AI DevOps Engineer
Southfield, MI, United States3w ago
Key details
- Location
- Southfield, MI, United States
- Work style
- Onsite
- Level
- Senior
- Job type
- Full-time
- Experience
- 6–8 yrs
- Function
- DevOps / Infrastructure
- Focus area
- AI / ML
- Posted
- 3w ago
Skills
ai agentsawscapacity planningcdkci/cdcloudwatchcost accountingcost controldatabricksdatadogdevopsdistributed tracingdockergithub actionsgitlab cigrafanaincident responseinfrastructure as codekuberneteslakehouselangfuselangsmithllmmodel context protocolobservabilityopentelemetryprometheusservice level objectivessreterraformtest automation
About the role
POSITION SUMMARY:
We are looking for an AI DevOps Engineer to own how our client’s AI systems get deployed, stay up,
and stay observable. You will own the infrastructure and operational substrate for our client’s
agentic platform — the CI/CD pipelines, the cloud environments, the release machinery, the
reliability practice, and the telemetry layer that turns opaque agent behavior into something you can
measure, alert on, and debug.
Our client’s engineers build the AI agents; you make shipping them safe, repeatable, and observable.
Right now, most of their apps have no CI/CD, the infrastructure is fragmented, and the agentic
systems being stood up have little more than print statements for observability. Building that
operational foundation is the job.
Agentic systems fail differently from ordinary services — nondeterministic output, silent quality drift,
runaway tool-call loops, and cost that spikes without warning. Standard DevOps is necessary but not
sufficient. This role exists because someone has to own reliability and telemetry for systems that
don’t fail the way the runbooks assume.
KEY JOB RESPONSIBILITIES:
Own CI/CD for the platform. Build the pipelines that take AI systems from commit to production —
automated testing, evaluation gates, security and dependency checks, controlled and canary
releases, and one-command rollback. Most of their apps have no pipeline today; you will establish
the pattern and make the safe path the default path.
Manage cloud infrastructure as code. Own the cloud environments (AWS primarily) the platform
runs on — provisioning, networking, secrets, environment parity, and cost controls — as versioned,
reviewable infrastructure-as-code, not hand-tuned consoles. You are accountable for environments
that are reproducible, least-privilege by default, and cheap to stand up and tear down.
Run the reliability practice. Own production reliability: SLOs, on-call and incident response,
capacity and cost management, self-healing loops that detect and recover from failures, and
blameless post-incident review. You will help define what “up” and “healthy” even mean for a
nondeterministic system.
Build the agent telemetry and observability layer. Instrument the platform so agent behavior is
legible: structured traces of agent runs and tool calls, token and cost accounting, latency and
success metrics, output-quality tracking over time, and the dashboards and alerts that surface a
regression before a user does. When an agent misbehaves in production, the telemetry you built is
how the team finds out and figures out why.
Set the operational standard by example. On a small, high-leverage team, your pipelines and
dashboards are the template. You establish the deployment patterns others adopt, the observability
every new system gets wired into by default, and the operational discipline that lets a lean team run
production systems well.
REQUIRED KNOWLEDGE, EDUCATION, EXPERIENCE, SKILLS, AND ABILITIES:
- 6–8 years in DevOps, SRE, platform, or infrastructure engineering, with a track record of
running production systems you were accountable for
- Deep CI/CD experience — you have built and owned pipelines (GitHub Actions, GitLab CI, or
similar) that gate, test, and safely release real production software
- Strong cloud operations, ideally AWS — provisioning, networking, secrets, and cost
management as infrastructure-as-code (Terraform, CDK, or similar)
- Hands-on observability experience — metrics, logging, distributed tracing, dashboards, and
alerting (OpenTelemetry, Prometheus/Grafana, Datadog, CloudWatch, or similar) — and the
instinct to instrument first
• SRE fundamentals: SLOs, incident response, on-call, capacity planning, and blameless
postmortems
NICE TO HAVE:
- Experience operating AI or LLM systems in production — token and cost accounting, prompt
and evaluation-score tracking, or LLM observability tooling (LangSmith, Langfuse, Arize, or
similar)
- Familiarity with the failure modes of nondeterministic systems: quality drift, runaway loops,
cost spikes, non-reproducible output
- Experience with Databricks or a similar lakehouse platform, and with tool-integration layers
such as MCP
- Container and orchestration experience (Docker, Kubernetes, or serverless equivalents)
- Experience in a non-software-company engineering organization — internal tools, corporate
IT transformation, or similar
- Experience standing up an internal platform or golden-path deployment pattern that other
teams adopted
WHY THIS ROLE:
You will build the operational foundation an entire organization’s AI runs on — the pipelines, the
environments, and the telemetry — with a clear mandate and a direct line to the Director of APEX.
The platform is early, and that is the appeal. You are not tuning someone else’s mature platform;
you are building the deployment and observability substrate our client will run AI on for the next
decade and defining what running AI in production looks like there.
How well do you match this job?
Skills
Experience
Level
Title
Location
Salary
Freshness
Similarity
Free account needed to apply or save. It takes a minute, and we'll bring you right back to this job.