Show HN: Raven – The harness of harnesses, built for RSI
Raven is an open-source meta-harness that orchestrates multiple specialized AI agents and uses an iterative loop to self-improve its prompts and strategies.
summary is enough The repository introduces an ambitious multi-agent architecture, but the discussion offers a necessary reality check on the hype around fully autonomous workflows.
summary
Raven is a pre-alpha meta-harness designed to coordinate built-in and third-party AI agents via directed acyclic graphs for complex multi-step tasks. It incorporates an evolutionary component that iteratively tests modifications to agent prompts, tools, and strategies against validation benchmarks. While the authors present benchmarks across coding and simulation tasks, Hacker News discussions strongly debate the realism of unattended execution and whether iterative prompt-tweaking qualifies as recursive self-improvement.
key points
- Raven orchestrates specialized built-in agents and third-party tools like Claude Code within a shared task graph and memory layer.
- A separate Curator module evaluates candidate changes to agent prompts and toolsets against local benchmarks before deployment.
- Commenters heavily push back on the marketing term recursive self-improvement, noting the process relies on standard iterative loops rather than weight updates.
who should read it
AI engineers and developers building or experimenting with multi-agent orchestration frameworks.
discussion skeptical
consensus
Commenters are skeptical of unattended multi-day workflows, arguing that real engineering requires active, continuous human steering rather than trusting a single prompt.
pushback
Some practitioners welcome open meta-harnesses that unify multiple native agent ecosystems and persistent memory layers under one hood.
Well for starters the use of the term "Recursive" is very dubious.
This is as far as the eye can see all very iterative, there is no tail-call or anything fancy, it is loops. Also "Recursive" I feel kind of tries to imply the LLM's weights are being pushed around in some feedback, because to recurse you have to invoke the thing you're recursing into at its very start right? That would be reinforcement probably, and there is none of that in any such RSI so far, at least not public. Please someone contradict me with examples.
So please: "ISI" for iterative self improvement is fine. Also "ISI" …
It dissects the technical accuracy of the term recursive self-improvement, explaining why the mechanism is merely iterative looping.
Generated by an LLM from the article and thread. Check the sources before relying on it.