@0xNixxx: Anthropic runs the same agent 10 times to get 3 that work Run 10 → Keep 3 → Read the 7 → Fix the harness → Run again - …
Summary
Anthropic uses a method of running the same AI agent multiple times in parallel, keeping successful runs, analyzing failures, and iteratively improving the process to enhance performance.
View Cached Full Text
Cached at: 08/30/26, 10:24 PM
Anthropic runs the same agent 10 times to get 3 that work
Run 10 → Keep 3 → Read the 7 → Fix the harness → Run again
- Run 10: same harness, same task, run in parallel
- Read the 7: the failures are unrelated, so every one gets read by hand
- Fix the harness: the change goes into its prompt, never into the task
- Run again: until they are willing to walk away and leave it running
the agent does this to itself one level down. it bins its own work and restarts the moment it stops climbing against the grader
the grader is a second model. it opens the app in Playwright and clicks through whatever the first one built
bookmark & watch it tonight
Similar Articles
Effective harnesses for long-running agents
Anthropic introduces a two-part solution using an initializer agent and a coding agent to enable the Claude Agent SDK to effectively handle long-running tasks across multiple context windows by maintaining a clean, incremental state.
@zodchiii: Anthropic engineer showed how one person can run 5 AI agents, that code, test, review, and deploy at the same time. In …
An Anthropic engineer demonstrated how a single person can orchestrate five AI agents to simultaneously code, test, review, and deploy software, and compiled a guide on building effective agent teams.
@zodchiii: Three Anthropic engineers just spent 16 minutes on what makes AI agents actually succeed in production. If the people w…
Anthropic engineers share insights on making AI agents succeed in production, highlighting proven patterns from their work on Claude.
I've been running production AI agents for months. Anthropic's "dreaming" feature solves the exact failure I kept hitting
Anthropic unveiled 'dreaming' and other updates for Claude Managed Agents, enabling AI agents to learn from past sessions and self-correct, alongside reports of 80x annualized growth.
@0xwhrrari: Anthropic engineers showed how to build agents that run for days with graph engineering "More than 30% of our code is a…
Anthropic engineers shared insights on building long-running agents using graph engineering in a workshop, highlighting that over 30% of their code is written by agentic graphs to accelerate development.