AI Bubble Pops, Zuck Freezes Hiring, Newsom’s 2028 Surge, Russia/Ukraine Endgame

The MIT study cited found 95% of generative AI pilots never make it to production, with 70% of budgets funneled into sales and marketing tools that show poor ROI. Chamath attributes this to board-level AI FOMO trickling down as unfocused experimentation, and to a fundamental mismatch between probabilistic AI and the messy, hard-to-codify nature of sales and marketing workflows. Back-office processes with clear rules and edge cases are proving far better initial targets.

Sachs and Friedberg both argue the superintelligence/AGI-in-2-years story has been debunked by incremental model improvements and performance clustering across vendors. They highlight the rise of SLMs and networks of specialized models tuned to specific tasks or industries, which can deliver higher accuracy and drastically lower costs per token. Vertical AI vendors and domain-specific copilots (e.g., tax, CPAs, back office) are achieving measurably higher success rates than generic LLM overlays.

Friedberg emphasizes that generative models don’t autonomously run businesses; they augment humans. Code still must be debugged and integrated, workflows must be supervised, and deterministic systems (like game engines or traditional software) often need to be coupled with generative models. In video and film, for example, using AI to generate assets and then rendering in Unity-like engines allows control over continuity, lighting, and framing that pure ‘generate a movie’ prompts cannot yet provide.

Sachs calls recent AI stock pullbacks and the mixed GPT-5 reception a “healthy correction,” not a bust. He still sees an investment supercycle but warns against fantasies of rapid takeoff and recursive self-improvement. Chamath raises the risk of sunk-cost lock-in: model giants are pouring tens of billions into LLM-centric architectures and data centers just as alternative representations, custom silicon, or radically different approaches might emerge from small teams, potentially stranding incumbent investments.

The hosts describe Meta offering $100M comp packages, billion-dollar acquisition offers for pre-product AI startups, and $30B rumored valuations being turned down. Sachs notes these rare offers only appear when mega-caps feel strategically vulnerable, and warns many founders have never lived through a bust. Once strategic urgency fades, valuations will need to be justified by fundamentals—billions in real revenue—not by being an acceleration asset for a desperate Big Tech buyer.