Back to blog

October 19, 2026

One Piece of Information, Two Jobs: Why a Simple Cleanup Almost Broke a Different Feature

We're currently building a system that reads incoming email for a business and turns it into tasks automatically: a pickup time, a form deadline, a sign-up link, the kind of thing that normally gets buried in a long newsletter and missed. Pulling that information out automatically saves real time, but it also means the system has to make a lot of small, careful decisions about what counts as useful information and what doesn't.

One of those small decisions almost broke something else entirely, and it's a good lesson in a mistake that is very easy to make with any automation.

The cleanup that looked obviously safe

Emails are full of junk links. Unsubscribe links, "view this in your browser" links, social media icons at the bottom of a newsletter. None of that is useful to a task, and all of it clutters up the result if you don't filter it out.

So we built a filter to strip those junk links out before the system processed the email, keeping only the links that actually mattered, like a sign-up form or a payment page. On its own, that sounds like a clean, obviously correct improvement. Nobody wants a task cluttered with an unsubscribe link.

The conflict nobody saw coming

Here's what we missed at first. That same email body gets used by more than one part of the system. A separate feature reads a short code out of the body of certain emails, to confirm that a message is genuinely being forwarded by the right person rather than a stranger. That code lives inside the same text the link filter was about to clean up.

If the link filter had been applied everywhere without exception, it risked removing or altering the exact piece of text that other feature depended on to work at all. Two completely different jobs, reading the same piece of information, for two completely different reasons. Fixing one would have silently broken the other, and the break wouldn't have shown up until someone tried to use the feature that had quietly stopped working.

Why this kind of conflict is easy to miss

It's easy to think of an email as one single blob of text that a system either uses or doesn't. In reality, a well-built automated system often reuses the same piece of information for more than one purpose. A cleanup that looks completely safe from the point of view of one feature can be exactly the wrong move for another feature that nobody was thinking about at the time.

This is the same shape of mistake as changing a setting that controls one kind of notification, only to discover it quietly also controls a second kind of notification nobody meant to touch. The fix for one thing becomes a silent problem for something else, because both were leaning on the same underlying piece of data.

How we caught it, and the fix

We caught this because every change to the system goes through a second review pass before it ships, specifically designed to look for exactly this kind of conflict. That review flagged the issue before it ever reached a real inbox.

The fix wasn't to abandon the link cleanup. It was to make it conditional: the aggressive link filter only applies for the accounts where it's genuinely needed, and the accounts that depend on reading a code from the email body keep their messages untouched. The two features now run side by side without stepping on each other, instead of one quietly disabling the other.

Turning a new feature on for one real account before everyone else

This same project also taught us something about rollout, not just design. Once the link filtering, time extraction, and calendar features were built and tested on made-up test emails, we didn't flip them on for every customer at once. We turned them on for a single real account first, let real emails flow through, and checked the actual results by hand: were the times right, were the places right, were the right links kept and the junk ones dropped.

Only after that real-world check passed did we start widening access, a few accounts at a time, instead of all at once. A handful of smaller issues turned up along the way, like a type of link that wasn't yet clickable in one view, things that didn't show up in testing but did show up in real use. Catching those on one account is a minor fix. Catching them after turning the feature on for everyone is a much bigger cleanup.

What this means for your business

If you're automating a process that touches the same piece of information in more than one place, like an email, a form submission, or a customer record, it's worth asking a specific question before you change anything: is this exact piece of data used an