I want AI to do my laundry and dishes so that I can do my 1-1s and discussions.
The only thing that matters in community organizing is: meeting people and strengthening your relationship with them. Literally everything else (e.g. deciding on socials dates, reconciling calendars, tracking who's attending, figuring out if a cohort can absorb a drop-in) is dishes. Whenever possible, we should not be doing dishes!
Epistemic status: anecdotal experience. I explain in boring detail how I use AI-assistance to run one (1) uni group.
I've been running EA at Penn, and frontier LLMs have been immensely useful at minimizing time spent on ad-hoc/logistical tasks, or as I like to call them, "work about work" -- a term coined by... Dustin Moskovitz? Anyways, in other words, it's not real work! I mentioned this offhand to a mentor of mine on a call and they said that I should text it to other organizers. I decided on an elaborate post with examples and a template in hopes it's helpful.
I divide the dishes in my mind into two:
- Macro dishes: planning a year (calendar, budget, strategy). With AI-assistance I finished 80%+ of the 2026-2027 macro dishes in 2h35m (last semester took way longer with no AI-assistance).
- Micro dishes: on-the-go planning like sudden incompatible availability, tracking attendance, sending reminders, sharing opportunities, etc. The example I use here is from the virtual summer fellowship (which you should run!) we're currently wrapping up.
TL;DR
- Make a chatbot account for the group. Connect the club's Gmail, Drive, and Calendar. Slack and Airtable if you use them.
- Make it do macro (calendar, budget) and micro (schedule, reminders, attendance, opportunities) ad-hoc tasks.
- Prompt it with context dumps and tool references via transcriptions.
- A 2026-2027 calendar template that I think can be useful for other EA groups is available here.
Setup
Get a $20/month subscription, and connect the chatbot to your group's:
- Gmail (you need to make it a habit to CC the club email when useful)
- Google Drive (you need to make it a habit to be organized!)
- Google Calendar
- optionally Slack, Airtable (Todoist or other to-dos would work; I don't use them)
All of which seems very intuitive; these LLMs can do agentic work now, so why isn't everyone using them? I have no idea, but from first-hand experience I rarely such LLM use by EA community organizers.
The account should be dedicated to the group for two reasons:
- Chatbot accounts can connect to only one Gmail/GDrive/GCal/etc. account at a time (plus context builds up, and tokens run out, and you don't want your personal usage turning off a few scheduled reminders), so you should separate your personal and your work accounts.
- Succession. Too often, uni groups die; apparently, there are currently only seven "established" EA uni groups (established means very active if I understand correctly). I personally had to spend tens of hours plowing through the old Google Drive and Gmail (read everything) to catch up. Imagine instead if a new organizer could just ask the club's chatbot; I personally would've appreciated that!
Note that consumer plans' ToS prohibit sharing logins. Team/Business plans are a compliant solution but at $25/user/month and with a minimum of 2 seats for ChatGPT and 5 seats for Claude. Succession still works but only if info is left in org-wide Projects, so put them there. (I personally have a single-user account that I need to eventually replace because I don't take my own advice; it's a personal-work mixed disaster, and not on Team.)
Privacy disclaimer: you should keep members' sensitive information off the AI. Don't CC the club email on sensitive comms. For what it's worth, you can opt out of personal data processing by OpenAI using their privacy portal. Anthropic doesn't do that by default.
Macro dishes: planning a year
Calendar
See ours! One row per day, from Aug 3, 2026 until the end of Spring 2027. Eight columns:
- Date
- Consider: holidays and observances that affect attendance. Rosh Hashanah, Yom Kippur, Sukkot, Ramadan, Passover, Diwali, Lunar New Year, Election Day.
- School: the academic calendar (the AI can just hunt it from the web; mine pulled in Penn's academic calendar with no problem). Orientation, first day, course selection (to remind students to select EA-ish classes!), drop deadline, term break, end of classes (often the limit on permitted on-campus student activity), reading days, finals.
- Fellowship: deadlines, review windows, decisions, every cohort session, end-of-fellowship dinners/celebration.
- Outreach: tabling, fresher fair, preceptorial, short talks at lectures, mass-emails.
