Hey there, just thought I’d share one classic “donkey work” task that I’ve been exploring using the RA for. I teach actors how to use their voices in the Theatre program at the university where I work. I have lesson plans for all my courses that I teach as markdown files in Obsidian. Every file is a class, and each one is number and dated. Every year, I go through and review my plan, and update each file with the date it will occur on. We have a midterm break (I’m in Canada, so it’s Fall Reading Week in October, and Winter Reading Week in February) and they move around each year.
I copied my plans and folders from this year’s files into the root directory of my system, and set the RA the task of reviewing each plan and updating the date of every class. Any dates embedded in a plan (for an upcoming assignment deadline, for example) were also flagged and updated.
Then I moved my .docx course outlines/syllabi into the course folders so that the RA could review those. The overview of the semester is embedded in those documents, so I got the RA to update that table and all the dates in it.
The only negative is that going through the course outline (a Word file) seems to burn through tokens pretty quickly. So I’m having to do one course at a time. But it’s worth it. Next year, I’ll be on sabbatical, and I’m handing my courses off to a part-time faculty member to cover for me. She’s willing to follow my plan, so being able to hand it off to her with all the dates updated already is going to make her life MUCH easier.
2 Likes
For future consideration, perhaps you could convert the Word documents to Markdown and then export to Word if you need to print them or hand them off to others. I have found, as I posted here, that AIs work natively and therefore faster with Markdown files, and, I presume, burn fewer tokens.
Appendix;
After I wrote the above, I asked Claude if my assumption was correct. Here is its response:
Native format. Markdown is effectively native for Claude. It is plain text with lightweight syntax that Claude reads and generates directly. Word documents (.docx) are zipped XML packages that must be parsed before Claude can read them and reconstructed when written back. So yes, Markdown is the format Claude works in most naturally.
Speed. For editing tasks, Markdown is faster end-to-end. There is no conversion step on either side, fewer places for formatting to go wrong, and the round-trip is a simple text read and write. With .docx files, each read and write involves extraction and repackaging, which adds latency and occasional formatting hiccups (stray styles, lost comments, altered spacing).
Tokens. This is where the nuance matters. Tokens are counted on the text content itself, not the file wrapper. A paragraph of prose costs roughly the same number of tokens whether it arrived as Markdown or was extracted from a .docx. Where Markdown actually saves tokens:
- No inherited Word styling artifacts (tracked-change residue, inconsistent smart quotes, invisible formatting runs) that sometimes get pulled into the text stream.
- Simpler syntax for structure (
# versus a heading style) when structure needs to be represented in text.
- Cleaner diffs when Claude returns edits, which means fewer tokens spent re-emitting boilerplate.
The savings are real but usually modest, not dramatic. The larger practical wins are speed, fidelity, and fewer formatting surprises.
Bottom line. For a writer’s workflow, batching edits in Markdown and exporting to Word only when a recipient requires it is the right call. The user gains speed and reliability and loses nothing that Word offers at the editing stage.
3 Likes
Thanks for this insight. Very helpful!
1 Like
I’m also faculty and used Claude to help me develop new draft Syllabi quickly. I took several old Syllabi, and asked it to create skills based on how the class worked and me just rambling a bit about them using Speech to Text. So then I ended up with a skill for each class.
Then I created a new task/chat where I asked it to combine those skills together into one that would apply no matter what I was teaching.
Then I took that master syllabi skill and rambled on about a newer class I was teaching. I mentioned what the problems were last time that I wanted to fix, some ideas for assignments and outcomes, a general structure for the course, etc etc. Just all thoughts I’d had about it, scattershot. Then it came up with a pretty decent syllabi that really felt like it was based on what I already do - but better.
I’ve also been teaching for like… 20 years? More? 
And realized when I give assignments to students, I do it in a shorthand. It makes sense to me, but for a 20 year old who is encountering the subject (or college) for the first time, it could be way more clear.
So I’ve used the AI to translate my shorthand into much more structured assignments. Like starting with “Write me a one page reflection on how the reading relates to your project.” I know what that means and it seems pretty obvious. But I get back something that’s about a paragraph and gives several specific prompts like “Are there strategies discussed in the reading that you could adapt for your project?” that can help students actually think it through a bit more.
1 Like
thanks for sharing this, it’s giving me a few good ideas for ways to improve my own assignments.