Notes
Short pieces about the methodology and architecture decisions behind the AI systems I ship — specs, evals, multi-agent orchestration, LLM integration, and the discipline of directing coding agents.
June 8, 2026
Apple rented its brain
At his farewell keynote, Tim Cook showed a rebuilt Siri — running on a custom 1.2-trillion-parameter Google Gemini model Apple pays about a billion dollars a year to use. Sit with that. The company whose entire identity is owning every layer of its stack just decided the AI model is the one piece not worth building. That's the most credible verdict you'll ever get that the model is a commodity — and a clean lesson in what's actually worth owning.
- business
- ai-native
- architecture
June 8, 2026
AI is brilliant at ideas and bad at being right
We worried AI would automate the boring work and leave humans the creative heights. The research from 2026 says we had it backwards. When AI agents were set loose on real research, they generated novel, clearly-written ideas — and then fabricated or invalidated their own experimental results in about 80% of cases. AI turns out to be a fantastic source of ideas and a terrible judge of whether they're true. Once you see that split, how you should use it becomes obvious — and so does the mistake almost everyone is making.
- methodology
- eval
June 8, 2026
When the government wants a piece of your AI lab
This week, US officials and OpenAI revisited a remarkable idea: the federal government taking an equity stake in the company. A senator went further, proposing a 50% government share in leading AI labs. Strip away the politics and a quieter shift is happening — AI is sliding from 'product' to 'national infrastructure,' something states treat like oil or the power grid and want to own. If the foundation you build on is becoming a strategic asset that governments fight over, 'it's just an API' is no longer a safe way to think about it.
- business
June 8, 2026
Route by difficulty, not by default
When Apple rebuilt Siri, it didn't pick one model and send everything to it. A timer request stays on your phone. A medium query goes to Apple's private servers. Only the hardest reasoning reaches Google's giant model. That three-tier split isn't an Apple quirk — it's the pattern every serious AI product is converging on, because sending every request to one big model overpays on the easy ones and over-exposes the sensitive ones. The fix is routing, and most builders skip it.
- architecture
- ai-native
June 8, 2026
The cloud has a smokestack
We call it 'the cloud,' which makes AI sound weightless. It isn't. Every prompt runs through enormous buildings that burn real electricity and evaporate real water — and the buildout is now big enough to strain power grids, raise people's electric bills, and trigger more than 300 state bills in a single year. AI is quietly one of the most physical industries on earth, and that physical limit — not algorithms — is becoming the thing that decides how far it can go. Worth thinking about, even from a keyboard.
- business
June 8, 2026
The labs are going public, and the public isn't sold
Anthropic just filed to go public at a valuation near a trillion dollars, with OpenAI close behind. At the same time, 57% of Americans say AI's risks outweigh its benefits — while using it more every month. Two things worth thinking about: what happens to a company's 'safety-first' principles once a stock price depends on relentless growth, and what it means that the foundation you build on is now run for shareholders who priced in a future that hasn't arrived. This isn't market commentary. It's about the ground shifting under everyone building on these models.
- business