As a lean team, I can tell you our issues log is not exactly pristine, or at least it wasn’t. Uncaught exceptions, leftover informational messages, and bits of debugging noise pile up fast. That happens even in larger companies, I’ve seen it myself.
We use Sentry across everything, web, APIs, mobile apps, and internal tools. Critical items get addressed right away, and every once in a while we would go through the list, fix the minor stuff, remove leftover notifications, and move back to priorities without ever really leaving it fully clean.
Until now.
The big opportunity right now, and really ever since frontier models became usable for production work, is the ability to describe and build ideal tools that fit your business and workflow. Those tools now ship about as fast as you can describe them.
To tackle the above, we follow a workflow that works specifically for us:
- Issue logged into Sentry
- Webhook fires to an internal tool
- Depending on severity, I might get a Telegram notification asking for approval. Telegram, by the way, is still the absolute best and fastest way to get all kinds of notifications and make ops decisions.
- Codex fires off the process against a mapped repository, fixes the issue, and creates a PR
- PR gets reviewed and approved the next day, tested if necessary outside of the TDD written by the agent, deployed, and closed
That’s all, that’s the post.