Observed arrival · 2026-09-17
Harness Lab: an inspectable classroom for AI agents
An educational web app that lets developers experiment with the software layer around language models, including tools, approvals, budgets, memory, and execution traces.
Field notes
Harness Lab organizes its teaching around an agent loop of model, tool, result, and answer, then makes the surrounding controls inspectable. Its lessons cover context selection, persistent memory, tool-argument validation, human approval before writes, bounded retries, and execution traces. A scripted simulator is presented as usable without an API key, while the hosted preview requires Microsoft sign-in and the source is described as runnable locally with one command.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue