{"id":2107,"date":"2026-11-17T06:50:00","date_gmt":"2026-11-17T06:50:00","guid":{"rendered":"https:\/\/devopsschool.org\/blog\/?p=2107"},"modified":"2026-11-20T14:50:00","modified_gmt":"2026-11-20T14:50:00","slug":"aiops-three-paths","status":"publish","type":"post","link":"https:\/\/devopsschool.org\/blog\/aiops-three-paths\/","title":{"rendered":"Rules, Platforms or Learning: Three Paths Into AIOps Compared"},"content":{"rendered":"<p>A learner in one of our evening cohorts once asked a deceptively simple question: &#8220;If I want to get good at AIOps, which route should I take: write my own rules, buy a platform, or learn the machine-learning side?&#8221; The honest answer is that these are three different paths with different strengths, and the smartest people walk parts of all three. This comparison tells the story of each, with evidence from <a href=\"https:\/\/www.xopsschool.com\/aiops\/\">the AIOps entry in the Ops Atlas<\/a>, so you can pick where to start.<\/p>\n<h2>Path one: rules and labels, the craft route<\/h2>\n<p>Our learner began here, and so does the Atlas. Its payoff-ranked list opens with four steps that involve no ML: shared label names, collapsing duplicates, bundling alerts while muting echoes sensibly, and joining alerts to the service map and the change record. The strengths of this path are obvious. It is cheap, it works with the tools you already have, and every decision is explainable to a colleague at three in the morning.<\/p>\n<p>The weaknesses are real too. Hand-written rules need maintenance, and they can hide traps. One runbook cited by the Atlas used the word cluster for two ideas at once, the affected cluster and the checking server, so a muting rule ended up hiding a real outage. Rules also cannot describe everything, such as traffic that moves with the week and the season.<\/p>\n<p>Skills you gain: label design, query writing, testing for missing data, reading dependency maps. Those habits overlap with the ones we encourage in <a href=\"https:\/\/devopsschool.org\/blog\/devops-problem-solving-how-to-build-critical-thinking-skills\/\">our write-up on sharpening analytical thinking for DevOps work<\/a>.<\/p>\n<h2>Path two: the platform route<\/h2>\n<p>The second path is to adopt a commercial or open-source platform that bundles correlation, routing and dashboards. The upside is speed and breadth. The Atlas points to a research firm&#8217;s 2025 evaluation of AIOps platforms that looked at ten providers on twenty-six criteria, which gives a sense of how crowded the field is. Features vary: one vendor caps a correlated incident at three hundred alerts, another builds fault trees over its own service map, another suggests causes from a list of four types.<\/p>\n<p>The downside is dependence. The Atlas catalogues a run of retirements: a managed anomaly service ended in October 2025, another was withdrawn on 1 October 2026, and a third closed to new customers on 29 October 2026 with existing customers supported until September 2027. Acquisitions have also reshaped products. Its takeaway: hold your rules, labels and scoring set in forms you own, so changing engines never means a restart.<\/p>\n<p>Skills you gain: vendor evaluation, integration, change management. Avoid memorising one product&#8217;s menus, though; our <a href=\"https:\/\/devopsschool.org\/blog\/devops-for-it-transformation-practical-skills-roadmap-and-strategy\/\">guide to the skills and strategy behind IT transformation<\/a> argues for transferable depth over tool-specific tricks.<\/p>\n<h2>Path three: the learning route<\/h2>\n<p>The third path is anomaly detection, learned grouping, cause ranking and, increasingly, language-model assistants. It is the most exciting and the least settled. The Atlas collects sobering research. Academic results are mixed. One 2022 paper found neural log detectors unreliable at times, while a 2024 one showed that basic detectors counting template frequency held their own on well-known test sets. Another team compared 21 graph methods for tracing causes in microservices and found no clear winner across scenarios. As for agents, a benchmark gave models from early 2025 only an 11.4% success rate on 102 operations tasks, and the top diagnostic mark on a later leaderboard sat near 56%.<\/p>\n<p>Skills you gain: statistics, evaluation, experiment design and, crucially, scepticism. Before any learned grouping goes near production, it wants roughly 100 hand-labelled past incidents to test against, and it asks that you move every learned policy along a lifecycle of proposal, silent scoring, a canary on one or two services, live use, review and retirement, agreeing rollback conditions at the start.<\/p>\n<h2>What each path teaches about people<\/h2>\n<p>Technology choices are also choices about teamwork. The rules path forces conversations about ownership: who exactly is responsible for this service, and what should they do when it alerts? The platform path forces conversations about budget, procurement and support contracts. The learning path forces conversations about evidence, because nobody should trust a model whose accuracy nobody has measured.