Senior DevOps engineers owning multi-cluster fleets are buried in Terraform, runbooks and SOC2 evidence. Bricolage writes infrastructure code, optimises Dockerfiles, consolidates pipelines and produces audit-grade runbooks while you keep production decisions.
Agent cast
Watch it happen
The team
Forty microservices across dev, staging, prod. Terraform IaC. GitHub Actions pipelines. Eighty-plus container images in ECR. Prometheus + Grafana + PagerDuty.
Plus internal tooling, developer support, security compliance and an endless queue of "my deployment is broken" questions from forty-five engineers. One person doing the work of three.
Idle
The chat is empty until the user types.
Step 01 · Runbooks
Cluster operations, deployment procedures, incident response, database operations, CI/CD troubleshooting, container registry, monitoring and alerting, and security operations.
Each references actual infrastructure: real cluster names, real namespace conventions, real Helm chart paths, real AWS commands specific to this company's setup. Judgment calls (rollback thresholds, escalation tiers) flagged for review rather than guessed.
Step 02 · Developer ops
When a developer's deploy failed, the system traced the root cause: the CI pipeline tagged images with git SHA but the Helm values specified a semantic version tag — a classic mismatch.
The fix was coded (update the pipeline to tag with both), and a developer troubleshooting guide was created with exact kubectl and AWS CLI commands for ImagePullBackOff, CrashLoopBackOff, no traffic, slow deployments. Next time a developer asks, they can be pointed to the guide.
Step 03 · Image audit
Five services over 300MB each. Twelve services running as root containers (a SOC2 finding). Eighteen services missing health checks. Three pinning to :latest.
Optimized Dockerfiles cut the largest image from 380MB to 120MB — sixty-two percent reduction. Multi-stage builds, non-root users, health checks, distroless bases applied across the portfolio. CI pipeline savings: forty-five minutes per full run.
Step 04 · CI/CD consolidation
Before: forty services with copy-pasted GitHub Actions workflows, 3,200 lines of duplicated YAML. After: one template and forty callers (five lines each), 320 lines total — ninety percent reduction.
When the system needs to add vulnerability scanning, it goes into the template once and applies to all forty services automatically.
Step 05 · Incident
Critical service throwing 502s. Structured diagnostics: pod logs, pod events, resource usage versus limits, correlation with recent deploy.
Diagnosis: OOMKilled, memory limit exceeded. Emergency fix coded (kubectl patch), postmortem template generated, service's facts updated with incident details for future reference. The platform engineer made the decision; the system provided the analysis.
The outcome
Security posture assessment identified six compliance gaps. Three remediated immediately (PodSecurity policies, network egress, audit logging). Three more with remediation plans. A security controls matrix mapped infrastructure controls to SOC2 criteria — audit prep that normally takes a week, structured in two hours.
Runbook writing: three days → twenty-five minutes. Dockerfile optimization: two weeks → two hours. CI/CD template: one week → thirty minutes. Developer support: one hour/day → ten minutes/day. Sixty to eighty hours per month back for architecture, performance and scaling work that requires human judgment.
Up next
A teammate that understands your codebase and ships complete features.
