Skip to content

About this recipe

Status
tried
Stage
Implement with agents
Output check
Expert-checkable
Autonomy
Drafts and checks; the author decides
Last reviewed
2026-10-30

Recipe 9

Write, review and teach from your own notes

Try this first. Let the agent work from your bullet points and earlier texts, and have students do tasks they have not mastered by hand first.

Summary

Text drafted by agents tends to look alike. Start from your own notes and argument, and give the agent earlier texts as examples of your style. Agents do the work around the argument well: formatting, consistency and checks against guidelines. The author writes the arguments and makes the judgments. Students do tasks they have not yet mastered by hand first.

Steps

  1. Check the manuscript against the journal's guidelines before submission. Editors and reviewers may run the same check.
  2. Write reviews and reports from your own bullet points and argument, and give the agent earlier texts as examples of your style. The German Research Foundation (DFG) allows reviewers to use AI this way, to expand their own notes and edit language, if confidential material is protected and the use is disclosed. Check your funder's rules before uploading a proposal.
  3. Make notation and references consistent across long documents, and replace preprint citations with the published versions.
  4. Use agents to build teaching material: interactive visualizations, tutorials and flashcards.
  5. Ask students to write research plans and conference abstracts, build arguments and critically evaluate scientific writing by hand before they use agents for these tasks.

How to check the output

The author reads every sentence that goes out under their name. A script can check formatting. Only a person can check an argument.

Examples

  • In a textbook with many authors, an agent made the notation and bibliography consistent and replaced preprint citations with published versions.
  • A 900-slide course was rebuilt so that R scripts, web pages and PDF handouts come from one source.
  • A lecturer builds small interactive web pages that let students explore the machine-learning algorithms discussed in class.
  • A newcomer to the centre uses agents to write interactive tutorials on unfamiliar topics.

Known problems

  • A thesis report drafted with an agent from the reviewer's own bullet points was almost identical in structure and wording to the second examiner's report. Bullet points alone were not enough. Adding earlier texts as style examples helped.
  • Students' slides are starting to look the same. See Where to draw the line.
  • Students and junior researchers are still building the skills they need to check agents. Handing that work to agents can reduce what they learn. See Where to draw the line.

Skills

Each skill page has a brief. Give it to your agent to build a version for your own institute.

Best practices on aiforscience.eu that this recipe follows:

Change history

  • 2026-10-30: Renumbered to follow point 9 of the submitted Comment (formerly "Writing, reviewing and teaching").
  • 2026-10-28: Updated to match the presubmission draft of the Comment.
  • 2026-09-25: First draft, from the first round-table.