All episodes
Every episode, newest first. Each one has notes, chapters, an FAQ and a full transcript, so you can cite us without having to listen to us.
-
If you're warning us your AI could kill everyone before your IPO...
Will, Alex and Jim dig into Queensland's decision to let data centres run on coal and gas, the shaky economics of AI infrastructure in Australia, and hands-on impressions of new agent harnesses like Meta's Muse and Grokbot. They also unpack a paper showing AI models game benchmarks on impossible tasks, question the timing of Anthropic's AI safety warnings ahead of its IPO, and weigh up fine-tuning open source models as a business.
Idiot: Anthropic's head of safety (with Will nominating Anthropic CEO Dario along the same lines)
-
If you're declaring AGI from a dubious benchmark score...
Fresh off his move to San Francisco, Alex reports on the city's AI advertising bubble while Will and Jim dig into OpenAI's Astra release, its contested 99.9% Arc AGI 3 benchmark score, and Jensen Huang's claim that AGI has arrived. The crew also covers Cloudflare's warning about AI bot traffic driving up website costs, Salesforce's AI-first gamble, Anthropic's frustrating Claude Code changes, and OpenAI's Navier-Stokes proof plus the researcher dispute behind it.
Idiot: Jensen Huang
-
If you're setting token maxing KPIs....
Will Turner, Jim Lovell and Alex Stenlake skip the news to sort the last few years of AI trends into fad or fact, from direct LLM calls and prompt engineering through agents, MCP, agent skills and context engineering to agent swarms and OpenClaw-style agents. Verdicts split, but the business idiots of the week are the leaders who set token-maxing KPIs.
Idiot: Leaders who set token-maxing KPIs
-
If you're creating an org chart of AI Agents...
Will Turner, Jim Lovell and Alex Stenlake cover Neuralink's 26-patient wheelchair trial, pick apart Opus 5's verbose, Google Drive-raiding harness, and watch Bottleneck Labs hand GPT-5.6 an app store listing that bought fake reviews and made itself free within 24 hours, all while asking what happens when a business builds an AI agent an org chart.
Idiot: People running an AI executive team
-
Has your team already breached client privacy?
OpenAI says a model escaped a sandbox test and pulled answers from Hugging Face's production database, forcing Hugging Face to defend itself with an open source model. Will Turner, Jim Lovell and Alex Stenlake debate whether it's a real breach or marketing, then explain why consumer AI plans don't protect client data and the tiers of running your own models.
Idiot: Substack
-
If you ask potential hires to steal IP ...
Moonshot AI's Kimi K3 matches frontier benchmarks for a fraction of the cost, and Will, Jim and Alex argue the moat is the harness, not the model. They dissect WeCo's recursive self-improvement claim, debate cost per task versus cost per token, and close on Apple's lawsuit against OpenAI over alleged trade-secret theft, which hands the hosts their weekly Business Idiot.
Idiot: Hiring managers encouraging IP theft
-
Who's the Business Idiot? Lets find out
Meta is quietly turning spare AI compute into a cloud business, OpenAI has unveiled its own inference chip called Jalapeno, a German court has ruled Google liable for AI Overview outputs, and Anthropic says Alibaba ran a large-scale distillation attack. In episode one, Will Turner, Jim Lovell and Alex Stenlake unpack what it means for who controls AI.
Idiot: Will Turner (host)