TL;DR
Get business pricing on networking and server gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
AI SRE tools are gaining attention as companies report more deployments and on-call pages, with some building internal agents and vendors expanding incident-response products. Current use centers on investigating alerts and triaging pages, not independently making high-stakes production changes; effectiveness varies by company and is difficult to measure.
AI-assisted on-call tools are gaining traction as companies contend with rising deployment and page volumes, but their practical role is mostly to help investigate incidents and triage alerts—not to independently change production systems. A report by The Pragmatic Engineer, based on interviews with incident-response vendors and engineers, says the fully autonomous fixes shown in conference demonstrations are not yet a routine reality.
The report points to more frequent code deployments as a driver of interest. It says many technology companies have seen pull-request frequency double over a year as AI coding agents speed up software development. Rootly data cited in the report also indicates that pages have risen from a year earlier and that outside-business-hours pages have nearly doubled. The report does not provide the full dataset or a common measurement period for those figures, so they should not be treated as an industry-wide estimate.
In practice, an AI SRE agent typically collects information from incident tools and company systems, then helps responders investigate possible causes, often through Slack or Microsoft Teams. Multiple agents may explore different hypotheses, while engineers guide the investigation. Some systems use a context layer—such as a knowledge graph and stored incident history—to supply company-specific information. Several companies are also building internal tools rather than relying entirely on vendors.
The report says alert triage and suppressing unimportant pages are increasingly common uses. Whether an agent should make code changes or remediate an incident without human approval remains disputed. The author describes seeing Microsoft demonstrate Azure SRE Agent at its May 2025 Build conference, where it investigated an alert and shipped a fix, but says they had not heard of comparable autonomous production use becoming established afterward.
Fewer Pages, Harder Incidents
If deployment frequency rises, engineering teams may face more opportunities for failures and a heavier on-call load. AI investigation tools could reduce the time responders spend gathering logs, reviewing prior incidents and identifying likely causes. If triage also improves, engineers might receive fewer low-value alerts and focus on incidents that need human judgment.
But fewer pages do not necessarily mean less operational risk. The report warns that incidents reaching an engineer may be more difficult, while greater reliance on automated investigation could weaken responders’ incident-handling skills. It points to incident-response simulations as one way to keep those skills practiced. The benefit of AI SRE therefore depends not only on speed, but on whether teams can trust its findings and retain control over consequential actions.
Products Worth Considering
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
DJI Osmo Pocket 4 Gimbal Camera Pro Kit, 18-Piece Bundle
As an affiliate, we earn on qualifying purchases.
From Build Demo to On-Call Tools
At Microsoft Build in May 2025, a stage demonstration showed Azure SRE Agent tracing an alert to a cause and deploying a fix without waking an on-call developer. The report’s author questioned whether engineering teams would allow an AI system to make production decisions that could worsen an outage. Interest has since become more visible: the report says OpenAI built an internal AI SRE bot called Sevbot, an ING engineer described work on an in-house system, and vendors have been promoting incident-response products.
The term “AI SRE” is itself disputed. A site reliability engineer’s work extends well beyond responding to alerts, and the report says vendors interviewed—including Rootly, incident.io and ResolveAI—acknowledged that the label is imperfect. The author disclosed being an investor in incident.io and said they were not paid by companies mentioned. The report frames these products more narrowly as AI-augmented on-call or incident investigation tools.
““The fully autonomous agents demoed on conference stages are still not a reality.””
— The Pragmatic Engineer report
Accuracy and Autonomy Remain Open
The report does not establish how accurately AI SRE products identify root causes across companies, or how often their recommendations lead to successful remediation. Measuring accuracy is difficult because systems, incident histories and definitions of a useful answer differ by organization and engineering team. Vendors and teams use approaches such as testing against past incidents, evaluating results and checking hypotheses with other agents or models, but those methods do not provide a universal performance measure.
It is also unclear how broadly the cited rises in deployment and page volumes apply. The report presents figures and observations from particular sources, not a standardized survey of the technology industry. Nor does it show that AI SRE tools themselves reduce outage rates. The degree of human review, the safeguards around production changes, and the cost and speed trade-offs between different systems remain company-specific.
Teams Will Test Before Trusting
The next phase is likely to be continued use of AI for investigation and alert triage, with teams testing results against their own incident records and operational requirements. The report describes model selection, concise prompts and monitoring cache performance as ways teams manage the competing demands of speed, cost and accuracy. These are engineering choices, not evidence that one approach works reliably everywhere.
For now, the report’s account supports a cautious distinction: agents can assist responders, but autonomous production remediation has not been shown to be standard practice. More evidence would be needed to judge whether these systems consistently reduce pages, shorten incident resolution or prevent repeat failures. Companies adopting them will also have to decide which actions require human approval and how to preserve responders’ skills.
Key Questions
What does “AI SRE” mean?
It is a loose label for AI-assisted on-call and incident-response tools, especially systems that investigate alerts, collect relevant context or help triage pages. The report notes that the name is imprecise because SRE work covers more than on-call response.
Are AI SRE agents fixing production outages on their own?
The report says fully autonomous production fixes are not routine. While a Microsoft Build demonstration in May 2025 showed an agent shipping a fix, common uses described in the report focus more on investigation and triage, with humans guiding the process.
Why are companies interested in these tools?
The report links interest to faster software delivery and rising pages at some companies. AI agents may help responders examine incidents faster and filter less important alerts, though the report does not establish that they lower outage rates.
Do AI SRE tools work reliably?
There is no single answer from the report. Accuracy is difficult to measure because each company has different systems and incident histories. Teams use evaluations and tests against past incidents, but results and safeguards need to be assessed in each organization.
Source: newsletter.pragmaticengineer.com
