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 drawing interest as faster code deployment coincides with more incidents and on-call pages. The Pragmatic Engineer’s report says current systems mainly help investigate and triage incidents; fully autonomous production repair remains largely a conference-demo vision, with accuracy and safety varying by team.
AI-assisted incident response tools are moving from conference demonstrations into vendor products and some companies’ internal systems, as engineering teams face more deployments and rising on-call pages, according to a report by The Pragmatic Engineer. But the systems in use are chiefly aimed at investigating and triaging incidents; the report says agents that independently make and ship production fixes are not yet a routine reality.
The report describes a widening set of tools from incident-management companies including Rootly, incident.io and ResolveAI, alongside internal projects at technology and financial firms. It says OpenAI built an internal AI SRE bot called Sevbot, and an engineer at ING told the publication the bank was developing its own system. Those examples indicate growing interest, but do not establish how widely the tools are deployed or how well they perform.
The tools commonly work through Slack or Microsoft Teams, where an agent can gather incident information, investigate possible causes and help engineers respond. Systems may send several agents to examine different hypotheses, while engineers steer the investigation. Vendors and teams also use a “context layer”—such as stored operational knowledge and memory—to give agents relevant information about a company’s services.
The report says teams are exploring ways to improve reliability, including testing hypotheses with separate agents, using multiple models, and evaluating systems against past incidents. But there is no simple, shared measure of an AI SRE’s accuracy: useful performance depends on each company’s infrastructure and on what its engineers expect the agent to handle. Faster responses, lower costs and reliable findings can also involve tradeoffs.
More Pages, Harder On-Call Decisions
The interest in AI SRE tools is tied to a practical pressure on engineering teams: more frequent code changes can mean more opportunities for failures, and more incidents can mean more work for the people on call. The report presents a simple reliability relationship: incident rate depends on both the rate of changes and the chance that any one change causes a failure. Increasing deployment frequency can raise incident rates even if the risk per change stays the same.
Rootly data cited by the publication shows pages have risen from a year earlier, with outside-business-hours pages nearly doubling over that period. The source does not provide a broader industry-wide measurement in the supplied material, so the figures should be treated as Rootly’s data rather than a universal estimate. If tools can handle routine investigation or silence unimportant alerts, they may reduce interruptions. The report says the incidents that still reach engineers may be the more difficult ones.
There is also a workforce concern. If engineers rely on agents to investigate incidents, they may get less practice diagnosing failures themselves. The report raises this deskilling risk and points to incident-response simulations as one way teams could maintain response skills.
Products Worth Considering
Apple iPad Air 11-inch (M2), 512GB, Wi-Fi, Space Gray
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
As an affiliate, we earn on qualifying purchases.
From Build Demo to Incident Triage
At Microsoft’s BUILD conference in May 2025, the company demonstrated Azure SRE Agent responding to an alert, identifying a possible cause and shipping a fix without waking a human engineer. The report’s author said the demo seemed artificial and questioned whether teams would allow an AI to make production decisions without oversight, given the possibility of worsening an outage.
The report says interest became more visible later, with vendors presenting AI incident products and companies building internal agents. It also notes that “AI SRE” is an imperfect label: site reliability engineers do considerably more than respond to alerts. The vendors interviewed reportedly acknowledged that they had not found a more widely accepted name for this category of tools.
The publication’s author disclosed an investment in incident.io and said they had not been paid to mention any company. That relationship is relevant because incident.io is among the companies discussed. The report also notes that AI-based alert triage is not entirely new; it says Meta was operating its systems with AI in this area two years earlier.
“The fully autonomous agents demoed on conference stages are still not a reality.”
— The Pragmatic Engineer report
Accuracy and Autonomy Remain Open
The available evidence does not establish how accurate these systems are across companies, how often their recommendations need correction, or whether they reduce incident duration or on-call workload in measured deployments. The report says evaluation is highly specific to a company and even to individual engineering teams. It does not provide comparable results from controlled tests or a common benchmark.
It is also unclear how much authority teams give agents in production. Investigating an alert and suggesting a remedy are different from changing code or deploying a fix. The report describes debate about whether agents should make code changes and says fully autonomous repair is not yet a routine capability. The safety controls and human approval requirements used by individual organizations are not detailed.
Rootly’s page figures offer one company’s data, but the supplied report does not specify the underlying counts or a full methodology. The causes of the increase in pages, and the extent to which AI-generated code is responsible, cannot be determined from those figures alone.
Teams Will Test Triage and Oversight
The next step for teams adopting AI SRE tools is likely to be testing them against their own incidents and operational practices. The report describes backtesting, evaluation runs and independent checks of agent hypotheses as approaches under development, while noting that each organization needs to decide what counts as a useful and safe result.
Companies will also need to decide which actions agents can take without approval, how engineers review recommendations, and how they preserve incident-response skills. For now, the clearest near-term use described in the report is helping triage pager alerts and speed up investigations—not handing an agent unrestricted control of production systems.
Key Questions
What does “AI SRE” mean?
It is a loosely used label for AI tools that support reliability work, especially alert triage and incident investigation. The report says the name is imperfect because SRE work extends well beyond on-call response.
Can an AI SRE fix an outage by itself?
The report says fully autonomous agents that investigate and ship production fixes, as shown in conference demonstrations, are not yet a routine reality. The authority given to individual tools varies and is not fully detailed.
Why are companies paying more attention to these tools?
The report links their growing interest to higher deployment frequency and rising pages. It cites Rootly data showing pages increased from a year earlier and after-hours pages nearly doubled, while cautioning that this is vendor data, not a universal industry measure.
How do teams check whether an AI incident agent works?
Approaches described include testing against past incidents, evaluating outcomes and checking an agent’s hypothesis with other agents or models. There is no single accuracy measure that applies across companies, according to the report.
Source: rss
