A monthly, evidence-led view of the roles and skills employers ask for—from cloud and DevOps to platform engineering, SRE and the emerging AI operations stack.
The live pilot below was collected in 2026-08 from 8 public employer job boards. It is actual data, but remains unaudited and source-panel biased; use it to review the method, not as a final market estimate.
LLM/GenAI appeared in 291 pilot postings and agentic systems in 212. These counts require audit because this source panel includes AI-native employers.
This chapter will measure whether AI creates new operational roles, enters existing roles as a required capability, or changes the responsibilities of the same familiar job titles.
We will report title adoption separately from mentions in job descriptions. A phrase appearing in responsibilities is not evidence that a new occupation exists.
Automate delivery and infrastructure
↓TOWARDBuild governed AI-assisted delivery, secure model access and automate agent workflows
Provide paved roads for application teams
↓TOWARDProvide self-service AI platforms for models, prompts, retrieval, evaluation and inference
Protect availability and latency of services
↓TOWARDDefine reliability for probabilistic systems: quality, drift, cost, safety and model performance
A specialist role adjacent to DevOps
↓TOWARDAn operational discipline increasingly embedded across platform and reliability teams
Each term is classified as required, preferred, responsibility or incidental mention. We will also report co-occurrence—for example, how often Kubernetes appears with vLLM, or SRE appears with model evaluation.
AI, machine learning, generative AI, transformers, embeddings
MLOps, model registry, feature store, experiment tracking, drift
LLMOps, prompt management, RAG, vector databases, model gateways
AI agents, agent orchestration, tool use, MCP, multi-agent workflows
GPU scheduling, accelerators, inference serving, Kubernetes operators
evaluation, tracing, hallucination monitoring, guardrails, red teaming
MLflow, Kubeflow, Ray, KServe, vLLM, LangChain, LlamaIndex
Bedrock, SageMaker, Vertex AI, Azure AI, OpenAI APIs
Are MLOps and LLMOps growing as standalone titles?
Which DevOps roles now require AI or LLM knowledge?
Is agentic DevOps language moving from experiments to hiring requirements?
Which traditional skills remain prerequisites for AI operations?
How do AI requirements differ by seniority, geography and industry?
Which tools are durable signals versus short-lived product mentions?
Every audited monthly snapshot will remain frozen and comparable. The site will show direction, velocity and persistence—not just a fresh ranking that erases last month.
DevOps, Platform, SRE, Cloud, MLOps, LLMOps and AI Platform as a share of the monthly relevant-job sample.
Month-over-month change in skill penetration, with minimum sample and confidence thresholds before a trend is called.
AI terms in titles versus requirements, responsibilities and preferred qualifications—kept as separate series.
Entry, mid, senior and lead demand over time, including the skill bundle associated with each level.
These editions provide useful context, but they will not be drawn as one continuous time series unless their samples and taxonomies can be reconciled. They remain available as source material and historical snapshots.
Historical evidence remains useful when each claim is matched to the strongest comparison the underlying data can support.
Strictly comparable snapshots using a stable core source panel, query set and taxonomy. Appropriate for numeric change, momentum and confidence intervals.
Older raw data—especially 2023—is reprocessed with the current taxonomy. Comparable fields become benchmark points, not an invented smooth line.
AI summarizes changes in titles, responsibilities and language across archived reports. Every synthesis links to evidence and is human-reviewed.
Comparable data supports a quantitative claim.
Evidence suggests movement, but coverage changed.
Useful qualitative evidence; not a direct numerical comparison.
Percentages alone hide what candidates need. Every monthly edition should split demand by seniority and distinguish required skills from preferred ones.
Linux, Git, cloud fundamentals, scripting
IaC, containers, CI/CD, observability
platform design, SLOs, security, cost
strategy, developer experience, governance
Your original principle remains the anchor: companies reveal demand in their job descriptions. The 2026 edition adds reproducibility, deduplication, contextual classification and an explicit QA layer.
Capture title, description, company, location, date, salary and source from approved feeds and public career pages.
Canonicalize locations and companies, strip boilerplate, detect reposts, and keep one record per real opening.
Separate role identity from skills mentioned. Assign role family, level, domain, work mode and industry with confidence.
Match a versioned skills dictionary, then use contextual extraction for requirements, preferences and responsibilities.
Human-review a stratified sample, measure precision and recall, document taxonomy changes and publish confidence bands.
Freeze a monthly snapshot, compute weighted trends, publish the living site and preserve every prior edition.
A skill counts once per posting. We retain frequency, but also record whether it appears in the title, required qualifications, preferred qualifications or responsibilities. This prevents long descriptions and repeated boilerplate from dominating the report.
We use job descriptions as evidence of stated employer demand, based on the working assumption that organizations generally describe the roles and capabilities they believe they need. We do not assume every description is precise or correct. The report preserves ambiguity, distinguishes explicit requirements from inferred responsibilities, and treats findings as signals—not a perfect description of work.
Version the dictionary monthly. Never rewrite history: each snapshot keeps the taxonomy version used to produce it.
Canonical family plus the employer’s exact title.
Responsibilities and operating practices, independent of tools.
Tools, platforms, languages, frameworks and concepts.
Freeze scope, sources, role families, skills dictionary, sampling rules and QA targets.
Collect the first 2026 sample, calibrate classifiers and manually audit a stratified 10% sample.
Release findings, methodology, downloadable tables and a known-limitations note.
Ingest, dedupe, classify, review exceptions, freeze snapshot and show change versus prior month.