This week's data: Payrolls +29K Wage growth 3.0%, slowest since 2021 ISM prices paid 77.9, highest since March Core PCE 3.0% 10-year 5.28%, up on the day Growth slowing. Prices rising. Bonds not helping. The Fed has one tool and two problems.

Sources: BLS Employment Situation, Oct 2 bls.gov/news.release/empsit.… ISM Manufacturing PMI Sept 2026: PMI 54.5, prices paid 77.9 (investinglive) investinglive.com/news/us-se…

Average hourly earnings: +3.0% year over year. Slowest since 2021. CPI: 3.7%. Real wages have been negative since April, and the gap just widened. Workers are losing purchasing power while the Fed raises rates to fight prices they didn't cause. This is the stagflation trade.

Payrolls: +29K in September vs 85K expected. July revised to -10K. Three-month average: 51K. Unemployment 4.2%. Health care added 17K. Everything else, roughly zero. The Fed hiked two weeks ago into a labor market that is barely adding jobs.

GREG BROCKMAN: SOME PRODUCTS SCALE WITH THE MODEL. MOST DON'T. Greg Brockman, who leads product strategy at OpenAI, recently drew a line that separates durable AI products from temporary ones, and it is a useful filter for anyone evaluating this category. According to Brockman, some activities scale well as the models get smarter. Others exist to work around limitations in the model. The second category is very hard to make durable. There have been startups at various points built almost entirely on that second thing. Take the model, add some prompts, add handcrafted scaffolding around its weak points. That can be genuinely valuable at a given moment. It just does not survive the next capability jump, because the thing it was compensating for stops being a problem. The more interesting observation is about interfaces. According to Brockman, each level of model capability suggests a new interface almost on its own. He traces the sequence. In 2022, models became coherent enough that reading the responses was worth your time, and that is what made a chat interface the right product. At the end of 2025, the coding agent era began, when AI shifted from roughly 20% of software production to something closer to 80%. That shift demanded a very different product, and the product that captured it became the one everyone used. Now, according to Brockman, we are entering the era of the personalized agent. A cloud-based AI, being built into a number of different form factors. The underlying question he keeps returning to is the one worth carrying. Not whether a product is useful today, but where it will scale as the models get smarter. @OpenAI @gdb @siliconvalleymm

OpenAI can build broad intelligence, but every industry still has huge pockets of inefficiency. That is where domain expertise matters. The biggest AI opportunities may come from people who understand one workflow deeply enough to redesign it from the ground up.

His app was so ignored at first that he emailed users personally and hand-recruited people in Apple Stores. It became Pinterest. His name is Ben Silbermann. 🧵

He hand-recruited his first users one by one until Pinterest finally caught fire. That's the kind of builder worth celebrating.