All a standup really needs to tell you is who did what yesterday and who's doing what today. Getting there is the annoying part. Either you book a meeting for it, or you open a thread and watch half the team reply while the rest forget. Meanwhile the answer is already sitting in Linear and GitHub. It just needs collecting, tidying into a readable post, and a polite nudge for whoever stayed silent.

Machine does that collecting for us. Every weekday an agent looks at each person's work in Linear and GitHub, pings anyone whose day left no trace, and drops a short standup into Slack. It's the standup our own team reads each morning, and nothing about it is specific to us: any team can copy it.

The problem

Ask a room of engineers to repeat out loud what their tickets and pull requests already say, and you've invented the standup meeting. Swap it for an async thread and the meeting goes away, but the chore moves onto each person. You have to remember to post, boil your day down, and actually hit send. Whoever skips it leaves a hole.

The usual fixes only get you partway. A bot that asks "what did you do yesterday?" every morning still needs a human to answer it. A dashboard of activity has the raw data, yet it won't summarize anything or point out who has gone quiet. And neither one looks at the places where the work actually happened, so the update is never better than the effort people put into typing it.

What we built

Each weekday morning, a schedule on Machine wakes an agent. It reads the Linear issues and GitHub activity for every person on the team, writes a few lines about each of them, spots anyone with an empty record, and posts the whole thing to a Slack channel. People with a quiet day get a friendly nudge to fill in what the tools can't show. The agent looks but doesn't touch: Linear and GitHub stay exactly as it found them.

How it works

The pattern

Every morning a cron trigger opens a fresh session that can only read Linear and GitHub through its connectors. The team list and the post format are kept as skills and memory. The agent writes the summary, nudges whoever is missing, and posts it to Slack, which is the only thing it ever writes.

Guardrails

This agent sees what everyone worked on in two tools and then posts to a channel the whole team reads. So we keep its reach small:

• Isolation. Each run gets a sandbox of its own. Inside it the agent reads the day's activity and writes the summary, and the Slack message is the only thing that leaves.
• Scoped secrets. The keys for Linear, GitHub and Slack are stored encrypted in the Secrets Manager and only handed to the agents you grant them to, when the run begins.
• Read-only across the tools. The Linear and GitHub connections can't write. Apart from the standup post, the agent leaves no mark in any system.
• Everything is code. The team list, the format and the permissions for each tool are files in the repo. They're versioned, and any change goes through a reviewed change request, not a toggle on a settings page.

The outcome

These days the standup shows up in Slack before anyone asks for it, built from work that was already logged, and anyone who went quiet gets flagged rather than forgotten. No one sits through a meeting to repeat their tickets, and no one writes up a day the tools already captured.