TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
AI SRE tools are being adopted to investigate incidents and help triage on-call alerts, as some companies report more pages alongside faster software deployment. The current use is mainly investigation and human-guided response; claims that agents can independently fix production failures remain ahead of what is commonly deployed.
AI SRE tools are moving from conference demos into company incident-response work, but their established role is mostly to investigate alerts and support human responders, not independently make production changes. A report by The Pragmatic Engineer says startups including Rootly, incident.io and ResolveAI are building these systems as companies contend with more deployments and pages; how reliably the tools reduce incident workload remains difficult to establish.
The term AI SRE has taken hold for tools that apply AI to on-call work, although the report says vendors and practitioners do not particularly like the label. Site reliability engineering covers more than responding to alerts, and the products discussed focus chiefly on investigating incidents, assembling relevant information and coordinating with engineers in tools such as Slack or Microsoft Teams.
The report points to rising software-change rates as one reason for growing interest. It says many companies have seen pull-request frequency double over a year as AI coding agents accelerate development. Rootly data cited in the report shows pages rising year over year and outside-business-hours pages nearly doubling. Those figures are company-supplied observations; the report does not provide a broader independent dataset or establish that AI-generated code alone caused the increase.
Companies are using different approaches to build investigation agents. These can delegate work to multiple agents, draw on internal documentation and incident history through a context layer, and allow engineers to steer investigations in chat. The report says some organizations, including OpenAI and ING, have built or are developing in-house systems. It also describes reliability practices such as testing competing explanations with different models, evaluating results against past incidents, and balancing speed and cost through model routing and shorter prompts.
How AI Is Changing On-Call
If deployment activity and pages rise together, faster incident investigation could help engineers keep pace without simply adding more people to rotations. Agents that gather logs, trace likely causes and summarize an incident may reduce time spent assembling evidence. Tools that identify or silence low-priority alerts could also reduce interruptions, although the report describes this as an emerging use rather than a measured industry-wide outcome.
The trade-off is that fewer routine pages may leave engineers with a more difficult set of incidents when they are called. The report flags a possible deskilling risk if responders rely on agents without practicing incident diagnosis themselves. It suggests simulation-based incident training as one way to retain those skills. These are operational concerns, not established findings that AI SRE has already reduced staffing or weakened teams.
Popular right now
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Vitamix 5200 Professional-Grade Blender, 64 oz. Container, Black
As an affiliate, we earn on qualifying purchases.
From Build Demo to Triage
At Microsoft Build in May 2025, Microsoft demonstrated Azure SRE Agent investigating an alert, identifying a possible cause and shipping a fix without waking a human on-call engineer. The report’s author said the demonstration seemed unlikely to reflect production practice, given the risk of an unsupervised change worsening an outage. The report says discussion of SRE agents later became more visible, with vendor conference booths and companies describing internal projects.
In practice, the gap between assisting an investigation and changing production systems matters. The report says the fully autonomous agents shown on conference stages are not yet a common reality. Human-guided investigation is more typical, while the degree of automation varies by company and incident. The author also disclosed an investment in incident.io and said the report was not paid for by the vendors it covered.
“The fully autonomous agents demoed on conference stages are still not a reality.”
— The Pragmatic Engineer report
Accuracy and Autonomy Remain Open
There is no shared, clear measure of whether an AI SRE agent is accurate. The report says performance depends on each company’s systems and even on individual engineering teams, while incident investigations can have several plausible explanations. Backtesting, evaluations and adversarial checks may help, but the material does not supply comparable results showing how often the products identify the correct cause or prevent outages.
It is also unclear how much the reported increase in pages is caused by more deployments, what portion of those changes were produced with AI, or whether the page trends apply across the industry. The report describes vendor data and conversations with selected companies, rather than a controlled study. It gives no common benchmark for time saved, costs, false alerts or production fixes made safely without human approval.
Evidence Needed From Deployments
The next test is whether these systems perform consistently in real incidents, not only in demonstrations. Companies adopting them will need to decide which actions agents can take, when a human must approve a change, and how performance will be evaluated. Useful evidence would include comparable measures of investigation time, alert quality, incident outcomes and the rate of incorrect recommendations.
For now, the report points toward human-supervised investigation and triage as the practical direction. As vendors and internal teams refine their agents, on-call engineers will still need to review findings and handle consequential decisions. Whether that eventually means fewer pages, faster recovery or merely a different set of tasks remains to be shown.
Key Questions
What does AI SRE mean?
AI SRE is a loose label for AI-assisted on-call and incident-response tools. The products described in the report chiefly investigate alerts, gather context and help engineers coordinate a response.
Can an AI SRE agent fix a production outage by itself?
Some demonstrations have shown agents proposing or shipping fixes, including Microsoft’s May 2025 Build demo. The report says fully autonomous production response is not common practice; human-guided investigation is more typical.
Why are companies considering these tools?
The report links interest to higher deployment activity and rising pages, including more pages outside business hours in Rootly data. Faster investigation may help teams manage the workload, but the figures do not establish a single cause or an industry-wide trend.
Is there proof that AI SRE reduces incidents?
The report does not provide a shared benchmark or controlled evidence that these tools reduce incidents. It says accuracy is specific to a company’s systems and that measuring it remains difficult.
Source: newsletter.pragmaticengineer.com
