This Day in AI Podcast

There are genuine observations about model behaviour in agentic workflows - thinking token economics, Grok's uncapped output bug, Anthropic's 4.7 regression - but they're buried in substantial filler: a diss track, a closing song, the prolonged Moshi gimmick, and repeated off-topic banter. The insight-to-noise ratio is poor for a 47-minute runtime.

“It also burns a lot more thinking tokens, which is the other problem. I'll find that Opus, I'll have like four tabs open. All of a sudden they all feel like they're stalled and you realize it's just epic amounts of thinking tokens which count as output tokens.”

“there is no output token limit. So it's like one of these models where it will just keep outputting tokens uh, if you don't sort of find a way to stop it from doing that.”

A handful of mildly interesting observations - Anthropic's first-ever regression with 4.7, the structural unsustainability of fixed-price subscriptions over variable token costs - but the broader takes (agentic is the future, prices will eventually fall, orchestration layers add value) are widely recycled in AI discourse. No contrarian or first-principles arguments are developed with any rigour.

“it just is unsustainable in the long run to charge fixed price subscription for something with a variable underlying cost, like tokens”

“I think the telling sign is just that the constant need for the major providers like Anthropic and OpenAI to build their own application level stuff”

There are no external guests. The two regular hosts are indie developers building their own AI productivity tool (SIM Theory) at apparent small scale, which gives them some practitioner credibility but not at the operator level a B2B audience needs. The 'new co-host' Moshi is an AI voice model, adding zero expert signal.

“I think that in that layer, uh, in terms of AI productivity being productive, like being, helping the user be more productive on top of the models work asynchronously, work agentically, is that value add”

“I do have an exclusive of what that phone might look like. Um, and for those listening, this is actually the Facebook phone, if you remember that flop”

Pricing figures for Grok 4.3 and GPT 5.4 mini are cited, and the Anthropic model deprecation timeline (18 days vs Google's year-plus notice) is a useful concrete data point. However, all performance evidence is purely anecdotal from personal workflows, and no benchmarks, revenue figures, or third-party data are introduced.

“1 million connects window multimodal pricing is A$25 per million inputs, 2.50 on the output tokens. H tokens are 20, 20 cents per million”

“the thing about a GPT 5.4 mini. Like it is just so much cheaper than like uh, 75 cents per million input uh for GPT 5.4 mini”

The episode is a casual co-host chat rather than a structured interview; there are no external guests to push or challenge. The hosts occasionally prod each other but rarely follow through, and the Moshi AI-co-host segments consume airtime with gimmick rather than substance. The hosts themselves acknowledge it is their worst episode and that they forgot to record the first half.

“Oh, just tell us the truth. Do you think it'll fail or not?”

“Um, I forget what we misled. Everyone did mislead.”