Orchestrate the full research loop with parallel agents
Sakana's Agentic Scientist uses a scaffold of parallel agents to run the whole research loop, rather than one agent writing a paper.
Done with a custom-built or unnamed agent
- What they did
- The post announces the third version of Sakana's automated AI Scientist system, called the Agentic Scientist, and links to its GitHub repo. It says the design changed from a single agent writing a paper to a scaffold of parallel agents that orchestrates the full research loop. The post gives no detail on the individual agents' roles or on where people step in.
- How it went
- The post gives no results, numbers or evaluations, and no account of limitations. It is an announcement of the release only.
- Worth knowing
- For mechanics, setup steps and costs, read the linked GitHub repository, because the post itself says nothing about them.
Try it yourself with Manus
I want to run a small research project on [research question or topic]. Split the work across parallel sub-agents: one surveying existing literature, one proposing hypotheses, one running analysis or experiments on [data or sources I can provide], and one drafting the write-up. Finish with a short report that cites its sources, lists what was tested and what was found, and flags weak points, and show me the plan before you start.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Bluesky · Oct 15, 2026
Related use cases
Find connections across 100 non-fiction books
Claude Code browses a personal library with CLI tools and surfaces trails of related excerpts across books.
Research AI training case studies and build a summary deck
A team had Agent Mode research case studies from McKinsey, PwC and similar firms and make a slide deck in 32 minutes, with some image and overlapping text issues.
Run first-pass topic research, then verify references
A commenter uses Manus for first-pass research, then has Genspark double-check every reference and flag problems.
Research content from Slack and draft Notion guide ideas
A team DM'd Lindy in Slack to summarize a year of guides and research a model, and it created a Notion guide idea after asking for approval.
Research a person and write a bio
An Inc. writer had Muse research them via its browser and write a bio, but objected that it accessed private messages unasked.
Produce a sourced research document with parallel agents
Grok coordinated 16 specialized agents that searched and parsed 896 sources, debated and peer reviewed, and delivered a complete research document in one minute.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.