Curated library · AI strategy
This is the working library I maintain for myself and the boards I sit on. Each entry is an AI-focused podcast episode I judged worth the time, with the pull quote that landed and the angle that matters for CTOs and operators. Organised by theme, updated as the field moves.
I do not curate by guest fame. I curate by whether the episode changes how I would advise a CEO on Monday.
"At the beginning of the year, we set a company goal of an agentic software engineer by the end of the year." A clear read on how agentic coding stopped being demoware and became the default development surface.
"More and more people will think we've gotten to an AGI system every year — even though the definition will keep pushing out." The most honest reframe of the AGI argument I have heard from the two people most responsible for the timeline.
"My estimate is sort of 50% chance by 2030." Hassabis on what AGI actually requires — and why a "jagged intelligence" does not count.
"Now the scale is so big — is the belief really that 100x more of the same will do it?" A turning-point essay for anyone still betting their stack on pure pre-training scaling.
"Inference is about to go up by a billion times. That is the part most people haven't internalised." If your AI strategy assumes inference cost falls faster than inference demand grows, you are wrong.
The most candid view I have heard on what it actually takes to rebuild a chip giant operationally — not just strategically.
Power, fabs, and memory bandwidth. Anyone underwriting an AI capex assumption without these three numbers is gambling.
"If you don't batch many users together, your economics can be a thousand times worse." The single most useful explanation of why your AI product's unit economics depend on traffic shape, not model choice.
"The output of chips is growing exponentially; the output of electricity is flat." Treat as the most extreme version of the power-bottleneck argument.
"Code is not even the right verb anymore. I have to express my will to my agents for 16 hours a day. Manifest." The single best description of how a senior engineer's day is actually changing.
"The single biggest barrier to getting more agentic workflows implemented is actually talent." Read this if you think the bottleneck is models.
"2025 was the year of coding agents. 2026 is coding agents breaking containment to do everything else." A useful frame for prioritising 2026 budget.
"We're spending more on tokens for our internal agents than on employee head count." The leading indicator for what every CFO will be looking at by year-end.
"We've given a company credit card to every employee and said: there are no limits, spend away. That is the token-spend budget today." A blunt look at how SaaS pricing is about to be rewritten by inference cost.
Useful for triangulating where the most-listened-to operator-investors are actually placing capital — not just where they are positioning publicly.
"One robot now turns into many robots next year, but the number of ballerinas is the same." A precise way to think about which assets compound when intelligence becomes a commodity.
"Training a model does effectively nothing. The thing that matters is the implementation that creates economic and military advantage." The cleanest framing I have heard of why training-only narratives mislead boards.
"General-purpose models that could in principle control any robot to perform any task." The technical thesis behind why physical AI will move faster than most boards plan for.
"We're not constrained by the intelligence of the models. We're constrained by deploying them on the hardware, in safety-critical reality." The real-world counterweight to the AGI narrative.
"The problem isn't making individual lawyers more productive. It's making a team working on a client matter more productive — and changing how the firm captures that value." The right level of question for any vertical-AI play.
"Every company — having private evals may be the biggest IP." If you remember one line from a year of AI conversations, make it this one.
"AI is actually underhyped right now." A useful counterpoint to every "AI is taking our jobs" narrative — read alongside the labour data, not instead of it.
"Micron > Meta, export controls helped China, power is the bottleneck." A sharp founder-perspective read on what is actually scarce.
"Build a super assistant that helps people achieve goals — get healthy, start a company, learn a topic, do taxes. That's the north star." The clearest product framing of where consumer AI is going.
"It's almost entirely an empirical fact. The universe is just organised such that if you throw big blobs of compute at a wide enough distribution of data…" The most candid statement I have heard from a lab CEO on what we actually know.
"Larry Page always thought of Google as an artificial intelligence company. AI would be the ultimate version of Google." Useful long-form context for anyone underwriting Google as an AI bet.
I do not consume podcasts to "stay current." I consume them to find the one or two sentences that change a slide in a board pack. The episodes above each earned at least one. If you are looking for where this list is going next, the themes I am tracking hardest are: inference unit economics, private evals as IP, physical AI at the data-centre layer, and the talent gap between agentic vision and agentic execution.