To speed up research, we have to shorten
the loop from idea to feedback.
That lets us test more ideas, learn faster, and push the frontier along the way. Agents will be at the core of that process, performing the gruntwork, guided by the scientist’s hand. rapidreview is building systems to make this approach to AI research easier.
The core objective of research is to test scientific ideas — to understand what works, learn from the result, and push the frontier of knowledge and performance. Everything else in a lab exists to serve that test.
The shorter the loop from an idea to its feedback, the faster we learn and move forward. A group that closes the loop in hours doesn’t just publish sooner; it asks better questions, because every answer arrives while the question is still fresh.
Agents change both halves of the loop: experiments start better designed, and results stop waiting on logistics. Better experiments plus faster execution is a faster feedback loop — and a faster loop compounds into more intelligence coverage. The figure above is our whole argument. Everything we build lives somewhere on it.
Most wasted compute is an experiment that shouldn’t have been designed that way. The fix is not more GPUs; it is starting from what the field already knows.
A dashboard of news as it lands. A living map of research — every dot a paper, edges shared ideas. A recommendation engine that knows what a project needs next. Together they hand every new project the right prior research instead of a blank page. Good understanding of the relevant research, plus good understanding of the available data, is what a better experiment is made of.
LLMs are not strong at ideation by default, but they excel at logical reasoning — they can put two and two together. That makes agents excellent at executing specified targets: design the experiment precisely, and it will be run exactly, at any hour, on any machine in reach.
Agents run the experiments. People decide what they mean.
Experiment design and execution live in one place, so a run is one click. Results come back to a person, and the review sends the next round out sharper. Faster execution means more experiments per unit of time — and the loop, not the run, is the unit of progress.
The right hardware is whatever you can reach right now: a Lambda cluster, a single cloud GPU, the university machine, or the laptop on your desk. Experiments are sized to the hardware at hand, from a $1 smoke test to a $10,000 sweep.
Because a round now spans many machines at once, monitoring is part of the instrument — watching progress across a fleet by hand is the definition of intractable.
We ship the framework as two instruments. Maps is the ideation half: a living map of 500,000+ ML papers and the twenty million connections between them — semantic search, similar-work neighborhoods, and a research agent that reads the field with you. It is live today in early access, with a public API, a Python SDK, and docs.
Merv is the execution half, built against the two things agents still lack: reliability and visibility. Models get better at the task every month, so Merv is not another agent harness — it is the gating mechanism around whichever agent you run: project state, resources, and independent design and experiment reviews, with cloud sandboxes for the runs. It is open source and installs into the coding agents you already use.
Both are in daily use on our own research.
If accelerating AI research is a problem you think about, we should talk.