TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
AI SRE tools are gaining attention as companies report more deployments and on-call pages, and teams look for ways to speed up incident investigation. The tools are increasingly used to gather evidence and triage alerts, but hands-off diagnosis and production fixes remain debated, with accuracy and impact varying by company.
AI SRE tools are moving from conference demonstrations into real incident-response workflows, as companies face more deployments and on-call pages, according to a report by The Pragmatic Engineer. The tools are mainly being used to investigate incidents and triage alerts; allowing an agent to change production systems without human approval remains contested, and the report does not establish that fully autonomous incident response is broadly in use.
The report describes interest from several directions: OpenAI built an internal incident-response bot called Sevbot, an engineer at ING said the bank was developing an in-house AI SRE, and vendors including Rootly, incident.io and ResolveAI are promoting incident-management products. The publication says it spoke with those three vendors. Its author disclosed an investment in incident.io and said they were not paid to mention any company.
In practice, the category is less about replacing site reliability engineers than helping people investigate alerts faster. The report says agents commonly work in Slack or Microsoft Teams, where engineers can steer investigations. Systems may fan out work to several agents and use a context layer—such as a knowledge graph and stored operational information—to make findings more relevant. Some companies are building these investigation agents themselves rather than buying a vendor product.
Companies are also trying methods to check agent output, including having separate agents challenge an initial hypothesis, combining models, and running evaluations or backtests against past incidents. But the report cautions that accuracy is difficult to measure: what counts as a correct response depends on an organization’s systems and on the engineering team using the tool. Speed and cost also require tradeoffs, managed through approaches such as model routing, shorter prompts and monitoring cache use.
Why More Pages Are Driving Adoption
The commercial case is tied to a changing operational workload. The report says many technology companies have increased their pull-request frequency, and describes a simple reliability relationship: incident frequency depends on both the rate of changes and the chance that a change causes a failure. If more changes reach production, teams may face more opportunities for incidents even if the risk per change stays steady.
Rootly’s data, as cited in the report, shows pages rising compared with a year earlier and outside-business-hours pages nearly doubling. The report does not provide the underlying counts, sample, or methodology in the supplied material, so those figures should be treated as vendor-reported observations rather than a sector-wide measurement. Still, the trend helps explain why teams are testing tools that can collect evidence, correlate events and filter less important alerts before a human is woken.
If these tools work reliably, they could reduce time spent on routine investigation and spare engineers some unnecessary interruptions. The harder incidents may still land with human responders, however, and an incorrect automated action could compound an outage. The question for operators is not simply whether an agent can produce a plausible diagnosis, but whether it can do so safely and consistently in their own environment.
Products Worth Considering
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Demo to Incident Workflow
Microsoft introduced its Azure SRE Agent at its May 2025 BUILD conference. In the onstage demonstration described by The Pragmatic Engineer, an alert triggered an agent to look for a root cause and ship a fix without waking a human on-call developer. The report’s author found the scenario difficult to imagine teams allowing in production without oversight. The report says the topic then received little attention from the author until more recent examples and vendor activity appeared.
The phrase “AI SRE” has stuck as a label for AI-assisted on-call tooling, although the report says vendors and practitioners find it imprecise: site reliability engineering includes much more than responding to alerts. Alternatives such as automated incident investigation or remediation agents have not taken hold. More importantly, the label can blur a distinction between assisting an engineer and making a production change. The report says alert triage and investigation are becoming more common, while the fully autonomous systems shown in demonstrations are not yet a reality.
““The fully autonomous agents demoed on conference stages are still not a reality.””
— The Pragmatic Engineer report
Accuracy and Autonomy Still Vary
The report does not provide independent, comparable performance results for the named products, or detailed figures behind Rootly’s page trends. It is therefore unclear how much AI SRE tools reduce response time, prevent outages, or cut the number of pages across the wider industry. Vendor demonstrations and company-specific deployments do not by themselves establish that the same results will hold elsewhere.
It is also unclear how much authority organizations will grant these agents. Automatically dismissing a low-priority alert is different from changing production code or infrastructure. The report describes disagreement over whether agents should make such changes and notes a potential deskilling risk if engineers lose practice handling incidents. It also says AI-related outages are emerging as a new category, but gives no frequency or severity figures.
Testing Before Wider Automation
The next test is whether teams can show that their agents improve incident handling under real operating conditions, not just in demonstrations. The report points to evaluations, backtesting, cross-checking hypotheses and human direction in chat as ways companies are trying to improve reliability. Since success depends on each team’s systems and incident history, organizations will need to assess tools against their own cases and decide what actions require human approval.
For on-call engineers, the likely near-term change is more assistance with investigation and alert filtering, rather than a fully autonomous replacement for responders. The report suggests incident-response simulations, including training from Uptime Labs, as one way to maintain human readiness. How much page volume falls, whether automated fixes become acceptable, and how teams address new AI-related failures remain open questions.
Key Questions
What does AI SRE mean?
AI SRE is a loose label for AI tools that support on-call work, especially alert triage and incident investigation. It does not mean the tool performs the full range of site reliability engineering work.
Are AI SRE agents fixing production incidents on their own?
The report says fully autonomous agents shown in conference demonstrations are not yet a reality. Investigation and alert triage are more common uses; whether an agent should make production changes is disputed.
Why are companies interested in these tools?
The report links rising interest to more frequent code changes and increasing on-call pages. Tools that investigate incidents or filter less important alerts could help teams manage that workload, though the scale of the benefit is not established.
Do AI SRE tools work?
The evidence in the report is mixed and company-specific. Teams use methods such as backtesting and independent checks, but there are no comparable results in the supplied material showing how accurately the products perform across organizations.
What risks do these tools create for on-call teams?
A mistaken diagnosis or automated change could worsen an incident, while excessive reliance on agents may reduce engineers’ response skills. The report also notes the emergence of AI-related outages, though it provides no figures on their frequency.
Source: rss
