The Lab

How should we make decisions when AI is becoming part of every workflow — and who is accountable when it gets them wrong?In Episode 5 of The Lab, Justus, Max and Scott are joined by Dr Daniel van der Walt, Senior Lecturer in Civil and Environmental Engineering at the University of Canterbury. Daniel works at the intersection of enterprise risk management, infrastructure, engineering education and emerging technology.They unpack why human decisions are often emotional before they are rational, how uncertainty and bias should be handled, and when AI is useful — or dangerous — as a decision-making partner. The conversation moves through accountability in AI-assisted engineering, the future of universities and research, mentorship and organisational knowledge, decision registers, hiring for genuine competence, the enduring value of human effort, and whether AI will eliminate work or simply let us tackle harder problems.An expansive conversation about making better decisions in a world where the tools are changing faster than the institutions around them.

Luke, founder of Lodge, joins Max, Scott, and Justus to talk about leaving school at 15, building a company that wants to reinvent tax and accounting, and competing in a world where AI has made software easier than ever to create.They explore customer-led product development, viral marketing, software defensibility, hiring great people, and the lessons founders can take from competitive sport. The conversation eventually turns to rewriting the Linux desktop, open-source GPU drivers, and the slightly terrifying idea of riding in a vibe-coded self-driving car.

Joe Cole, Co-Founder and CEO of Tacit Intelligence, joins The Lab to discuss how the hidden judgment of human experts can be captured and used to train more capable, reliable AI systems.Drawing on Joe’s experience building technology for hundreds of professional sports teams and Olympic programmes, the conversation explores the surprising parallels between elite coaching, human learning and reinforcement learning.Joe explains why conventional datasets often miss the reasoning behind expert decisions, how Tacit evaluates AI for high-stakes environments, and why the future may belong to specialised models rather than one model that attempts to solve everything.The group also discusses sleep and performance, AI safety, digital twins, whether models could “dream,” the limitations of AI memory, and how businesses can encode their judgment into agents, skills and evaluation systems.

In episode 2 of The Lab, we talk through what it actually looks like to build software with AI agents in the loop.We cover the first MVP, why big codebases can make agents more useful, Scott’s prompting framework, enterprise-ready AI workflows, background agents, Cursor, Codex, merge conflicts, and the reality of letting agents run for hours at a time.A practical conversation about where AI coding tools are already working, where they still break down, and how we’re using them inside real product work.

In this first episode of The Lab, we talk about building real products with AI, what actually matters when hiring, the strange ways people trust confident AI systems, and why taste, persistence, and shipping still matter.