Host of What the AI?! podcast. Advisor and speaker helping leaders make sense of the AI moment. Previously: Upstart (founding team), Google, Stanford CS

Joined August 2007
384 Photos and videos
Two unprofitable frontier labs are cutting prices to win your business before their IPOs. If you buy AI at any scale, this is the rebid window. The best negotiation tactic is a good alternative — right now you have two, fighting over you, below cost.
19
The real AI business is on the ground: SpaceX rents its ~555,000 Colossus GPUs to Anthropic and Google — reportedly ~$920M/month from Google alone. The neutral compute landlord might be the realest AI bet inside the whole $2T.
1
100
Meanwhile both frontier labs are racing to go public while cutting prices to fight each other — making the profit question harder to answer right as the market gets to ask it. Great week to be a buyer. Stranger week to be a seller.
1
64
Full breakdown on this week's What the AI?! — plus Apple's Siri-on-Gemini and the Dario/Bezos/Musk fight over your job. Link in bio.
43
A teen told my co-host she now adds typos to her papers on purpose — so teachers won't think she used AI. We've made imperfection the proof of humanity. The better your writing, the more suspicious it looks.
1
52
Then GitHub Copilot went usage-based. Heavy users report 10–50x price jumps — $29→$750, $50→$3,000/mo. The power users driving the "gains" are now the expensive ones.
1
145
A year ago: "AI is 100 interns — fire the interns." Now some tasks cost $1,000 in tokens vs. a $250 junior analyst. Turns out the interns were cheaper.
1
66
The real question isn't whether AI coding works. It's whether "a seat for everyone" survives the usage-based invoice.
43
Same week: Anthropic filed to go public, and Alphabet sold ~$85B of stock to fund its AI buildout. When the company that prints cash has to raise cash to keep up, you've learned how big the bills got.
1
113
And Florida became the first state to sue OpenAI — naming Sam Altman personally, calling ChatGPT a "dangerous public nuisance." Same theory that took down Big Tobacco.
1
106
Full breakdown on this week's What the AI?! — link in bio.
62
We just recorded on AI ROI, and the line I can't shake is from Meta's CTO: "All motion is not progress, and token usage alone is not a measure of impact." Half the AI strategies I see are measuring motion and calling it momentum.
1
1
45
Every morning at 6 a.m. my Mac mini prints out my day. A small piece of software I wrote (with Claude) pulls my calendar and to-do list, lays them out on half a sheet of paper, and sends it to the printer. It's waiting for me when I sit down. Paper. No screen. Pen.
2
107
Of course, none of this is entirely new. People have been writing little automations for themselves for decades. What's new is who gets to do it. Now it's 10 minutes for anyone with a Claude Code or Codex subscription. The set of people who can build this just got much bigger. The skill that matters isn't syntax. It's taste — knowing what's worth automating in the first place. We're about to live in a world full of small, custom, slightly weird software that exists only for the person who wanted it. Single-player software.
1
1
38
Tomorrow at 6 a.m. my printout will be waiting for me — small and weird and exactly mine. What will you build?
21
You can measure lines of code. You can't easily measure whether those lines made the product faster or better for customers. Uber's COO just said the quiet part out loud.
1
77
This isn't "AI doesn't work." It's that a sophisticated buyer still can't measure what it's getting. Both can be true.
1
35
So the open question: is the missing ROI a measurement problem we'll solve — or is the spend already outrunning the value?
14