THE NUMBER
94 events.
That's what one day of Andy Maguire's GitHub activity looks like - commits, pull requests, code reviews, issue comments. His standup agent reads all of them and writes one paragraph for his team at PostHog. He built it himself, open source, and says whether it actually saves time is "debatable."
3 THINGS HAPPENING RIGHT NOW
Zapier's standup now takes "about a minute or so"
Zapier's engineering team replaced their daily synchronous standup with a Slack bot that sends three questions to each person and posts the answers to a shared channel. Tangents move to threads. Scott Williams, an Engineering Manager at Zapier, described the result: the standup now takes "about a minute or so, depending on how much you type." No one's on a call waiting for the part that only affects them.
Honeycomb tried async and changed their minds
Honeycomb, a software monitoring company, went async and reversed. Their case: without a fixed time to "stand up," updates get lost in scrollback - and if no one reads them, the whole exercise is pointless. They replaced it with an hour-long daily team sync - five hours a week, longer than most standups, and they say it's working. Not every team should automate the standup - some need more human connection, not less.
One developer built a standup agent over a weekend
Jesse Liamzon documented building an AI standup agent from scratch: each person types a short update; the agent breaks it down into Yesterday / Today / Blockers and emails a digest to each team member. The agent does what the meeting used to. No meeting - 3 minutes of written text per person replaces a 15-minute call. He published the full walkthrough on Medium.
THE DEEP DIVE
The Standup Agent That's Honest With You
Andy Maguire is an engineer at PostHog - a product analytics company, not a standup vendor. He built an agent because he kept losing time every morning reconstructing what he'd done the day before.
The agent reads his GitHub activity: commits, pull requests, code reviews, issue comments. It drafts an update, he refines it in a quick back-and-forth, then approves it before it posts to Slack. He tried auto-posting - the agent was accurate about 90% of the time - and stopped. Teammate-facing text needs someone to own it.
On whether it saves time: "Debatable." The reconstruction is gone. The approval is quick. But setup, debugging, and maintenance aren't free. He keeps using it because the mental work of reconstructing yesterday's activity is the part that drains him - and the agent handles that.
The real math: 94 GitHub events condensed into one paragraph. That's the pattern recognition part - the thing machines do faster than people. The judgment at the end stays human.
This pattern fits any small team where people produce traceable work: tickets, logged tasks, version control. The agent reads the record. You read the summary.
ONE THING TO TRY THIS WEEK
Your standup is already written. You just haven't read it yet.
Every commit you make during the day is a log entry - here's how to have Claude read them back to you as a finished standup.
Open Claude Code in your project folder and paste this:
Write a command-line script called `standup`. When I run it:
read today's git commits in this folder, draft a short standup
from them (what I completed, what I'm working on next, any
blockers visible in the commit messages), and save the result to
standup-today.md. One command, no arguments needed.
Install it so I can run it from any folder.
Claude writes the script and installs it. Approve it once.
At the end of the day, type
standupand step away. Ten seconds later, standup-today.md is ready.Read it, adjust a line, send.
You typed one command. The reconstruction happened without you.
Stuck? Reply to this email. I'll help.
WHAT'S COMING
Next issue: we put an agent and a human on the exact same work for two weeks and kept score - who was faster, who was better, and where the agent quietly fell apart. An honest scoreboard, no cheerleading.
Manu