- Penn EA Events: our public socials and talks. This column is connected to the GCal via a script that makes them viewable on our site.
- External Events: EAG/EAGx dates and their app deadlines, GCP workshops, local ACX meetups, retreats, etc.
- Organizing & Ops: board stuff, re-registration, school-specific deadlines/training, mentorship programs (e.g. CEA's Organiser Support Programme or Matt Reardon's Network for EA Support & Training).
All of that, over 300 days, was generated by a few half-baked prompts with lots of context and a reference to an old template. This should take you less time, cause you can feed your AI ours!
Example prompt
Here is a chunk of a prompt, to show that you don't need to spend time figuring out weird specifics e.g. how to fit two fellowships, each with two cohorts, into a semester:
for the fellowship assume we'll run two per semester (two in the fall, two in the spring) back to back, with 2-3 unused weeks for outreach at the start of each sem (but after that period the first fellowship starts, ends, with the second one happening right away, because the 2nd's one outreach would start within the last 2-3 weeks of the first) you have to figure out the optimal fellowship length for this to happen. make sure to account for school and "Consider:" columns, etc. for each of the four fellowships select the 2 days that result in the best-maximal duration (e.g. maybe mondays are always cancelled so go for tuesdays) because each fellowship would have two cohorts.
Budget
Same session. It read an old budget, a list of things I thought might be worth purchasing (e.g. club fair, posters, website domain), the new blank template (to be submitted to the funder), and the calendar it had just built. It produced 20+ line items, whose quantities and costs I then reviewed and adjusted by hand.
Micro dishes: running a fellowship
Availability coordination is hell
We had 22 admits, each of whom had internships, family travel, time zones, vacation, and/or pre-orientation. One group's only workable slot was 1-2 AM Eastern.
Instead of putting all of them into clean groups that some would then miss, we ran four groups with floating membership: everyone had a primary group (to activate that synergy), but people moved between groups week to week, depending on their availability. This resulted in more attendance compared to our last iteration, and I also think the maintained momentum lowered drop-off (though it could also be due to self-selection; summer fellows applying to a virtual EA fellowship before they even come to campus are more interested in EA, on average).
You can't do floating memberships on your own!
Groups aren't in sync. For example, Group A would be on its second meeting while Group B was on its third.
Imagine a student messages you: "I can't attend tonight's, have a family event, where can I go instead?" To figure that out, you need to find a group whose next meeting is running the same session number n as the one they're missing (you don't want the member to jump ahead or re-discuss content), that also falls before their own n+1th session (which may itself be with a different group), and that also fits the availability they submitted in the Google Form. Oh my god.
Here is what you can do: ask your AI to figure it out! Here is an example prompt that you can speak-transcribe into the AI:
make sure that any members' nth session is with a group that's doing its nth session (so that members' learning journeys make sense), e.g. if someone's first session is a 2nd session-content that's bad ! Review all students and make sure their 1-8 are actually 1-8 sessions, and update calendar invites and emails
That prompt was trying to fix an error I had made. It triggered a full rebuild of all 22 people and their 176 rows (22 × 8), plus corrected calendar invites, plus nine individual apology emails (for sending incorrect invites beforehand, oopsie).
If a member sends me an excuse over WhatsApp, I open my chatbot and say "John [Doe] can't do the next session, relocate him if possible. If not, place him in the most reasonable choice and tell him he's excused if not possible cause I know it doesn't fit his availability." Within minutes he has a reply telling him his new session is on Date X at Time Y, an instant invite, and a 12-hour reminder queued up.
Attendance tracker
AI can build a good tracker with two tabs, one for data, and another for analysis. The first would include one row per person per session (date, time, calendar event, host group, host session, curriculum session, facilitator, participant, email, attendance status, notes). The second would include live formulas per person. Scheduled sessions, attended, excused, absent, not recorded, attendance rate, groups attended, sessions attended, completion status, etc.
(Unrelated disclaimer: you shouldn't stress over your members’ stats or Goodhart them.)
Reminders
Twelve hours before each session, each group gets an email with the reading and the syllabus link, facilitator CC'd.
