Front pageThe desks
All coverage Reporting and analysis on the agent economy and frontier AI.
Agent economy / Friday edition
Forter puts the average agent order it saw in early November at $204. To get from one basket to a market size, you need a second number. Here’s what three widely cited reports actually count.
Nov 18, 2026 · 2 min read Frontier AI / Publishing
ChatGPT will link to the original articles. Across the publishers Chartbeat tracks, referrals from AI services accounted for roughly one pageview in 10,000.
Nov 18, 2026 · 2 min read Agent economy / Workplace productivity
Microsoft’s new dashboard starts with how long a task usually takes. A study of 5,172 support workers shows another way to measure what an AI assistant contributes.
Nov 18, 2026 · 2 min read Agent economy / Voice agents
The voice platform disclosed traffic and a thinner companywide margin. Buyers still need a way to price the work a call actually finishes.
Nov 17, 2026 · 1 min read Agent economy / Payments
Forter detects fraud for merchants. Its order data shows how quickly the mix can change when it learns to recognize another shopping agent.
Nov 17, 2026 · 1 min read Agent economy / Commerce
OpenAI gave ChatGPT a wallet. In Visa’s test of 22 models, more than half of failed shopping runs still ended with a decision to buy.
Nov 17, 2026 · 2 min read Frontier AI / Evaluations
Epoch AI asked six models to do its own research. Mapping a data center was the assignment where two could compete with a human contractor.
Nov 17, 2026 · 1 min read Frontier AI / Model economics
Astra’s token prices fell 20%. The bigger swing in the release data comes from how many tokens a job takes.
Nov 17, 2026 · 1 min read Agent economy / Retail
Its fashion agent can watch for a price drop in the background. Only participating retailers can take the final order through the agent.
Nov 17, 2026 · 1 min read Frontier AI / Evaluations
METR asked agents to inspect 55 test environments. One model could consistently tell which ones were designed to examine its behavior.
Nov 17, 2026 · 1 min read Frontier AI / Inference
Google’s two-model system finished a coding benchmark faster while using less quota. The savings depend on which steps can change hands.
Nov 17, 2026 · 1 min read