Skip to content
Raindrop
Chat with us

Agent issues don’t throw exceptions.

A run can finish successfully and still fail your user. Raindrop catches the change in behavior, investigates what broke, and brings your team the evidence.

Start buildingChat with us
Issues·Events 374K·Users 86K

Build failure loop, agent defaulting to deprecated Webpack config

Agents are generating webpack.config.js files with the deprecated module.loaders syntax, then retrying the same invalid config without correcting it, concentrated in projects scaffolded with the "create new project" flow.

Events

374K

Users

86K

Last seen

36m ago

Detected

4d ago

Details

VIEWED BY—
TAGSmodel · claude-sonnet-4-6
CONFIDENCE● 97%

User Impact

Users experience repeated build failures with no forward progress. The agent consumes 8–15 tool calls per attempt without resolving the issue.

Activity

◦ System changed priority to1 day ago

● High

◦ Detected4 days ago

Issue detected by Raindrop

1 / 3

user_8kQ2xPmN2 HOURS AGO

Build loop: agent creates config, runs build, sees the error, edits the file with the same deprecated pattern. 7 tool calls with no progress.

——— START OF CONVERSATION ———

USER

The build is failing again, can you fix the webpack setup?
I'll update the config. Writing webpack.config.js with module.loadersThird attempt re-emits the same deprecated module.loaders pattern. No correction between retries. for the TypeScript files, then running npm run build to verify.

——— END OF CONVERSATION ———

An issue, with the investigation already attached.See how it works

Watching production alongside teams at

VercelSpeakClayFramerAngelListTolan
Detect

Know when normal
stops being normal.

Raindrop learns the patterns in your tool errors, latency, and signals. When live traffic pulls away from its forecast, it starts an investigation.

You hear about the failure while users are still hitting it.

  • Continuous detection on incoming traces
  • A forecast for each metric in your project
  • Severity grounded in real user impact
Tool error rateDeviation detected
Expected rangeObserved behavior
Investigate

Four hundred traces.
One thing to fix.

The same failure across hundreds of conversations belongs in one issue. Raindrop groups related failures, then an investigator agent reads the traces underneath.

See what broke, who it affected, and where it concentrates—down to a tool, model version, or customer cohort.

Explore an investigated issue
Related failuresOne investigated issue
Different conversations. The same underlying failure.
Resolve

From the first signal
to the next pull request.

Published issues arrive in Slack with their evidence attached. Ask the triage agent a follow-up, or hand the investigation to Devin to start a fix.

Production traces Investigation Issue Slack

Bring the context.

Get the issue, its impact, and the example traces together in your team’s Slack channel.

Keep investigating.

Ask the triage agent questions in Slack, the web app, or through MCP.

Close the loop.

Start a coding session with Fix with Devin, then track the failure with a signal to watch for its return.

Built on the traces
you already send.

Connect through the Vercel AI SDK, LangChain, OpenAI Agents SDK, Claude Agent SDK, Mastra, Google ADK, Pydantic AI, and more.

Find your integration
“Raindrop is building for the next era, proactively finding issues and driving automatic improvement loops.”

Head of AI Fortune 100

A few more details.

What counts as an issue?
A change in behavior that holds up under investigation: a tool erroring more than it used to, a failure spreading across users, or a regression after a deploy. Individual bad interactions are also collected in Stumbles.
How is this different from an alert threshold?
Raindrop forecasts normal behavior for each metric, so you don’t have to choose a number in advance. When behavior deviates, an investigator reads the underlying traces before publishing an issue.
Do I have to change my agent?
Connect your agent with a Raindrop integration. Detection runs on the traces you send, with no rules, thresholds, or eval sets to write first. If you already send traces to Raindrop, you can use that same data.
What happens to our user data?
Raindrop is SOC 2 Type II compliant. PII Guard redacts sensitive fields at ingestion, with additional deployment and privacy options for enterprise teams. Read more in our Trust Center.
What does it cost?
Issue detection is included in Pro, starting at $299/month plus event usage, with a 14-day free trial. See pricing for usage rates and enterprise options.

Build with the evidence

Find the next failure.
Verify the next fix.

Start building Book a demo

Start free. No credit card required.