AGI - Advance, Grow, Innovate with AI

There are a handful of real data points and interesting observations - particularly on open-source vs. frontier model cost differentials and the self-reinforcing economics of AI coding tools - but these are buried under extended tangents about Elon Musk history, RoboCop jokes, and workforce development anecdotes that add no informational value for a B2B operator.

“the cost to run anthropic, forget even fable, just anthropic generally is like $25 per million tokens. Um, the cost to run some of the latest open source models is $0.30 per million token”

“if we can figure out how to use AI to write more code, we can then use AI to write better AI and we can accelerate our advancements in relation to other providers”

The framing around junior workers pairing with domain experts to extract tacit knowledge and the observation that open-source model economics will force a structural shift in AI product architecture are modestly interesting, but most of the episode recycles standard AI industry narratives - fear fundraising, approval ratings, 'use it and tinker' - without adding a genuinely fresh angle.

“the Frontier labs aren't 10,000 times better than the open source models. They might be a thousand times better, they might be a hundred times better, but they're not 10,000 times better”

“knowing the form of the skill is valuable and important, um, because it allows you to basically create the scaffolding for the AI to be able to operate effectively”

Oliver Belanger is the COO of a small, recently formed nonprofit AI consultancy and is essentially the host's coworker; he has a software engineering background and nonprofit sector exposure but has not operated at meaningful scale in any domain, making this closer to an internal team discussion than a practitioner interview.

“a guy who I met through this podcast, who's now my boss and my co worker and the chief operating officer of Citizen AI, Oliver Belanger”

“part of my background is software, uh, engineering and I got into the tech industry through a workforce development program”

The episode earns credit for citing concrete grant parameters (the $50M pool, 500k - $10M eligibility band, up to 10% of operating budget, July 15 deadline, October decisions) and the $25 vs. $0.30 per-million-token cost comparison, but the rest of the conversation - especially on AI lab dynamics and workforce trends - is speculative and unverified.

“they're specifically targeting small organizations, highly trusted small organizations in the uh, 500k to 10 million um, dollars dollars, uh, a year in operating budget”

“the cost to run anthropic...is like $25 per million tokens. Um, the cost to run some of the latest open source models is $0.30 per million token”

The conversation is almost entirely between two colleagues who agree with each other throughout; there is no meaningful pushback, no challenging follow-up, and the discussion drifts repeatedly from the stated topic into loosely related tangents, functioning more as a friendly chat than a structured, probe-driven interview.

“I don't disagree with you at all”

“I'm going to 100% agree with you and just rephrase it slightly”