Back to blog

Office Assistant Agent: PPT, explanatory notes, and data tables for group meeting reports—delivered according to a deliv

Office Assistant Agent: PPT, explanatory notes, and data tables for group meeting reports—delivered according to a deliverables list. When preparing a group meeting report, I used to start with "open the PPT template" and end with "there's no time left." Now I start with a deliverables list. The fou

AI office assistant agentpersonal databasecollege student AI assistantcollege student AIAI for making PPT

Office Assistant Agent: For group meeting presentations, I deliver the PPT, notes, and data sheet against a deliverables checklist

Preparing a group meeting presentation used to start with “opening the PPT template” and end with “I’m out of time.” Now I start with a deliverables checklist.

The four items on the checklist

1. A one-page outline (Word): what this presentation covers, how many sections, and one sentence per section.

2. A slide deck (PPTX): no more than fifteen slides.

3. A data sheet (Excel): the raw values behind the charts.

4. Half a page of speaker notes (Word): one opening sentence, the supporting slide number for each conclusion, and three questions I might be asked.

For each item: what I provide, what it produces, and what I revise

Outline.

I provide three sentences: what this presentation is for, who the audience is (same group or cross-group), and what counts as finished (my advisor has heard it and can give next-step feedback). It produces a one-page structured outline. What I revise is the order—it tends to arrange things by workflow, and I switch to a “conclusion first” order.

Slide deck.

I provide the outline and a few key figures. It produces a PPTX that can be previewed and further edited. I revise three things: delete decorative pages, change each slide title into a declarative sentence, and increase the font size of captions. I go back and revise formulas and units myself; I don’t leave that part to it.

Data sheet.

I provide the raw data. It produces a table. I revise the column names and units and check each value one by one—it handles formatting, not doing the calculations for me.

Speaker notes.

I provide the outline and my three conclusions. I generally use only half of what it produces: I keep the opening sentence and supporting slide numbers, and I rewrite the entire “questions I might be asked” section myself, because that is my prediction of how reviewers will think, not something it can infer.

I turn a “one-sentence request” into three sentences

What to do, who it’s for, and what counts as done.

The official description of the Office Assistant Agent is “describe your request in one sentence and get editable PPT, Word, and Excel deliverables,” with support for output in multiple formats and deliverables that can be previewed and further edited. These editable deliverables are crucial to me—the materials I hand in always need another round of edits according to the school template. In my experience, it’s best to include all three elements in that one sentence; otherwise the deliverable you get is “viewable” rather than “submittable.”

Three things I don’t skip

· Every number in a chart gets checked against the original file.

· A conclusions slide must have supporting slides; if not, add the data first—never write the conclusion first.

· Reference formatting is manually adjusted according to the school template.

Boundaries and costs

· Unpublished data is not uploaded. For sensitive values that must appear in the presentation, I use placeholder illustrative values and fill them in verbally when presenting.

· For a dozen-plus-page deliverable in one go, the credits consumption is noticeably higher than I expected. So now I do it in two steps: first produce the outline, confirm the structure and conclusions are fine, then produce the deliverable—if the order is reversed, rework costs more.

· The materials are not assembled on the spot: images and data come from MuchSci’s Personal Knowledge Base; DOI and citation information for the literature were completed in advance with MuchSci’s Paper Search Assistant; only then is the final deliverable handed to MuchSci’s Office Assistant Agent.

What am I doing the night before the presentation? Reading through the speaker notes and thinking through the three questions I might be asked one more time. The materials themselves were handed in that afternoon.

This is what I now understand a graduate student AI assistant to be: it turns “making materials” into a checklist-driven process; the judgment and trade-offs are still mine.

Back to articles