A research project normally needs a team. Someone reads the existing studies, someone spots the open question, someone designs and runs the test, and someone writes up the result. Most AI research tools stop at summarizing papers other people already wrote. Handing all four jobs to a single AI system is a practice called vibe research. No person checks any step, so the system can follow many lines of inquiry at once, day and night.
Analemma, a research lab, introduced itself by pushing the idea to its limit in public. Its system, FARS, short for Fully Automated Research System, ran live online for anyone to watch. It reached its target of 100 finished research papers in about 10 days of nonstop work. Human researchers stepped in only afterward, to check each paper before release.
The demonstration became an online workspace called Lemma, built for anyone with a question. Point it at a topic, and it takes on the same four jobs, citing real sources and testing ideas with code. After each stage it pauses so a person can review the result, unless someone sets it to run straight through to a finished study.
One test gave Lemma a question a private equity firm might ask: how AI could make local trade businesses more profitable and easier to scale. The trades included plumbing, roofing, heating and cooling, and electrical work. Asked first for an investment memo, Lemma pushed back, because an opinion cannot be tested. It reshaped the request into an experiment built on measurable numbers: booking conversions, repeat customers, and margins before and after automation. A tested experiment offers a result someone can weigh against what happens inside a shop or a service van.
Working through the reshaped question took about 12 hours. What came back was a quality score and three separate proposals, each backed by references and laid out as a plan to test further. Lemma’s deepest base of work is in computer science, especially AI research, and local trades sit outside it.
Papers produced this way still face the same checks as any other paper before their findings count as settled. A system that can try many possibilities makes attempts easier and more frequent, though that alone does not mean faster breakthroughs.