<\/p>\n<p>Mentors in our cohorts often say the best engineers can move between those conversations without friction. They can sit with a service owner and argue about labels in the morning, read a vendor contract in the afternoon and review a precision table in the evening.<\/p>\n<h2>A side-by-side look<\/h2>\n<ul>\n<li><strong>Cost to start:<\/strong> rules are cheapest; platforms vary from free to expensive; learning needs data and time.<\/li>\n<li><strong>Explainability:<\/strong> rules are the clearest; platforms mix rules and models; learned methods are hardest to explain.<\/li>\n<li><strong>Risk of silent failure:<\/strong> highest where labels are ambiguous or calendars are missing, whichever path you take.<\/li>\n<li><strong>Portability:<\/strong> best for rules you own; weakest for engines tied to one vendor.<\/li>\n<li><strong>Time to first win:<\/strong> days for rules and labels; weeks for a platform trial; months for anything learned.<\/li>\n<\/ul>\n<h2>What the learner chose<\/h2>\n<p>Before the cohort ended she wrote a short plan on a single page, and it is worth borrowing.<\/p>\n<ol>\n<li>Fix names and owners first; measure compliance weekly until it stops being embarrassing.<\/li>\n<li>Trial any platform against my own labelled incidents, never against its demo data.<\/li>\n<li>Add a learned component only where a rule cannot reach, and only after a quiet trial.<\/li>\n<li>Write down every automation limit before writing the automation.<\/li>\n<\/ol>\n<p>Four lines, none of them glamorous, and every one of them survived contact with a real on-call rota.<\/p>\n<p>Our student started with path one and kept a notebook. In month one she wrote a label contract and measured compliance. In month two she added de-duplication and noticed that an instance name inside the fingerprint stopped repeats collapsing. In month three she ran a platform trial with a hundred sample alerts and scored its grouping against her hand-labelled incidents using precision and recall. The platform&#8217;s compression looked better, but her scoring showed it merged two unrelated problems. A compression ratio, as the Atlas puts it, is a vanity metric because wrong merges cost as much as misses.<\/p>\n<p>By month four she had experimented with a seasonal baseline, rediscovered a calendar lesson from the Atlas lab, where an undeclared holiday produced 32 of its detector&#8217;s 34 false alerts, and written one guarded automation with strict limits. When a hiring manager asked about her AIOps experience, she did not recite product names. She showed a notebook of measurements.<\/p>\n<h2>Our advice for certification seekers<\/h2>\n<p>If you are weighing certifications, treat them as a map, not the territory. Whichever credential you pursue, build your own small lab so you can speak from experience. Use the first path to understand the problem, the second to understand the market, and the third to understand the limits of the current science. A security thread runs through all three paths as well. Alerting interfaces and automation webhooks must be authenticated, repair runners should avoid root, and any model that reads alert text is reading content an outsider may have written.<\/p>\n<p>No single route wins outright, and that is a fair summary of the whole field. A learner who can compare all three, measure honestly and keep humans in charge of the risky steps is the kind of engineer teams want to hire.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>A learner&#8217;s story comparing three routes into AIOps, rules and labels, platforms, and learned methods, with their strengths, risks and skills gained.<\/p>\n","protected":false},"author":5,"featured_media":2108,"comment_status":"","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[489],"tags":[],"class_list":["post-2107","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-devops-careers"],"_links":{"self":[{"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/posts\/2107","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/users\/5"}],"replies":[{"embeddable":true,"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/comments?post=2107"}],"version-history":[{"count":1,"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/posts\/2107\/revisions"}],"predecessor-version":[{"id":2251,"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/posts\/2107\/revisions\/2251"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/media\/2108"}],"wp:attachment":[{"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/media?parent=2107"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/categories?post=2107"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devopsschool.org\/blog\/wp-json\/wp\/v2\/tags?post=2107"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}