Post as thread on @avivl account.
I taught my AI assistant to do SRE.
Not just "summarize logs" — actual incident response, alert analysis, and engineering metrics.
Here's what an AI-powered operations toolkit looks like 🧵
1/11
Skill #1: Incident Response
My Clawdbot can now: • Check production health across 67 Cloud Run services • Correlate alerts with recent deploys • Suggest rollbacks or scaling fixes • Follow actual runbooks, not hallucinate them
2/11
The key: structured diagnostics.
It knows to check:
- Recent deployments (GitHub Actions)
- Error logs (Cloud Logging)
- Service metrics (latency, memory)
- External dependencies (API quotas)
In that order. Like an actual SRE would.
3/11
Skill #2: Alert Insights
Weekly analysis of production alerts: • Scans Gmail for alert patterns • Identifies noisy/flapping alerts • Cross-references with monitoring config • Recommends specific threshold changes
Turns alert fatigue into actionable PRs.
4/11
The magic: it reads our actual infra code.
Points to specific files: "Adjust error threshold in src/core/services/monitoring/error-reporting.ts line 47"
Not generic advice. Real code changes.
5/11
Skill #3: DORA Metrics
Tracks the 4 key DevOps metrics: • Deployment Frequency • Lead Time for Changes • Change Failure Rate • MTTR
Weekly reports with trends and per-service breakdowns.
6/11
Data sources it pulls from:
• GitHub Actions → deploy frequency, failure rate • GitHub PRs → lead time (created → merged) • Gmail alerts → MTTR (alert → resolved)
All automated. No manual spreadsheets.
7/11
The pattern: Skills = Runbooks as Code
Each skill is: • A markdown file with procedures • CLI commands it can run • Context about our specific infra
AI follows the runbook. Humans review the output.
8/11
What changed for us:
Before: Wake up to 47 alerts, spend 30min triaging After: "Golem, what happened overnight?" → 2min summary
Before: Monthly DORA review (if we remembered) After: Weekly automated report in my inbox
9/11
The meta insight:
SRE is mostly pattern matching + executing known procedures.
That's exactly what AI is good at.
Humans should design the runbooks and make judgment calls. AI should execute the checklist.
10/11
All of this runs on a $34/month GCP VM.
Skills are just markdown files. No fancy infra needed.
Your AI assistant can be your junior SRE — if you teach it how.
#SRE #DevOps #AI #Clawdbot #PlatformEngineering
11/11