Field report · November 2026

The State of the Harness

A language model on its own can only read text and write text. Everything that lets it do work in the world is the harness: the loop, the tools, the memory, the rules. This report maps what today's leading open-source and commercial harnesses can do, what each capability is for, and where each product stands.

Agent=Model+Harness

  • 40harnesses21 open source · 19 commercial
  • 50capabilities
  • 8dimensions

Download The State of the Harness Report

We compared 40 AI agent harnesses across 50 capabilities. Get the full field report.

Required

By getting the report, you agree that Helvia.ai stores and processes these details to send it to you and may contact you about it. See our Privacy Policy.

Your copy is ready

Thanks for your interest. Download the report, or open it in your browser.

Download the PDF (2.9 MB) Open in a new tab
Cover of The State of the Harness, a Helvia.ai field report, November 2026

Eight things the data says

We checked 40 coding agents, personal assistants, coworker apps and general agents against 50 capabilities, with a source behind every mark.

  1. The basics have converged
  2. Open source is not behind
  3. Open source wins on model freedom, commercial on the organization around the agent
  4. Safety is strong at the door and thin behind it
  5. Nobody controls what the agent writes
  6. Memory arrived before memory governance
  7. The frontier is half-built
  8. Agents have left the laptop

Also inside: the eight dimensions of a harness, open source against commercial, a trust section on safety, memory and data flow, profiles of all 40 harnesses, the full capability catalog and matrix, and our method.