Tech Overflow

Season Two Wrap

A season finale should feel like a recap, but ours turns into a snapshot of how fast tech is reshaping real work and daily life. Hugh’s back on the ground in Los Angeles in a new Paramount role, taking robotaxi rides like it’s normal, while Hannah steps into a Product Director job at Ocado tackling last mile logistics and the delivery experience. We talk honestly about what we’ve learned after 18 months of making Tech Overflow and why the “curious minds” approach works best when we keep it practical. Waymo becomes our unexpected lens on product design, autonomy, and human behaviour. What happens when there’s no driver and no social friction? Why do other drivers treat a self-driving car differently? And what does “polite” software feel like as a passenger when the rest of the city learns it can always cut in? Then we go deep on AI agents, vibe coding, and the gap between “LLMs make building easy” and actually shipping something useful. Hannah tries to build an agent to book Pilates classes and discovers that the hard part is not motivation, it’s foundations: terminals, tooling, and knowing how to break the problem into steps. From there we unpack AI at work, including token usage as a blunt adoption metric, the meaning of tokens, and why most organisations are still learning how to use AI as a co-pilot rather than an autopilot. Listener Q&A covers LLM tiers like Claude Haiku, Sonnet and Opus, local models you can run on your own machine, plus data privacy, enterprise terms, and API retention. We also answer a classic question clearly: how contactless payments and Apple Pay work, end to end. Subscribe for season three, share this with a friend who’s trying to “use AI properly”, and leave a review if Tech Overflow helped you make sense of modern technology. What should we build or explain next? Like, Subscribe, and Follow the Tech Overflow Podcast by visiting this link: https://linktr.ee/Techoverflowpodcast [https://linktr.ee/Techoverflowpodcast]

Venture Investing With The VC Who Invested in Insta and Figma (with John Lilly)

Entrepreneurs get the glory, investors get the spreadsheets, and John Lilly says that’s exactly how it should be. John (former VC, now angel and board member, and newly involved with Gigascale Capital) joins us to demystify venture capital for curious builders: how funds are raised from limited partners, why returns follow a power law, and what investors actually do between writing a cheque and a company becoming real. Along the way, we unpack his simple working framework: see, win, decide, then help build.  The best moments are the stories. John explains how a small real-world signal helped him chase Instagram, how “winning the right to invest” can mean recruiting a key hire, and why a fast acquisition can still feel like a mixed outcome when you believe the company could be worth far more. He also shares the long arc of Figma, including an early “no”, a year of breakfasts, and the traits he looks for in founders: grit, follow-through, and the ability to learn without defensiveness.  Then we widen the lens to the present shockwave: AI and large language models. John talks about prototyping at speed, what it means for startup formation, and why the next constraints may be compute, chips, cooling, and power rather than ideas. Finally, we touch climate tech and hard tech investing, where energy, materials, supply chains, and data centre demand collide.  Like, Subscribe, and Follow the Tech Overflow Podcast by visiting this link: https://linktr.ee/Techoverflowpodcast [https://linktr.ee/Techoverflowpodcast]

Microsoft’s Rogue AI — What We Learned from Tay (with Derrick Connell)

