Observed arrival · 2026-11-21
The Agentic: a dated record of AI-agent findings
A daily, weekly, and monthly digest of AI-agent research, evaluations, product claims, and security findings.
- For
- AI-agent researchers and model evaluators
- Worth noticing
- One entry compares 22 coding harnesses across five open models and says the cheapest and most accurate choices vary by model.
Field notes
The homepage offers Daily, Weekly, and Monthly views, with entries dated 17–20 November in the supplied extract. Items carry topic labels including Evaluation, Inference pricing, Model provenance, and Agent security, and some summaries preserve the distinction between a reported test and a company's own measurement. One listed study says a pipeline recomputed 659 astrophysics papers and identified 872 errors; the extract does not show the underlying paper or a site-wide sourcing method.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue