Your agents should never make the same mistake twice.
Joined December 2025
- Tweets24
- Following13
- Followers153
- Likes47
moda retweeted
Issue 4 of our weekly update is out. This week is all about the dashboard.
- A Home page you can customize
- Choose the unit for each signal rate
- Traces and search now show user tasks
- User health across conversations
- Problems has its own page
Full write-up below
moda retweeted
check it out!!
Moda now files problems straight into @linear.
One click (or auto-sync) and the issue arrives with the cause, the impact, and evidence linked to traces.
Close it and Moda watches for it to come back. If it does, the issue reopens with the new evidence.
Full write up in the comments
moda retweeted
we now fit in better to your engineering stack, take a look
Moda now files problems straight into @linear.
One click (or auto-sync) and the issue arrives with the cause, the impact, and evidence linked to traces.
Close it and Moda watches for it to come back. If it does, the issue reopens with the new evidence.
Full write up in the comments
moda retweeted
you can now track pr regressions right in linear
Moda now files problems straight into @linear.
One click (or auto-sync) and the issue arrives with the cause, the impact, and evidence linked to traces.
Close it and Moda watches for it to come back. If it does, the issue reopens with the new evidence.
Full write up in the comments
Moda now files problems straight into @linear.
One click (or auto-sync) and the issue arrives with the cause, the impact, and evidence linked to traces.
Close it and Moda watches for it to come back. If it does, the issue reopens with the new evidence.
Full write up in the comments
Issue 3 of our weekly update is out:
- Open any Moda finding in Cursor, Claude Code or Codex
- Alerts get their own page, now with email
- A new custom signal builder you can test on past traces
- Fix checks: link a signal to Linear and see if the fix worked
- Clearer problem evidence
Full write-up below
Moda now lives in Slack.
Issue 2 of our weekly update is out.
Last week was about finding problems. This week is about what happens after you find one.
In this issue:
- Send a problem to Linear in one click, then tag Moda in Linear to watch the fix
- Our frustration detection now outperforms Opus 5.5 at 1/30 of the cost
- Custom signals are live for every team
- Mark any detection as a false positive, and Moda learns from it
- Opening a trace now highlights every linked message
Read more: moda.dev/weekly/002
We're starting weekly updates on the Moda blog to share what we're shipping and learning.
In our first issue:
- Jev helps us detect more specific signals
- Full conversation analysis catches problems that a single scan can miss
- We rebuilt the dashboard and trace views to make problems easier to investigate
- You can now ask questions about alerts directly in Slack
- Import your Braintrust history and find signals in past traces
Read more: moda.dev/weekly/001
moda retweeted
Here's how recursive language models had a real business impact for us:
At Moda we process millions of LLM chats daily and cluster them into use cases to help companies understand what users are doing and where their AI is failing.
Previously we tried the obvious approach: analyze each segment in each conversation, propagate back up. Turns out aggregation is lossy, roll up 10k summaries and all the sharp insights get averaged into mush.
RLMs flip it. The LLM explores the full dataset programmatically and recurses into what matters. Top-down with selective depth beats bottom-up aggregation.
Shoutout to @lateinteraction & @a1zhang for such a great paper