The edition

Sunday August 30, 2026

This is one edition of my daily email.

One free email each morning, around 8 AM US Eastern.

In this edition

Day 7Anthropic's revenue surge and IPO pushAnthropic Plans October IPO at Potential $2 Trillion Valuation
Day 5NVIDIA AVO and Prime Agent on ARC-AGI-3Agent harnesses near-solve ARC-AGI-3; WeaveBench shows 41.2% long-horizon gap

What moved

Day 7

Anthropic's revenue surge and IPO push

Ramp's business spending data, published August 29, shows Anthropic holding roughly 60% or more of business AI API spending versus OpenAI's 35%, with coding identified as the most contested area.
The same data put Anthropic's adjusted gross margin at approximately 40% in 2025, projected to reach 44% in 2026, against OpenAI's 33%.
The figures add to the financial case Anthropic is preparing to present to investors in a late September or early October IPO.

The gist

The planned offering, if completed at the bankers' cited figures, would surpass SpaceX's $85.7 billion raise, described in reporting as the largest IPO ever. The revenue run rate and market share data also put competitive positioning between Anthropic and OpenAI on record.

Continue this story →4 sources

Day 5

NVIDIA AVO and Prime Agent on ARC-AGI-3

WeaveBench, a 114-task benchmark of hybrid GUI-CLI long-horizon work, emerged August 30 showing a best pass rate of 41.2%, against the near-complete ARC-AGI-3 scores NVIDIA AVO and Prime Agent posted days earlier.
LongHorizon-Harness, described alongside it, frames the gap as a task-state problem, where task history becomes unreliable about what is finished, failed, or remaining, and argues that reliability over long task spans depends as much on the harness as on the model.

The gist

Harness design separates the best from the worst results more than model capability alone, as shown by the gap between near-perfect ARC-AGI-3 scores and the 41.2% ceiling on WeaveBench.

Continue this story →2 sources

Get the daily email

I send one email each morning with the stories that moved. If you would rather just read here, that works too.