- Engineering leader Geoff Cox said in a HackerNoon post published October 28, 2026 that an AI bot replaced more than 30 school emails a week, plus a messaging app with no email feed, with about seven digests a week.
- The bot can read his Gmail, calendar and the app but sends only from its own account to a shared list, and never changes calendar events or replies to schools, according to Cox.
- The failures that mattered were quiet ones, such as a digest that silently skipped a source, and each became a written rule, several of them enforced by automated checks.
Geoff Cox, an engineering leader based in Seattle, described how he and his wife replaced a stream of school emails and texts with one afternoon digest written by an AI bot, in a HackerNoon post published October 28, 2026. The post is labelled AI-assisted, and the figures below are Cox’s own.
The Problem and the Numbers
The family has two children at two schools, and messages arrive from the district, both schools, teachers, parent groups and an after-school club, according to the post. Cox counted roughly 200 school-related emails between September 9 and October 21, a bit over 30 a week. A sports team app sent another 120 over the same period. Some school mail was duplicated, because the after-school club often sent the same notice once for each child.
The schools also use TalkingPoints, an app for announcements and teacher messages that never reaches email. In one sample week it carried about 13 announcements and three active conversations with teachers.
What the Bot May and May Not Do
Each afternoon the bot reads new school mail in Gmail and new TalkingPoints messages, follows links to full newsletters and writes one digest. In his description, items are grouped by school with must-dos first, events carry a time or say none was listed, duplicates appear once and sports team mail is left out. A second email on Saturday evenings looks ahead several weeks. The bot also checks his Google Calendar, flags items already on it and offers a one-click link to add the rest. If nothing new arrives, it sends nothing.
From September 9 to October 20 the bot sent 44 digests, 40 daily and four weekly, which Cox puts at about seven emails a week in place of 30-plus emails and an app. He estimates it saves him 30 to 45 minutes of reading a week.
Most of the bot’s rules concern what it cannot do. It has its own Gmail account and sends only to a list Cox shares with his wife. It has read-only access to his Gmail. It never creates, changes or deletes calendar events, never replies to schools and never invents a detail. It works out each weekday from the date itself.
TalkingPoints needed the most care. It has no official way for a program to read it, so the bot reads it through a web browser on a cloud computer. Sign-in requires a code texted to a number the school manages. When the session expires, the bot stops and asks him to sign in, and he types the code himself. Cox considered giving the bot its own virtual number and decided against it, because a number the bot could read would act as a key to the account.
The Quiet Failures
The failures that mattered most, in his account, were ones that looked fine. On September 11 the bot took a weekday from a message’s text instead of its send date, and it now computes every weekday from the date. On October 10 the weekly email skipped TalkingPoints without saying so. The bot must now check it on every run, and a warning goes at the top of the digest if it cannot. On October 11 a digest arrived full of raw formatting code, and a small program now checks every email before it is sent. On October 15 the bot wrongly reported that TalkingPoints had signed it out, and it now confirms a sign-out by opening a page only a signed-in user can reach.
There were plumbing problems too. In early October the bot’s Gmail connection lost the ability to send, and a backup route was later blocked by the cloud computer. Five of the 44 digests were corrections or second versions, Cox says.
Where the Rules Live
Each bot has a profile file in a git repository of markdown notes, recording its instructions, access and the lessons above, so it can be rebuilt if its cloud computer disappears. Changes arrive as pull requests, and 185 automated tests run on each change. Cox published a template for such bots under the name memex-bots. Its rules include that the AI finds and judges while ordinary code scores and writes the output, every item needs a source or is dropped, and a failed run still opens a pull request with a “Run problems” section.
He runs the bots on Grok Bot for now and plans a follow-up post with step-by-step directions for building a school digest bot.
