Tag: evals

All the articles with the tag "evals".

• AI agents need incident reports, not just dashboards

  Published: 20 Sep, 2026

  • 6 min read

  

  OpenAI's new misalignment reporting framework is a useful signal for enterprise teams: if agents can act, they need traces, eval gates, ownership, and a boring incident loop.

• AI agents are not failing because they’re dumb. They’re failing because the system around them is sloppy.

  Published: 18 Apr, 2026

  • 6 min read

  

  Most agent products do not break on model quality first. They break on messy context, brittle tools, fake evals, weak recovery paths, and sloppy release discipline.