AI, context, and New Zealand’s data center future
Full Show Notes:
https://www.theintelligenceagepodcast.com/843
Mark Smith speaks with Oliver Hartwich about how AI has changed since January, why agentic tools have become the real shift, and what that means for workflows, context, and cross-model collaboration. They also dig into New Zealand’s position in the global AI and infrastructure landscape, especially around data centers, energy, solar, batteries, and water use.
Key topics
- In this episode: Oliver says the biggest change in 2026 so far has not just been better models, but the rise of agentic AI tools like Claude Code, ChatGPT work modes, and Perplexity Computer.
- He explains why AI now feels less like a novelty and more like a practical, increasingly capable working layer, especially when multiple models are used together and checked against each other.
- Mark raises the concern that New Zealand may be vulnerable because AI compute and data may not stay in-country across the major hyperscalers operating locally.
- Oliver discusses the post Fable moment, when access uncertainty highlighted how dependent users have become on frontier tools and how quickly expectations shift across providers.
- The conversation turns to context management across platforms, including when to preserve long-term memory and when to use a fresh AI for independent review.
- Oliver describes building a personal Claude Skill from his own writing archive, compressing books, reports, articles, and newsletters into markdown so the system can reflect his style and past thinking.
- They discuss using one AI to peer review another AI’s work, including ping ponging a skill between Claude and ChatGPT to improve quality and completeness.
- Oliver shares how Codex can be used for computer control and debugging by going directly into system settings and config files instead of relying on manual UI hunting.
- The discussion moves to MCP-style integrations, including using Site CITE AI to connect large academic literature databases into AI workflows for faster research and self peer review.
- Mark and Oliver compare the old PhD research process with the newer AI-assisted version, where initial literature collation can be compressed from roughly a year to a few weeks.
- They reflect on the tradeoff between speed and thinking time, arguing that the slower, manual research process also created space for reflection, sleep, and deeper synthesis.
- The final section focuses on New Zealand data centers, geothermal energy, grid stability, solar adoption, battery storage, and the challenge of building infrastructure with social license.
Resources:
Oliver Hartwich's essay on AI in education: https://oliverhartwich.com/2026/06/25/bildung-and-the-machine/
If you want to get in touch with me, you can message me here on Linkedin.
Thanks for listening 🚀 - Mark Smith
