Position title: AI Research Assistant
Reports to: Paper author (that is, you)
Work nature: Execution role, no decision-making
Translated into the language of organizational behavior, "AI helps you write a paper" is actually a job description. The clearer the responsibilities are written, the fewer pitfalls there are in collaboration. The following job description is based on the real capabilities of muchsci Paper Sailor in the execution stage.
Job responsibilities
1. Revise the manuscript text. Based on the real results you provide, correct the wording of the manuscript—note the premise: based on real results, not made-up ones.
2. Data analysis and subset filtering. Analyze the data, filter subsets by conditions, and re-run statistics. It handles all the grunt work of data processing.
3. Create publication-quality figures. 8 chart types plus multi-panel assembly, with SVG vector output and journal column-width adaptation—for example, Nature's 88 mm. Figures are the face of a paper; this item alone is worth the price.
4. Run scripts. Data cleaning, batch processing, format conversion—all the parts that count as "dirty work."
Division of labor: the boundary between execution and judgment
The most important part of this job description is "what it does not do":
| It does (execution) | You do (judgment) |
| --- | --- |
| Revising the manuscript text, data analysis, and subset filtering | Providing real experiments and data |
| Creating publication-quality figures | Judging whether the conclusions hold |
| Script batch processing and format conversion | Being responsible for the viewpoints and final draft |
The boundary is drawn at "real data is provided by you, judgment is made by you"—execution can be outsourced, but judgment always remains in-house. This is also the boundary muchsci has upheld by design: tools do not touch the judgment layer.
Collaboration scenarios and division-of-labor scripts
When experimental data just comes out: "Use this data for analysis and update the relevant paragraphs in the manuscript"—data goes in, the manuscript is updated, and both sides stay in sync.
When the paper needs figures: "Create publication-quality figures based on the results, 88 mm column width, SVG"—one sentence from data to finished figure.
When data needs subset filtering: "Filter the data subset for XX conditions and re-run the statistics"—when reviewers ask you to add analyses, this one is the fastest.
The value of a good assistant is to give your time back to judgment. Collaboration with clearly written job responsibilities is collaboration where you don't have to worry about crossing boundaries—with this muchsci job description, we recommend printing the full text and posting it at your workstation.