Automating an airline’s complex customer support(29:17) Use Case: A 100-million-call-a-year health insurer agent(38:40) Use Case: Why frontier models fail at EdTech (and how to fix it)(50:44) The case for smaller, specialised models vs. giant LLMs(57:13) Final advice: "Be more ambitious"Let’s stay connected!Find Battery on LinkedIn:https://linkedin.com/company/battery-ventures/⁠Find Battery on X:⁠https://x.com/BatteryVentures⁠Follow Anand on LinkedIn:https://www.linkedin.com/in/animkar/Follow Julien on LinkedIn:https://www.linkedin.com/in/julienlaunay/Follow Lior on LinkedIn:https://www.linkedin.com/in/liormallul/⁠

Ver todos los episodios