Ep. 44 | Building Tomorrow's Insurer | The 27-Hour Coding Sprint: How Agentic AI is Transforming Insurance Operations with Alex Taylor (QBE)
Podcast: Building Tomorrow's InsurerPublished On: Mon Feb 09 2026
Description: Are you still trying to figure out if agentic AI is hype or reality? Alex Taylor, Global Head of Emerging Technology at QBE Ventures, cuts through the noise in this no-nonsense conversation about what's actually working in insurance AI - and what's failing spectacularly.Discover why agentic AI isn't just "fancy RPA," how insurers are running shadow mode tests to prove AI can outperform human underwriters, and why the real barrier isn't technology, it's data strategy. Alex shares jaw-dropping examples from software development (27-hour autonomous coding sprints!) and explains how insurers are moving from chatbot failures to genuine operational transformation.Key insights: the difference between vibe coding and provable AI, why observability matters more than accuracy, Microsoft-Allstate's governance playbook, and the one thing every insurance CIO must do in the next 30 days.If you're responsible for AI strategy, digital transformation, or innovation in insurance, this episode delivers the practical framework you've been missing. No vendor pitches. Just real talk about implementation, regulation, partnerships, and what separates AI winners from the FOMO-driven crowd.Timestamps0:00 - Introduction - Alex Taylor & QBE Ventures1:30 - The shift from 'what's possible' to 'what works' in insurance AI2:15 - Why insurers underinvested in technology (and why it made sense)3:45 - The real problems insurers are trying to solve with emerging tech5:00 - Internal pressures: cost, complexity, and competitive speed6:20 - Customer expectations and the value proposition (spoiler: they don't care about AI)7:30 - What actually changed in the last 12-18 months8:30 - Agentic AI explained: beyond classical generative AI9:45 - The critical difference between agentic AI and RPA11:20 - The operating system experiment: 27 hours of autonomous coding13:00 - Inversion of control: humans as engineering managers14:30 - Build vs buy vs partner: how the calculation has changed16:15 - What the ideal tech stack looks like: people, process, tech, governance17:45 - The regulatory complexity and governance requirements18:30 - Snorkel's AI leaderboards and model certification19:45 - Case study: What didn't work (the chatbot mistake 99% made)21:30 - What actually works: agents as employees, not buttons22:15 - Metrics that matter: measuring AI against human baselines23:30 - Shadow mode testing: running parallel systems for 12 months25:00 - Partnership models: how CVCs accelerate experimentation26:30 - QBE's Lighthouse Program: 3-week proof of value27:45 - Cutting through the hype: what's real vs. overstated28:45 - The one thing to do in the next 30 days: know where your data is30:00 - Closing thoughts and where to follow Alex's content
The note was deleted
The note was saved
Your message was sent