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, with products focused mainly on speeding up incident investigation and triage. The Pragmatic Engineer’s report says fully autonomous agents that diagnose and fix production incidents without human oversight remain more a conference-demo vision than an established practice.
AI-assisted site reliability engineering tools are moving from conference demos toward use in incident investigation and alert triage, as engineering teams contend with more software changes and on-call pages, according to a report by The Pragmatic Engineer. But the report found no evidence that fully autonomous agents routinely resolve production incidents without human oversight: current systems are chiefly being used to gather evidence and help engineers investigate.
The report describes a growing market for tools that connect to services such as Slack or Microsoft Teams, collect information from engineering systems, and help responders investigate incidents. Companies named in the report include Rootly, incident.io and ResolveAI; it also cites OpenAI’s internal incident-response bot, called Sevbot, and an engineer at ING who said the bank was building an in-house system. These examples indicate activity, not a measured industry-wide adoption rate.
A typical investigation agent may break a problem into tasks and run multiple agents to examine different sources of information. Engineers can guide the investigation in a chat channel, while a company-specific context layer—such as a knowledge graph or stored operational history—can supply details about services and prior incidents. The report says teams are also testing approaches such as checking one agent’s proposed explanation with another model, evaluating results against past incidents, and tuning the balance between speed, cost and accuracy.
The report cites Rootly data showing that pages were up from a year earlier and that outside-business-hours pages had nearly doubled. It also says many companies have seen pull-request frequency double or more over a year, attributing the rise to AI coding agents. The material does not provide the underlying sample, exact measurement window, or comparison method for those figures, so they should not be treated as a universal industry measure. More changes can create more opportunities for failures, but deployment volume alone does not establish the cause of any particular incident.
Why More Pages Are Driving Adoption
On-call teams are the immediate audience for these tools. If an agent can rapidly assemble logs, deployment history and service ownership information, it may reduce the time responders spend searching across systems and help them decide which alerts need attention. The report says triage and investigation are the most common practical focus, including filtering or silencing alerts judged unimportant.
The potential gains come with tradeoffs. A system that makes an incorrect diagnosis or changes production code could worsen an outage, which is why the report describes disagreement about whether agents should be allowed to make changes. It also flags a deskilling risk: if engineers rely on automated analysis, they may get less practice diagnosing incidents themselves. The report suggests incident-response simulations as one way to maintain those skills.
Products Worth Considering
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Microsoft Demo to On-Call Tools
At Microsoft’s BUILD conference in May 2025, the company demonstrated an Azure SRE Agent responding to an alert, identifying a suspected cause and shipping a fix without waking a human on-call engineer. The report’s author described the demonstration as difficult to imagine teams trusting for unsupervised production decisions. The article does not establish that the demo represented a generally available or routinely deployed capability.
The newer activity described in the report is more limited: internal systems and commercial products aimed at investigation and incident workflows. The label “AI SRE” is itself disputed. Site reliability engineers work on more than alerts, and the vendors contacted by the report said they had not found a better term. The report’s author disclosed an investment in incident.io and said they had not been paid to mention the companies.
“The fully autonomous agents demoed on conference stages are still not a reality.”
— The Pragmatic Engineer report
How Reliable Are These Agents?
The report does not establish a shared benchmark for AI SRE accuracy. What counts as a correct investigation depends on a company’s systems, incident history and engineering teams, making results difficult to compare across products. Backtesting and model checks may help, but the report does not provide comparative performance results or independent evidence that one approach works better.
It is also unclear how widely the named tools are deployed, how often they make mistakes, and what safeguards govern any proposed production changes. The cited page and deployment trends come from the report’s sources, with no full methodology in the supplied material. It remains uncertain whether increased AI-assisted development is the main cause of rising pages across the industry or how much incident rates vary between companies.
Human Oversight Remains the Test
The next measure of progress will be whether teams can show that these tools help responders resolve incidents faster or reduce unnecessary pages without adding new operational risk. That requires company-specific testing against prior incidents, clear limits on agent permissions and a human review process for consequential actions. The report describes these as engineering challenges, not results already demonstrated across the market.
For now, the development to watch is assisted investigation rather than unattended repair. As teams gather experience, they will need to report what tasks agents actually perform, how their accuracy is measured, and whether on-call workloads change. Until that evidence is available, claims of hands-off incident resolution should be treated as vendor or demonstration claims, not established practice.
Key Questions
What does “AI SRE” mean?
It is a loose label for AI-assisted reliability and incident-response tools that help with tasks such as investigating alerts, collecting operational context and triaging pages. The report notes that SRE work is broader than on-call response, making the term imprecise.
Do AI SRE agents fix production incidents on their own?
The report says fully autonomous agents that investigate and ship production fixes without human oversight are not yet an established reality. It describes current use as focused mainly on investigation and triage, while permission to change production remains debated.
Why is interest in these tools increasing?
The report points to higher software change frequency and more on-call pages as possible drivers. It cites Rootly data showing pages rose year over year and out-of-hours pages nearly doubled, but the supplied material does not give enough methodology to generalize those figures to every company.
How can teams judge whether an AI SRE tool works?
Teams can test agents against their own incident histories, track whether their hypotheses are useful, and check results with other models or human responders. The report says accuracy is company-specific and does not offer a common benchmark or comparative product results.
Source: newsletter.pragmaticengineer.com
