$1,000, two weeks, one Fed chairman
What stopped me this week wasn't a lab demo, it was a small asset manager's side project. Squeezed by Fed Chairman Kevin Warsh's new taste for silence, F/m Investments built a chatbot that parses roughly 1,800 of his documents and transcripts: WarshGPT. Cost? Under $1,000. Timeline? Two weeks from idea to release. Model underneath? Anthropic's Claude — with a name that riffs on OpenAI's ChatGPT.[1]
I'll confess: the number that actually got me wasn't the $1,000, it's that the team treated this as a reflex, not a product launch. As CEO Alexander Morris put it, they'd 'made a pretty good business out of decoding Fedspeak' — so when Warsh went quiet, they wrote their own translator in two weeks. June's Fed statement shrinking to 130 words (down from a prior average north of 300), the share of policy sentences at the press conference falling from 27% to 5% per UBS: that's why F/m moved fast. The model layer is commoditizing, and value is moving up — this time, to a hedge-fund desk.[1]
Kimi K3 is the same coin's other face
The same week, Moonshot's open-weight Kimi K3 proved the same thesis from another angle: Moonshot admitted its model 'still trails' Claude Fable 5 and GPT 5.6 Sol, yet independent tests from Arena.ai and Vals AI found it competitive with flagship systems. Even OpenAI's Dean Ball called Kimi 'a very good model.' I see opportunity here, not threat: when a near-frontier model is open-weight and free, projects like WarshGPT multiply everywhere.[2]
Last week I wrote about a 975-billion-parameter model's oddly humble line in its own announcement; this week I watched that humility turn into a business model. My guess: pricing pressure on Anthropic and OpenAI will intensify for precisely this reason — thousand-dollar projects like WarshGPT and free open-weight models are narrowing the space around expensive APIs. What I'll be watching next week isn't model announcements — it's how many more 'two-week agents' show up.[1], [2], [3]