Justy: Exploring Next, episode 335. Sentry’s new Seer Agent is basically trying to turn, uh, production debugging into a normal conversation instead of a scavenger hunt.
Cody: Yeah, and that matters right now because incidents don’t usually feel like one clean problem. It’s alerts, logs, traces, maybe a deploy, maybe not, and everybody’s jumping around trying to piece it together.
Justy: Exactly. If you’re on a team shipping all the time, the pain isn’t just the bug. It’s the time lost figuring out where to even look.
Cody: Seer sits inside Sentry and takes a natural-language prompt. So instead of asking someone to manually browse through issue data, you describe what’s going on and it tries to work across the context it already has.
Justy: So the user story is pretty clear to me. You’re already paying for observability, you’ve already got the app in Sentry, and now you want a faster path from ‘something broke’ to ‘okay, this is probably it.’
Cody: Right. And I think the interesting part is that it’s not just a chatbot pasted on top. The value is in how it can connect the prompt to real production signals. That’s the only way this is useful.
Justy: Who’s the buyer here, though? I’d guess teams that already feel the sting of incident response. Smaller teams might like the idea, but if they only have a few issues a month, the adoption pressure is lower.
Cody: Yeah, and the barrier is trust. If Seer says, ‘this is probably the root cause,’ people need to know whether that’s a solid lead or just a confident guess. I think that’s the whole game.
Justy: And honestly, that’s true for any AI tool in dev workflows. If it saves ten minutes and points you to the right file, great. If it sends you on a weird detour, people stop using it fast.
Cody: What I find clever is the shape of the problem. Debugging is already language-heavy. Engineers write the issue in Slack, in tickets, in notes, and now the tool can meet them there instead of forcing a new interface.
Justy: [chuckles] Which is nice, because nobody wants another dashboard to learn at 2 a.m. when they’re already annoyed.
Cody: The trade-off is obvious, though. The more the agent abstracts, the more important it becomes that it shows its work. I’d want to see what evidence it used, not just a summary.
Justy: Yeah. If it’s going to win people over, it has to feel like a sharp assistant, not a mysterious one.
Cody: For a weekend build, I’d start small. Take a single service’s error logs and stack traces, wire them into a local LLM workflow, and ask it to summarize one incident plus the likely code path.
Justy: And if you want the solo-builder version, do it with one repo and one alert source. No big platform project. Just see if the model can actually save you time on a real mess.
Cody: That’s the test, honestly. Not can it sound smart, but can it help you move faster when the app is on fire and you’re trying to stay calm.
Justy: That’s our read on Seer. We’ll keep poking at what actually helps in the moment, and what just looks clever from far away. See you next time.