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 gaining attention as faster software releases and rising incident pages put more pressure on on-call teams. The Pragmatic Engineer’s report describes current systems as aids for incident investigation and triage, while fully autonomous production fixes remain a conference-demo vision rather than a broadly established practice.
AI SRE tools are attracting more attention as software teams face rising deployment activity and on-call pages, but the systems described in a new report are mostly used to investigate incidents and triage alerts, not to make unsupervised production changes. The Pragmatic Engineer spoke with incident-platform companies Rootly, incident.io and ResolveAI, and reports that some technology companies are also building internal tools; how reliably these systems work varies by company and remains difficult to measure.
The report traces the current interest to a practical pressure on reliability teams: AI coding tools have helped increase the pace of software changes, while incident pages are also rising. The author says many companies have seen pull-request frequency double in a year, and cites Rootly data showing pages are up from a year earlier, with outside-business-hours pages nearly doubling. The report does not provide a universal industry-wide measure for those trends.
In practice, the label “AI SRE” covers tools that gather information about an incident, search logs and other operational context, and suggest possible causes inside collaboration tools such as Slack or Teams. Teams may use several agents to investigate different leads, while engineers guide the process. Vendors described using a context layer, including a knowledge graph or stored memory, to help agents connect current alerts with relevant systems and past incidents.
The harder question is whether an agent should take action. The report says teams disagree about allowing AI to change code or systems, given the risk that a mistaken fix could worsen an outage. Approaches described for improving results include checking hypotheses with a separate, adversarial agent, using multiple models, and testing systems against past incidents. Yet there is no single accuracy measure that applies across companies: a useful evaluation depends on each organization’s systems and the incident types its teams handle.
Why On-Call Teams Are Testing AI
For engineering teams, the appeal is less about replacing SREs than reducing the time spent collecting evidence during an incident. If an agent can assemble relevant logs, identify likely services involved and summarize previous failures, an on-call engineer may reach a useful diagnosis faster. Alert triage could also reduce interruptions from pages that do not require immediate human attention.
But faster investigation does not remove the need for sound operational judgment. A tool that confidently points engineers toward the wrong cause can waste time, and an automated production change carries a different level of risk from a suggested diagnosis. The report’s distinction between assisted investigation and autonomous remediation matters: the former is already being pursued, while the latter remains contested.
The shift may also change the work engineers do when paged. If routine alerts are filtered out, the remaining incidents could be less frequent but more difficult. The report flags a deskilling risk if engineers lose opportunities to practice diagnosis, and points to incident-response simulations as one way teams could maintain those skills.
Products Worth Considering
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From BUILD Demo to Daily Triage
At Microsoft’s BUILD conference in May 2025, Azure SRE Agent was demonstrated investigating an alert and shipping a fix without waking a human on-call developer. The report’s author said the demo seemed artificial and questioned whether teams would permit an AI to make production decisions without oversight. The report says the fully autonomous version shown on conference stages is not yet a common reality.
Interest has since become more visible. The author reports that OpenAI built an internal AI SRE bot called Sevbot, an engineer at ING described work on an in-house system, and AI incident-platform vendors have become more prominent at technology conferences. These examples indicate activity, not broad adoption or independently verified performance. The report also notes that the name “AI SRE” is imperfect: site reliability engineering includes much more than handling alerts, and vendors have not settled on a more widely used alternative.
The author disclosed an investment in incident.io and said there was no other vendor affiliation or payment for the article. That relationship is relevant because incident.io was among the companies interviewed.
““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 a shared benchmark for how accurately AI SRE systems diagnose incidents or how often their recommendations help teams resolve them. Results depend on company-specific infrastructure, available operational data and incident history, making vendor claims hard to compare without common tests.
It is also unclear how broadly organizations will permit agents to make changes in production, or whether reported increases in pages and deployment activity reflect the same pattern across the industry. The examples of internal tools and vendor interest show experimentation, but do not establish that these systems consistently reduce outages, workload or response times.
More Testing, Limited Automation
The near-term development to watch is whether teams can show that AI-assisted investigation and alert triage improve incident response under real operating conditions. The report describes backtesting, evaluation against earlier incidents and checks by multiple agents as methods teams are using to test systems, though it offers no common standard for judging results.
For now, the source describes a cautious path: let agents gather evidence and flag likely causes, with engineers retaining responsibility for consequential decisions. As use expands, teams will have to weigh the benefits of fewer or faster-handled pages against the risks of incorrect recommendations and reduced hands-on incident practice.
Key Questions
What does “AI SRE” mean?
In the report, it refers mainly to AI-assisted incident investigation and alert triage—tools that gather operational context and help engineers assess what may have caused a problem. The term is broader and less precise than the work these tools currently perform.
Can AI SRE agents fix production outages on their own?
The report says fully autonomous agents that diagnose an incident and ship a production fix are not yet a common reality. Teams are debating whether agents should make code or system changes without human approval.
Why are companies exploring these tools now?
The report points to more frequent software changes and rising incident pages as reasons teams are investing in on-call tools. It cites Rootly data showing pages increased from a year earlier and outside-business-hours pages nearly doubled, but those figures are not presented as a universal industry measure.
Is there proof that AI SRE tools work?
The report describes current use and testing approaches, but does not provide a shared benchmark proving consistent results across organizations. Accuracy is company-specific, and evaluation methods such as backtesting need to reflect each team’s systems and incident history.
Could AI SRE tools affect engineers’ skills?
Potentially. The report raises a deskilling concern if fewer pages mean engineers get less practice responding to incidents. It suggests incident-response simulations as one way teams might keep those skills active.
Source: newsletter.pragmaticengineer.com
