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 pages, and vendors are building systems to investigate incidents through tools such as Slack. The reported use today is chiefly investigation and alert triage; fully autonomous production fixes remain uncommon, while accuracy, safety and effects on engineers are still unsettled.
AI SRE tools are moving from conference demos into vendor products and some companies’ internal projects, with their current focus on investigating incidents and triaging alerts rather than independently changing production systems. The development matters as companies face more deployments and pages, but the available reporting does not establish that these agents can safely resolve incidents without human oversight.
The Pragmatic Engineer reports that OpenAI built an AI SRE bot called Sevbot, an engineer at ING said the bank was working on an in-house AI SRE, and vendors in the incident-response market were increasingly promoting related products. The report spoke with Rootly, incident.io and ResolveAI; its author disclosed an investment in incident.io and said the companies did not pay for inclusion.
In practice, the systems described chiefly help engineers investigate incidents. They can gather and interpret information from operational tools, present findings in Slack or Microsoft Teams, and help triage incoming pages. Vendors and engineering teams use multiple agents, shared context or memory, and ways for engineers to steer the investigation. Whether an agent should also make code changes is disputed.
One prominent earlier demonstration came at Microsoft Build in May 2025, where Azure SRE Agent was shown identifying a cause and shipping a fix after an alert. The report’s author questioned whether teams would permit that kind of unsupervised production action. The subsequent industry activity described in the report points to narrower, human-directed uses—not routine, hands-off fixes.
More Pages, More AI Triage
The appeal is straightforward: when releases become more frequent, engineers may face more chances for failures and a larger on-call burden. The source presents the relationship as a simple model—incident rate depends on the number of changes and the probability that any one change causes a failure. It says many companies have seen pull-request frequency at least double over a year, but does not provide a single industry-wide dataset establishing that rate.
Rootly supplied data indicating that pages were up compared with a year earlier and that outside-business-hours pages had nearly doubled. The report does not specify the precise measurement window, baseline, or sample in the supplied material, so the figure should be treated as a vendor-reported observation, not a universal measure. If page volumes are rising, faster investigation and filtering low-value alerts could reduce disruption. But a tool that misses a serious signal or recommends an unsafe fix could add risk instead.
There is also a workforce question. The report warns that relying on agents could weaken engineers’ incident-response skills over time. If routine triage is delegated, teams may get fewer interruptions, yet the incidents that reach a human could be harder to diagnose. The practical test is not whether an agent can perform well in a staged demo, but whether it helps a particular team respond more accurately and safely during real incidents.
Products Worth Considering
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Build Demo to On-Call Tools
At Microsoft’s May 2025 Build conference, an onstage Azure SRE Agent demonstration showed an alert being investigated and a fix shipped without waking an on-call developer. The report’s author said the scenario seemed artificial and questioned whether engineering teams would allow an AI to make production decisions without a human. The author said there was little visible discussion of SRE agents after that event until more recent vendor and company activity.
“AI SRE” has become a common label for AI-assisted on-call tooling, although the report says vendors and practitioners do not find it especially accurate: site reliability engineering covers much more than responding to alerts. Alternatives such as automated incident investigation agents have not displaced the term. The report also notes that some companies build their own systems rather than buying from vendors, and compares current alert triage with approaches Meta used in its own systems two years earlier.
““AI SRE” looks on its way to becoming the agreed term for AI-augmented oncall tooling, even though we’ve not spoken to a single person who likes it.”
— The Pragmatic Engineer report
Accuracy and Autonomy Remain Open
The report does not provide independent, comparative results showing how accurately AI SRE products diagnose incidents across companies. It says evaluation is difficult because what counts as a correct investigation depends on each company and engineering team. Suggested techniques include testing agents against past incidents, using separate agents to challenge a proposed explanation, and trying a mix of models; these are methods, not proof of dependable performance.
It is also unclear how often current systems can take action safely, how frequently engineers accept or reject their recommendations, and whether they measurably cut time to resolution or reduce the total number of pages. The Rootly page figures are vendor-provided, and their exact period and comparison basis are not given in the source material. The report describes fully autonomous agents as not yet a routine reality, while noting that AI-based alert triage and suppression of unimportant pages are becoming more common.
How Teams Will Test These Agents
For now, the next step is likely to be company-by-company evaluation rather than a single industry verdict. Teams can compare agent investigations with past incidents, track whether suggested causes are useful, and keep human approval in place for consequential production changes. The report says vendors are working on speed, cost and accuracy, using methods such as model routing, shorter prompts and monitoring cache-hit rates.
On-call teams will also need to watch for changes in the incidents themselves as AI coding tools increase deployment activity. The report suggests incident-response simulations, including training offered by Uptime Labs, as one way to maintain engineers’ skills. Whether AI SRE systems reduce after-hours disruption without creating new reliability or deskilling problems remains to be established through operational experience.
Key Questions
What does AI SRE mean?
AI SRE is a loose industry term for AI-assisted on-call and incident-response tools. The systems described in the report focus mainly on investigating alerts and helping engineers triage incidents.
Can an AI SRE fix production incidents on its own?
A Microsoft Build demonstration in May 2025 showed an agent shipping a fix, but the report says fully autonomous agents are not routine in practice. Whether to let an agent make production changes remains disputed.
Why are companies interested in these tools?
The report links interest to rising deployment activity and vendor-reported increases in pages, including after-hours alerts. AI-assisted investigation and triage could help teams manage the workload, although the size of the benefit has not been established independently.
How do teams measure whether an AI SRE works?
The report says accuracy is specific to each team and difficult to measure. Possible approaches include testing against prior incidents and checking an agent’s hypothesis with a separate agent, but these techniques do not by themselves prove reliable performance.
Could AI SRE tools affect engineers’ skills?
The report raises a deskilling risk if engineers handle fewer incident investigations. It suggests incident-response simulations as one way teams could keep those skills active.
Source: newsletter.pragmaticengineer.com
