Writing
Notes written here, and Middle Layer, the newsletter I have been writing since March 2025 on compute, distribution, and the boundary between AI platforms and products.
Notes
Middle Layer · 10 essays since March 2025
Token Maxxing ↗
A practical answer to AI anxiety: use the tools more, on real problems, in your own domain.
Things Aren't Changing That Fast ↗
A counterweight to AI discourse: frontier builders may be moving fast, but most daily workflows still change slowly.
Where Consumer UX Meets Enterprise Value ↗
A note on B2B SaaS, consumer-grade interfaces, and why enterprise software is being pulled toward better product taste.
Carving a Niche in the Cloud: The Modal Approach ↗
Why some of the next great AI companies may win by making the hard infrastructure parts disappear.
I: Claude Is the Drug, Cursor Is the Dealer ↗
Coding tools as one front in AI's supply-chain wars, where wrappers, labs, costs, and distribution collide.
No Status to Be Had ↗
Why ChatGPT feels unusual in a status-soaked internet: it gives everyone the same blank interface.
Prompting on Hard Mode: The Prompt-to-Prompt Trick ↗
A simple way to get better outputs by asking the model to write the recipe before generating the answer.
The Strange Case of Random Rewards: AI's Weird Learning Process ↗
What a random-reward experiment with Qwen says about hidden capabilities, reinforcement learning, and how messy AI research still is.
Claude 4 Isn't Just Smarter. It's Strategic ↗
Anthropic's Claude 4 release as a signal about compute, distribution, safety, and the strategic layer of the AGI race.
CoreWeave: The AI Cloud Bet with Hidden Risks ↗
A breakdown of CoreWeave's GPU-cloud business, the AI compute gold rush, and the risks in being a specialist middle layer.