Back when I wasn't using AI in any capacity, every time someone moved groups I had to update the sheet, the calendar invite, and the reminder list separately... or just not do reminders at all and risk undergrads being undergrads and forgetting.
Rescheduling
One time, I couldn't run a session on a day and was too busy to notify everyone and reschedule them, so I typed:
I can't make this Jul 27 session (today), is there a way to redistribute the people (to another 3rd session, taht's before their 4th) and send them individual emails to the updated ones. would any people not do the 3rd as a result???
It moved all six people in a proper way, notified them, and updated the tracker sheet, GCal invites, and 12-hour reminders. 4/6 attended. When this would happen before, I'd just cancel the meeting and compress the fellowship.
Other small stuff
Say you get an email: "Dear Organizer, Please share this Amazing Opportunity™ with your members." It’s so helpful being able to read the email, quickly determine that it's beneficial for your members, and just say: "chatbot, look at my inbox and share the latest GCP opportunity in #opportunities in Slack, also plz imitate the style of messages previously sent in that channel."
Slop prompting
As you can probably tell, most of my messages were voice-transcribed on my phone. One example, verbatim:
Okay, you sure the groups are, now I want you to reaudit the groups and place people in the groups where they're attending most of the sessions. And you can think of it as attending most of the sessions they were for sure attend or most of the sessions they're invited to, because some of them are invited to stuff that they might not attend because it's not in their availability. You know what I'm saying?
That's a slop prompt, and it's enough. It worked (because it had context before and around it, plus the chatbot's intelligence): it reassigned primary groups, updated the sheet, and revised the emails. I looked at all of it, and it was perfect. No problems arose later.
Literally treat the chatbot as a cracked San Franciscan high schooler who can use tech tools and do logistical work with them really, really well, but who needs to be given constraints, requirements, etc. Just send transcribed messages; do not worry about speaking too much, or about saying something wrong. No need for the Marc Andreessen prompt.
Potential objections
"Does this not make organizing impersonal?" No. I mean, unless there was some personal connection in looking at GSheets and figuring out which sessions intro fellows can hop onto. Getting more time to talk to the members, face-to-face (or virtually), is what you're optimizing for. The AI is simply a secretary, in my view.
The comms it sends are where failure could happen. I'd suggest you write the drafts it uses, like the "Hi, reminder in 12 hours" one or the "Sorry, session's been cancelled" one; those read as automated anyway but being human-written is better. Most importantly, write the personal stuff yourself, e.g. if a member messages you to say "Hi I can't come because I have a family problem". You also wouldn't CC the club email then.
"My group is too small for all of this!" Probably true for macro dishes (sorry, I like the term); you don't need an elaborate calendar if you're running one small reading group. But you do if you want to run multiple, alongside other events (socials, talks, dinners, intro talk, tabling), and want to get your members to join you in going to EAG(x)s and retreats. I think automating small tasks (like around an intro fellowship) is always useful, though.
Templates
I'm sharing an example of the calendar used by EA at Penn, with some private information removed. I believe the rest of the files are AI-generatable quite easily and quickly. I like the calendar system, in particular, because it’s clear; works well for planning many parallel things and seeing the full picture; lets you specify details like location and time; and can be connected with the group’s public GCal (i.e. less dishes!).
To end on a note
In the words of productivity guru Dustin Moskovitz:
[We're at our best] engaged in tasks that are distinctly human... These are the tasks that AI can enable our days to be filled with, by automating the busy work that slows us down... A world with less "work about work" is a world with more breakthroughs... full of happy individuals, living their best lives.
Meta:
your top priority really should be understanding that. Worst case is they have a good reason to not use LLMs, and you might want to worry about that reason to.
Object:
Worst catastrophic error: privacy leak because of shared context.
"Read all my emails then tell my student Bob I'm late"=> "Hi Bob, I'm late, sorry, really hectic day dealing with all the changes going on in the department!" and Bob is thinking "?? what changes?"
Second worst catastrophic error: it deletes all your data.
Other problem: are your sure your AI writing isn't coming across as slop? Your example with forwarding the recommendation was lossy - the LLM did not forward your reason for thinking it was good for the group, it made up a new one. My bet is people can tell your slop is slop more than you realize and I really recommend assuming all writing you do is empathetic thinking involving your listener. It is not safe to outsource to LLM while being an organizer (or anyone who... thinks.)
