Your rivals rarely announce the small stuff. A plan gets repriced overnight. A new feature slips into a release note. The homepage headline changes its tune. If nobody on your team happened to look that week, you hear about it from a customer. And let's be honest: opening twelve tabs each morning is a habit that lasts about a fortnight.

So we handed the job to an agent on Machine. Once a day it reads the competitor websites, changelogs and pricing pages we care about, compares them with what it saw the day before, and posts a brief note in Slack covering only the real changes.

The problem

Watching the competition is everybody's job, which means it is nobody's. Anyone can read the pages, yet they live in a dozen places and change whenever they feel like it. On most mornings nothing has shifted at all, so the person checking burns their time proving a negative.

The usual fixes fall apart. Page-change alert tools ping you for every comma and font tweak, and the one change that matters drowns in the noise. A shared checklist of pages works only while someone remembers it exists. What you actually want is a check that happens every single day, remembers the last visit, and speaks up only when there is something worth your two minutes.

What we built

An agent wakes up on a schedule each day. Every run gets its own sealed-off session, pulls down the competitor pages on our list, lines them up against the copy saved last time, and sends Slack a short rundown of the differences. Styling tweaks are ignored. New prices, new features and new positioning get flagged by name.

How it works

The pattern

Start the check with a cron trigger. Store the page list and yesterday's snapshot as skills and memory. Give the agent tightly scoped connectors for the web and Slack, then let every run compare and report. You read a summary; nobody goes clicking around.

Guardrails

A read-only agent still gets every control the rest of Machine uses:

• Isolation. Every run lives in a sandbox of its own, and the one Slack summary it is permitted to post is the only thing that leaves.
• Scoped secrets. Credentials are stored encrypted in the Secrets Manager and only handed to the agents you grant them to, injected into the sandbox when the run starts.
• Human approval gates. Posting a summary is the only outside action it takes. Anything more would need a person to say yes first.
• Everything is code. The competitor list, the pages to read and the rule for what counts as a real change sit as files in the repo. They carry a version history, and you change them through a reviewed change request, not a toggle in a dashboard.

The outcome

When a rival reprices a plan or ships a feature, we know the next morning, not whenever someone stumbles on it. Since every run compares against the one before, the channel stays quiet when the market is quiet and only lights up when something has genuinely moved.