Dhrumil

Building AI Products at @Sentry

[Dhrumil @dhrumilPM]

https://t.co/UMaUd5vCoK

[bolt.new @boltdotnew]

Your users might be hitting errors you never see.

Turn on the @sentry connector in https://t.co/7HmmEBwGQU, then ask: "What errors have happened in the last 24 hours? How can we fix these?"

https://t.co/7HmmEBwGQU pulls the real production errors and fixes them in the same chat.

[Dhrumil @dhrumilPM]

@i_mika_el @sentry We require code reviews on all main Sentry monorepo PRs -but we still find that there are many long tail bugs from edge cases that Seer can easily one shot

Aim to publish some numbers in the future

[Dhrumil @dhrumilPM]

@i_mika_el @sentry Def depends on the stage of the codebase. For e.g in our CLI - Seer goes and opens patches that the team could often just stamp

https://t.co/akiHiPdp5b

[Dhrumil @dhrumilPM]

Every app has a long tail of bugs. Seer is quietly clearing them from the @sentry CLI — 11 last week.

First to fall: sentry alert issues view crashed with a TypeError when a rule had no conditions. Seer shipped the guard.
→ https://t.co/JKHQYBeHaR

[Kent C. Dodds 🐨 @kentcdodds]

@kodykoala @sentry @cursor_ai Let's dive into this together! https://t.co/X4T291DVDE @sergical ❤️

[Dhrumil @dhrumilPM]

@kentcdodds @kodykoala @sentry @cursor_ai Seer🤝Kody

It was great to see Kody build this integration itself 🙌

[Dhrumil @dhrumilPM]

Don't remove humans from these AI loops - you don't want more problems. Instead help your devs review and fix faster.

This was a fun and easy Claude routine to build using our Seer Agent and all done with Slack as the common surface

[Sentry @sentry]

Everyone wants to spend less time fixing bugs, including us.

So we built an automated debugging workflow using our Seer agent in Slack, Claude routines, and Seer autofix.

Check out the full blog by @dhrumilPM to see how we did it 👇 https://t.co/GCZfg140TI

[Dhrumil @dhrumilPM]

@jasonkneen As we build our agent, we’ve found tracing useful. Once you have 1,000s of users, having an agent look through every trace isn’t very efficient. Tracing helps Claude find the right traces to evaluate behavior.
Logs can work too, but likely need some skill work to guide agents.

[Dhrumil @dhrumilPM]

@oldstackjournal Def think Agents+ MCP can handle this for us now as long as we have rich traces

https://t.co/r1SSu5wXJ6

[Dhrumil @dhrumilPM]

Agent Monitoring + MCP with @sentry = an LLM judge for your agent in 5 minutes.

I set up a @claudeai routine that reads the messages users send our AI agent, categorizes them, then reads the full conversations to see how the agent did — and reports to Slack.

You just need the right telemetry and a way for the agent to fetch it. Sentry's got you on both

[Dhrumil @dhrumilPM]

@colemurray 💯 Agree. This is what we’re seeing as we build Seer Agent and our MCP at @sentry.

Having the right context about app problems- where/why/when/what params etc lets agents fix better and efficiently

[Dhrumil @dhrumilPM]

*Paid customers*

[Milin Desai @virtualmilin]

We exceeded our annual paid users goal just halfway through the year.

Turns out the Sentry concept of issues with context that help debug and fix faster is even more valuable as the world builds more software (humans and agents).

Sentry is for EVERY builder out there. The Fortune 500,0000.
- The tattoo parlor getting its first booking site online.
- A YC startup shipping multiple times a day.
- An F500 team rebuilding on a modern stack.

🙏 @sentry community.

Job not finished.

[Sentry @sentry]

Your agent works great... until it doesn't.

Then you're staring at a black box wondering why it hallucinated a tool call, burned 40k tokens, or just silently died.

Agent Tracing fixes that. Here's what it looks like in 3 different scenarios 🧵

[Dhrumil @dhrumilPM]

I know many devs like Codex. Tried it recently and gave up after a day. As a PM, Claude Opus feels like a Staff Eng who works with me to build cool stuff

Codex went and built crap without planning well or asking enough questions - More work for me to prompt vs collaborate 🤷‍♂️

[Dhrumil @dhrumilPM]

Aka - you can search for logs directly related to that slow db call

Or you can see all the logs for that particularly expensive generative AI call

And if you ask Seer - it just cross queries for you as it investigates

[Sentry @sentry]

Did you know you can cross reference logs when you're searching your traces?

We shipped this and a bunch of other cool highly-requested enhancements for Logs this summer 👀

Check it out here: https://t.co/tH0VLYVwDa

[Dhrumil @dhrumilPM]

I think Claude is done with my PM skills right now 😂

[Dhrumil @dhrumilPM]

@dvsj_in @sentry Just curious- were you trying to query errors or logs cz sentry has both now 😅

But tags is applicable to errors only

[Dhrumil @dhrumilPM]

Getting a first hand look at some greats devs use AI to code and it's mostly saying 'no, not like that' and 'okay good job buddy' in a slightly strained messages.

[Dhrumil @dhrumilPM]

@jrowe6720 @sentry Def agree with this - this is how we want seer to work too

[Dhrumil @dhrumilPM]

@jrowe6720 @bcherny @sentry Love to see this ! Any feedback abt the MCP or anything we could do better in the product itself?

[Dhrumil @dhrumilPM]

Code mode's having a moment (some more litigiously than others). We're building it into Seer @sentry. It's faster, cheaper & acts for you.

Pointed it at 54 JS issues → triaged them all, clustered dupes, archived the 20 noisy ones with a comment each. Shipping soon 😇