Could be true, yes. I was hoping I'd get some pushback/feedback here or elsewhere if this is true (which it could be), so thanks for your comment!
On the risk of privacy leak, you can bound what your LLM can do via permissions. Personally, I do not let it send emails without my manual approval, which allows me to glance review all outbound comms (takes a couple seconds, helps avoid errors). For what it's worth, I've never caught a privacy leak. Worst thing I've caught is an overly-long email subject line.
On the erasure of data, version control and trash bins mitigate this risk. Choosing perms is useful here, too -- I personally do not permit my LLMs to delete any files. If an edit is destructive, I'd simply revert that edit via version control (again, for what it's worth, I've never had an LLM "go haywire" and commit destructive edits).
Could be. I do not mind simple reminders to be "sloppy" -- here is something I glanced reviewed and thought it looked good (after which it was schedule-sent):
For sharing opportunities on Slack, it looks something like this (copy-pasted from Penn EA's #opportunities):
Does either of these come off as slop to you?
I personally think that for sending reminders, updates, or opportunities, you're just trying to share some information in a concise manner, and LLMs do that well (based on my personal experience).
Am I missing out on strengthening personal connections because, I don't know, I could've done the same but with my tone (and humor or something)? Maybe. But I'd trade off that small benefit for saving hours. Plus, I think there are more effective ways of strengthening personal connections than having my personal tone in reminders or #opportunities messages.
Edit to add:
Actually it didn't! Because I didn't ask it to include a reason; I told it to share it, and so it did. Personally, when I share opportunities via Slack, I normally don't include reasons for why they're good. When something excites me or if I want to share the reason for why something is good, I just write the message. Also, setting reasonable perms / glance-reviewing would help you avoid an LLM passionately arguing for an opportunity or something. (And again for what it's worth I've never experienced that)
I'll point out, you now are describing the actual engineering work needed to achieve privacy and reliability, which includes context limiting for privacy's sake, and a backup scheme. That might need to be "above the fold" stuff if you're trying to get people to use LLMs for administrivia.
You may need to consider badness v. likeliness of mishaps - for example, I think it can wipe your calendar, and while that's unlikely and it's hard to imagine the exact prompt that would cause it, it's not really unprecedented either. Anxiety about these things will go up as you move outward from tech enthusiasts, and I think it's unclear whose instinct is right here - maybe they're too cautious, maybe you're taking too much risk. If you're an entrepreneur/seller, you want empathy to your customer's risk tolerance. If it's higher, they care more about this type of problem. Maybe you would try to sell those customers on readonly workflows.
I really think the key concept in modern AI prompting / coworking / engineering is verifiability. Software engineers are mostly not stepping back from reading / reviewing AI code, and people sending out real emails really aren't going to want to step back from verifiability.
From a salesmanship point of view - you probably should stress the actual problems it can solve more, and not the fact that it's generally capable and can do lots of stuff if you just connect it to everything. That's a bit of a "draw the rest of the owl" move, and I guess some people who are have tech adopter / tinkerer mindsets will respond to your push, but if you're trying to work your way through skeptics with less time experiment, you really need to solve problems they actually have, not give them half-solutions to nonspecific problems. (Key book, aimed at entrepreneurs: "The Mom Test.")
Short answer: I think that was a good automation, because the "sent by Claude" transparency helps a lot.
Analysis:
In my mental model, "slop" is when the author thinks the AI substitute is human quality, and the reader can tell it's not. If you say "Sent by Claude" then that pretense isn't really there and you and reader are aligned this was an automated process not intended to be empathetic communication. There still might be quality loss, as is true for any automation, LLM or not.
Worth considering where you did make tradeoffs:
It may be better to describe the tradeoff explicitly when it comes up, rather than handwave like, this was easy to automate and there's no slop effect.
100% agree. I think the "personal plea" part is what I think I lose out on when I do this, so I do share the best opportunities manually as a result. Also agree with thinking about both the magnitude and liklihood of an error. Thanks for your reply!