Incident review
Timing…
Customer scope…
Root cause…
Mitigation…
Tone…
0 of 5
custom evals
Every agent is graded against what “good” means at your company.
Backed by
Combinator
Team agents for
every team
For companies that want to 10x, Echo agents learn the
work, do it end to end, and scale it
Incident Response
PagerDuty
Triggered - [prod] API failures
Reason: Customer failure threshold exceeded.
On-call: Echo, Manuel
EchoAgent
🔴 Impact: 2 Priority customers
• Kestrel Bank: 98% of requests failing with 503s in the last 30m
• Juniper Health: 92% of requests failing with 503s in the last 30m
Investigating…
EchoAgent
Root cause: This morning's deploy (PR #4821) shrank the database connection pool from 200 to 20. Requests are queuing for a connection and timing out.Suggested mitigations:
1. Roll back #4821, about four minutes, or
2. Raise the pool size and redeploy, about 20m.
MA
Manuel
@Echo perfect, roll it back
EchoAgent
Resolved. Rolled back #4821 and the error rate is down to 0.2%. I'll keep watching for the rest of the day and work out what #4821 needs before it ships again.
works where
you work
Slack, email, Linear, Codex: Echo joins the tools your team already uses. It watches how work moves forward there, then picks up the next step the way your team would.
- Slack
Linear
GitHub
Gmail
Notion
Google Calendar
Google Sheets
Google Docs
Google Drive
Fathom
Granola
Ashby
Conductor
Claude
OpenAI
# incidentsEchoAgent
ENG-248 is open. @Claire can you review?
Checkout errors this morning?Kestrel BankDana Ruiz We saw a wave of failed checkouts around 9:15. Is this on your side?
Claire Draft by EchoSendHi Dana, sorry about the trouble. The API is back, and the calls that failed from 9:12 to 9:41 were retried automatically.
ENG-248In Progress🚨
Echo Fix merged, error rate back to baseline. Closing.
team agents,
not personal ones
Personal agents keep what they learn to themselves. Echo agents belong to your team: anyone can teach them, everyone benefits, and the know-how stays with the company.
# engMA
Manuel
@Echo error rates spiked after the deploy. Can you triage?
EchoAgent
Timing points to this morning's deploy. Rollback PR is up, ENG-248 opened.
CL
Claire
FYI that deploy cut the connection pool to 20, that's what tipped it over.
EchoAgent
Noted. Added it to the runbook so the next on-call sees it.
JE
Jessica
@Echo did any customers notice the checkout errors on Monday?
EchoAgent
Two: Kestrel Bank and Juniper Health, for about 30 minutes. @Claire's fix is in. Want a note for your calls?
bespoke to your
company
Your team's way of working is your edge. Echo learns your playbooks, priorities and voice, not a generic version of the job.
LA
Lauren
@Echo from now on, loop finance in on anything over $20k before it goes out.
EchoAgent
Done. Added it to the deal desk runbook. It applies to everyone in Sales from now on.
Deal desk agent · LumenUpdated · shared with Sales
TonePlain. No hedging. One question at a time.
ApprovalsFinance signs anything over $20k.
Runbookdeal-desk-v4.md · edited by Lauren
ToolsSheets, Gmail, Drive, HubSpot
ModelClaude, with OpenAI for forecasts
gets better every day
Proposed change
Timing failed on 2 runs this week
Drafting a fix…Post the first update in #engJoin #triage and post there when an alert fires
ApproveNot yet
self-improvement
It scores its own work and fixes its own misses.
Learned from your team
Prefill queue caps at 500From Priya's edit to a summary
Finance signs anything over $20kMaya told it in #deal-desk
Join #triage when an alert firesPriya approved the fix
Skip the weekend digestIgnored four weekends running
continual learning
It learns by watching what your team edits, approves and ignores. You can tell it too, but you don't have to.