One and half year ago on our 1:1 with @paulg , before @ChatGPTapp came out; PG said this: “Every company will start their product with @montereyai. If they don’t, they would lose the competition” Building a reliable ai product for high stake decision is hard. But we are making it happen.

🤩

The launches of @ChatGPT’s Intelligent UI and @AnthropicAI’s Motion and Dashboards show a pretty striking difference in product DNA. Probably also a glimpse into the org designs. - OpenAI is pushing toward a new interaction model. Why force users into predefined interfaces when AI can build one on the fly? - Anthropic is making existing workflows smarter. More immediately useful, but still largely operating within familiar software paradigms. Making the software itself adaptive is the bigger better bet.

Benchmarks are becoming a new growth channel. They used to be academic infrastructure or a few like @huggingface put those in core. Now every ai company realizes if you can own the benchmark, you can help define what the best means for an entire category. The playbook: - Build a benchmark around the capability your product wins at - Make it credible enough that others want to compete - Become the scoreboard the industry references - Turn every model launch into distribution for your benchmark, and your company Competing on benchmarks is becoming competing to own the benchmark.

The current AI personal assistants prioritize task efficiency as the primary goal for activities like shopping and traveling. What if having fun is the goal, aligning with what the general public seeks from these activities?

Miss the early days of builders sharing ai products + knowledge. Now it’s all endless metaphysical discussions of ai.

My life is defined by how many acquisitions I go through every yr at this point. 😆 Congrats team!

I like how engineers in euro do cigar/smoke break 😆

Second liquidation of angel investments 🫡 congrats team!

Everyone wants to build a painkiller. Sometimes being a really, really good vitamin is enough!

To people who are fascinated by @bendingspoons’s action on @airtable, this is a great piece covering Bending Spoon’s operation machinery, and founding principles.

Patiently waiting for custom voice mode so my codex pet's inner monologue has a matching sound @OpenAI 😁

Midjourney is absolutely the best building and acquiring communities. - AI Spa: A community in the health space, physically. - Co-Star: A community in the Astro/social space, psychologically.

Elizabeth is a remarkably calm and clear thinker. Absolutely love the episode.

The New "Change a Button" Test 💊 Something @ravi_mehta said at @MiroHQ Canvas stuck with me. For years, the classic test of product / company velocity was one question: "How long would it take you to change a button on your homepage?" It measured how fast a company can make a decision and ship it, and of course the infrastructure that lets it. Today I think the question has changed: "How long does it take your product to move to a new AI model?" It means: - knowing when a new model is genuinely better for your use case - catching regressions before your users do - updating the prompts and workflows built around the old model - rolling it out safely - building enough confidence for the team to trust the change What amazes me is watching companies at millions-of-users scale do this elegantly while new models ship every few days. How long does it take your company to switch?

Market is converging on a two-layer AI coding stack: - Orchestrator: understands the codebase, plans work, manages context, delegates tasks. - Execution engine: writes code, edits files, runs tests, and iterates. If Codex continues to widen the gap as the execution engine, the orchestrator becomes increasingly interchangeable. And we are back to the old principle: value follows active usage. Execution agents are in the critical path of every edit, every test, every debugging session, and every iteration. Orchestrators, by comparison, are consulted far less frequently. Once the work is decomposed, the majority of token usage, compute, and user time shifts to the execution layer. If that pattern holds, it changes the economics of AI coding.

Have a sudden hunch @amazon will buy @ouraring - Oura completes Amazon’s healthcare and wearable ecosystem w shared consumer personas - Buying beats building in the competitive predictive health AI race. - Oura data powers more personalized AI assistants / shopping experience

Announced investments / partnerships since the post: @EliLillyandCo x @nvidia @A24 x @GoogleDeepMind 😉