This week Hannah is joined by guest host Derrick Connell to discuss how Microsoft's Tay went wrong, how Satya Nadella reacted in the moment, and what Derrick learned about innovation. Derrick also shares stories from shifts in technology and discusses his new book Twenty One Summers. Derrick shipped a chatbot that survived for 18 hours as it went horribly wrong. It sounds like a punchline until you realise it helped rewrite how the industry thinks about AI safety. Hannah sits down with Derrick, a former Corporate Vice President who spent nearly three decades at Microsoft and led teams across Search and AI, to unpack what innovation looks like when you are shipping into the real world and the real world fights back. We talk through the massive platform shifts Derrick lived through, from Windows and Office shipped on discs to cloud services that ship daily, plus what it took to build Bing while Google held the vast majority of the search market. Along the way we get practical about product development methods, why agile experimentation changed the pace of software, and how “scrappy” teams innovate when they are not expected to win. Then we go deep on conversational AI. Derrick explains why China’s WeChat environment made early chatbots thrive, how training data and user behaviour shaped outcomes, and why the US launch of Tay on Twitter was vulnerable to bot attacks and manipulation. We also get into the leadership playbook after a public failure, the importance of asking “What did we learn?”, and how that moment pushed Microsoft to publish early AI ethics guidelines that influenced responsible AI practices across the industry. If you care about AI product management, innovation leadership, chatbot design, LLM guardrails, and what it takes to build safer AI systems, this conversation will give you both a story and a framework. Subscribe, share with a friend who builds AI, and leave us a review so more curious minds can find Tech Overflow. Like, Subscribe, and Follow the Tech Overflow Podcast by visiting this link: https://linktr.ee/Techoverflowpodcast [https://linktr.ee/Techoverflowpodcast]

You’re Not Searching the Web (How Google Search Really Works)

Google Search feels like magic because it is solving an impossible problem on your behalf: you show up with a complex information need, type a couple of words, and expect a great answer almost instantly. We unpack what’s really happening in that split second, from the early days of cluttered 90s search engines to why Google’s clean interface, speed, and relevance changed everything. We walk through the core machinery that makes web search work: crawling (and why it has to be “polite” to websites), building and refreshing a copy of the web, and using an inverted index so results can appear in around 200 milliseconds. From there we get into ranking, including PageRank and why links became a proxy for credibility, plus the constant battle against spam and how SEO sits in a tricky grey zone between good practice and gaming the system. Then we zoom out to what’s changing now. Human judges still evaluate search results at industrial scale to improve quality and train machine learning systems, while query understanding rewrites and repairs your input so the ranker has a fighting chance. Finally, we tackle the shift away from the ten blue links era as AI summaries and LLMs like ChatGPT reduce clicks, introduce new trust issues, and force new monetisation choices that could reshape search again. If you enjoyed this deep dive, subscribe wherever you get your podcasts, share the episode with a curious friend, and leave us a review so more people can find the show. Like, Subscribe, and Follow the Tech Overflow Podcast by visiting this link: https://linktr.ee/Techoverflowpodcast [https://linktr.ee/Techoverflowpodcast]

AI Is Already Better Than You Think with Ramez Naam

AI did not creep in quietly, it arrived like a tidal wave. We talk with Ramez Naam, computer scientist, science fiction author, futurist, and climate tech investor, to pin down what today’s large language models really are, why they’re the fastest adopted general technology in history, and why “impressive” is not the same thing as artificial general intelligence. Along the way, we challenge the idea that AGI is right around the corner, even as these tools already outperform any single human on breadth of knowledge and rapid synthesis.  We get practical about capability and risk: where ChatGPT, Claude, and Gemini shine, where they still fail, and why supervision and verification are the new baseline skill for knowledge workers. We also unpack why the AI model race keeps flipping leaders, why user data may not create the kind of network effects people assume, and what recursive self-improvement would need to be real rather than wishful thinking.  Then we go straight to the biggest near-term shock: coding with AI. Vibe coding and modern developer tools are collapsing the distance between an idea and a working app, which raises hard questions about software engineering careers, junior hiring, and what “good” looks like when you are managing an army of bots. Finally, we zoom out to the energy and infrastructure behind AI, from data centres and grid bottlenecks to the case for solar-and-battery powered compute, including why Australia could be well placed.  If you’re curious about the future of AI, AI jobs, AI reliability, data centres, and what the next ten years might realistically hold, this conversation will give you a grounded framework. Subscribe, share the episode with someone who debates AI with you, and leave a review with your most surprising takeaway. Like, Subscribe, and Follow the Tech Overflow Podcast by visiting this link: https://linktr.ee/Techoverflowpodcast [https://linktr.ee/Techoverflowpodcast]