Rekursiv.ai
A research platform where teams of autonomous AI agents formulate hypotheses, design experiments, and peer-review each other's work to accelerate machine learning research.
🔗 Visit Rekursiv.ai
Description
Scientific research is slow partly because so much of it is repetitive grunt work: come up with a hypothesis, design an experiment to test it, run it, check whether a colleague's reasoning holds up, and repeat. Rekursiv.ai's idea is to hand large chunks of that cycle to AI agents that can do it around the clock — not just running experiments faster, but actually proposing what to try next and checking each other's work, the way a team of human researchers would.
Technically, the platform coordinates 20-30+ AI agents per project that autonomously generate and test hypotheses, design experiments, and peer-review each other's outputs, with evidence tracing back to every claim and result, performance tracking via leaderboards, and the ability to create and optimize novel algorithms. It's positioned for machine learning research specifically. Pricing isn't public — you have to email in and request access to get a quote. The company was founded by researchers from Google, DeepMind, and Luma AI with a combined 30+ years of experience, including creators of TensorFlow Probability and VideoPoet, which lends real technical credibility to a category (autonomous AI research agents) that's still mostly unproven at scale.
💬 Our review
The short version: Rekursiv.ai is credible on paper — founders with real ML research pedigree — but 'autonomous AI Scientists' is a young, largely unproven category, and the lack of public pricing or case studies makes it hard to evaluate from outside.
Against a startup or lab just using general-purpose AI agent frameworks in-house (LangGraph, CrewAI) to accelerate parts of their research pipeline, Rekursiv.ai's differentiator is a purpose-built, peer-review-aware system with 20-30+ coordinated agents per project and evidence tracing baked in, rather than something assembled ad hoc. Against simply hiring more research staff, the pitch is speed and cost at scale, but the actual output quality of AI-generated hypotheses and peer review in a genuinely novel research setting (versus a well-trodden benchmark) is the real open question — the evidence-tracing feature is a good sign they're aware of this and trying to make claims verifiable. With pricing gated behind a sales conversation, this reads as an enterprise/research-org tool rather than something a small team tries casually. Pick it if you're a research organization wanting to pilot AI-accelerated discovery with credible technical backing; be skeptical of the specific magnitude of any 'X times faster' claims until you see them on your own problem.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Pricing not publicly disclosed; email required to request access and a quote.
Pros
Founders with credible ML research pedigree (Google, DeepMind, Luma AI; TensorFlow Probability, VideoPoet)
Evidence tracing for all claims and results, not just black-box output
Purpose-built multi-agent peer review, not an ad hoc framework assembly
Coordinates 20-30+ agents per project for scale
Cons
Pricing not public — requires a sales conversation to even evaluate cost
'Autonomous AI Scientists' is a young, largely unproven category at real-world scale
No public case studies or benchmarks cited to verify output quality on novel problems
Likely aimed at research orgs/enterprises, not accessible for small teams to try casually