Meta-agent: self-improving agent harnesses from live traces
We built meta-agent: an open-source library that automatically and continuously improves agent harnesses from production traces. Point it at an existing agent, a stream of unlabeled production traces, and a small labeled holdout set. An LLM judge scores unlabeled production traces as they stream. A proposer reads failed traces and writes one targeted harness update at a time, such as changes to prompts, hooks, tools, or subagents. The update is kept only if it improves holdout accuracy. On tau-bench v3 airline, meta-agent improved holdout accuracy from 67% to 87%. We open-sourced meta-agent.…
Meta-agent: self-improving agent harnesses from live traces launched on April 6, 2026 with 14 votes, #1,096 of 21,248 launches that month and more than 90% of that year's launches.
In plain words
Meta-agent is an open-source library that automatically improves AI agents by analyzing production traces. Users point it at an existing agent, provide a stream of unlabeled production data, and a small labeled test set. An LLM judge scores the unlabeled traces while a proposer generates targeted updates to prompts, tools, hooks, or subagents, keeping only changes that improve accuracy. The library currently supports Claude Agent SDK with plans for additional frameworks. It demonstrated a 20-point accuracy improvement on a benchmark airline task.
written from the facts on this page · September 2026
From the sources
Continual harness optimization. Contribute to canvas-org/meta-agent development by creating an account on GitHub.
Meta-Agent turns execution traces into persistent improvements to the system around a language model. It learns at two timescales: The base model remains frozen. Meta-Agent keeps what it learns in harness code and versioned proposer instructions, backed by an on-disk experience store. Research question: Can an agent improve both its harness and the process used to improve that harness? An agent is more than its model. The harness decides what context the model sees, which tools it can call, how it tracks state, when it checks its output, and when it stops. Meta-Agent treats that harness as an editable program. It runs the current harness, records task-level traces and outcomes, proposes a…from github.com
In the maker’s words, at launch
We built meta-agent: an open-source library that automatically and continuously improves agent harnesses from production traces. Point it at an existing agent, a stream of unlabeled production traces, and a small labeled holdout set. An LLM judge scores unlabeled production traces as they stream. A proposer reads…
Does a similar job
all alternatives →
Progress AI ObservabilityAug 2026 · telerik.com · ▲168Hi HN! We’re the team behind Fiddler, and we recently built an Agent Cache feature. It works by capturing an LLM request/response pair and, once caching is enabled, serving subsequent matching calls locally so duplicate requests never hit the provider. It’s meant to reduce three pains that show up together in agent development: token cost, feedback-loop latency, and output non-determinism. We’d love to hear your thoughts.
- OS
Aviator Agents2025 · aviator.co · ▲97When we asked developers from large engineering teams, almost everyone is using Claude Code, Cursor, or Copilot. But adoption is still inconsistent. Some of us would have 15 agents running in Claude Code at the same time; some would still refuse to use any of them and write code manually. There's a fragmented AI development problem: - Five developers building similar features, and all of them start from scratch - No way to share "here's how we prompt AI to follow our architecture" - No trail of how the AI changes were generated - People leave the company, so does the their knowledge Aviator…
20+ Claude Code agents coordinating on real work (open source)Feb 2026 · github.com · ▲53Single-agent LLMs suck at long-running complex tasks. We’ve open-sourced a multi-agent orchestrator that we’ve been using to handle long-running LLM tasks. We found that single LLM agents tend to stall, loop, or generate non-compiling code, so we built a harness for agents to coordinate over shared context while work is in progress. How it works: 1. Orchestrator agent that manages task decomposition 2. Sub-agents for parallel work 3. Subscriptions to task state and progress 4. Real-time sharing of intermediate discoveries between agents We tested this on a Putnam-level math problem, but the…
TrailsJan 2026 · trails-red.vercel.app · ▲81Hey all, Justin here. I previously built Phind, the AI search engine for developers. One of the biggest problems we had there was figuring out what went wrong with bad searches. We had tons of searches per day, but less than 1% of users gave any explicit feedback. So we were either manually digging through searches or making general system improvements and hoping they helped. This problem gets harder with agents. Traces are longer and more complex. It takes more effort to review them, so I'm building a tool that lets you analyze LLM outputs directly to help developers of LLM apps and agents…
More ai this month
the category →
Paintings by AI models: programs run through a simulation of oil paint.
AI · 21d ago · stillwet.art
PixVerse R2▲381A real-time world model you can explore and change
AI · 29d ago · world.pixverse.video
Source: Show HN post ↗ · launch text quoted in part
Launched alongside, April 2026
the whole month →
- AG
Thought the resources for GPU arch were lacking, so here we are
Life & fun · Apr 2026 · jaso1024.com
Built a ~9M param LLM from scratch to understand how they actually work. Vanilla transformer, 60K synthetic conversations, ~130 lines of PyTorch. Trains in 5 min on a free Colab T4. The fish thinks the meaning of life is food. Fork it and swap the personality for your own character.
AI · Apr 2026 · github.com
I made a laptop stand out of solid concrete, in the style of brutalist architecture, with themes of urbex and urban decay
Life & fun · Apr 2026 · sam-burns.com
With social media and now AI, its important to keep the indie web alive. There are many people who write frequently. Blogosphere tries to highlight them by fetching the recent posts from personal blogs across many categories. There are two versions: Minimal (HN-inspired, fast, static): https://text.blogosphere.app/ Non-minimal: https://blogosphere.app/ If you don't find your blog (or your favorite ones), please add them. I will review and approve it.
AI · Apr 2026 · text.blogosphere.app