Zeitgeist · July 10, 2026
The Pendulum Swings Again
Friday's conversation kept returning to the same tension: how fast sentiment moves, whether it's picking a favorite lab, calling a bubble, or reading a hire as either reassuring or alarming. Nothing on the list stayed simple for long.
Cost Per Intelligence Is the Only Scoreboard That Matters
- ▸The team pulled up DeepSWE, a new "vibes" benchmark from DataCurve that ranks models by how they actually feel to use relative to price, and it became the frame for the whole model discussion.
- ▸GPT-5.6's top tier came out near the top of that chart, while Claude Sonnet 5 landed as both expensive and comparatively weak by the same measure. That gap, more than raw benchmark scores, is what's driving the team's actual usage.
- ▸The practical takeaway: the team is seriously questioning whether a Claude subscription still makes sense once quota, price, and day-to-day task completion are weighed together, especially with a frontier model available in Codex that isn't available in Claude Code.
The Anthropic-to-OpenAI Pendulum
- ▸The team openly acknowledged how quickly its own allegiance swings between labs, from Anthropic loyalist to OpenAI convert in the span of a release cycle, and called the pattern out as somewhat absurd even while living it.
- ▸The comparison drawn was to esports: once a lab becomes the top player, it instantly picks up outsized criticism. Some of Anthropic's current wave of skepticism was read through that lens rather than as a verdict on the product itself.
Is This an AI Bubble, or Just a Bottleneck?
- ▸A Reddit post under 24 hours old, with roughly 5,000 upvotes, argued that sudden price spikes are an early sign the AI bubble is about to burst. The team pushed back hard on that read.
- ▸The counterargument: rising prices reflect high demand against constrained infrastructure and energy supply, not a lack of investor money or a ceiling on model capability. A real bubble would look like manufactured demand quietly evaporating, not usage climbing while everyone scrambles for compute.
- ▸Jevons paradox came up as the likely shape of things: cheaper, more efficient models tend to drive more total usage and spend, not less, with Coinbase's rising token consumption despite falling per-token costs cited as a live example.
Bernanke Joins Anthropic's Oversight Trust
- ▸The team dug into Ben Bernanke's appointment to Anthropic's AI Oversight Trust, noting he holds no equity or financial interest in the company.
- ▸The read was mostly about signaling: of recent Fed chairs, Bernanke carries unusually broad, bipartisan credibility, and his history leans toward propping up struggling institutions rather than letting them fail. Hiring him over someone like Alan Greenspan was seen as a deliberate, safer choice.
- ▸The larger point: this is Anthropic positioning itself to Washington as an institution built to last, not a startup lab winging it.
Talent Is Flowing, and Not Always Home
- ▸Mercor's reported move toward a $20 billion valuation, up from $10 billion roughly nine months earlier, prompted a quick multiple check. The team landed on something close to a 10x revenue multiple and called it a reasonable price for a company supplying synthetic training data to the labs.
- ▸A Nobel-winning U.S. chemist relocating to lead an AI institute in China drew more concern. One view: America's edge has long been its ability to attract top talent, and losing that trend would be a real warning sign.