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 attracting interest as deployment activity and on-call pages rise, with startups and some companies building agents to investigate incidents. The systems are being used mainly for triage and investigation; the source report says fully autonomous agents that safely fix production problems without human oversight remain out of reach.
AI SRE tools are moving from conference demonstrations into engineering teams’ incident workflows, as companies contend with more deployments and rising on-call pages, according to a recent report by The Pragmatic Engineer. The systems are mainly being used to investigate and triage incidents; the report says agents that independently make production changes without human oversight are not yet a routine reality.
The report describes a developing market for tools that connect AI agents to incident systems and team conversations, often in Slack or Microsoft Teams. Agents can gather logs and service information, propose possible causes and help engineers investigate. Some systems use several agents in parallel, while engineers direct or review their work through chat. A shared context layer, including stored knowledge and prior incident information, is intended to improve results.
The author reports hearing about OpenAI’s Sevbot and an in-house AI SRE project at ING, alongside increasing visibility for vendors including Rootly, incident.io and ResolveAI. Those companies provided input for the report. The author disclosed an investment in incident.io and said they had not been paid to mention any company.
The report points to a Microsoft Azure SRE Agent demonstration at the company’s May 2025 BUILD conference, where an agent appeared to identify an alert’s cause and ship a fix. The author questioned whether teams would allow that level of unsupervised production action. The report’s current assessment is more limited: automated triage, including filtering less important pages, is becoming more common, while autonomous remediation remains unsettled.
Why On-Call Teams Are Looking to AI
More frequent code changes can put pressure on reliability teams. The report cites a basic relationship—incident rates depend on both the number of changes and the chance that each change causes a failure—and says many companies have seen pull-request frequency at least double over the past year as AI coding agents increase development activity. It also cites Rootly data showing pages are up year over year and outside-business-hours pages have nearly doubled. The source does not provide the full dataset, measurement window or comparison details for those figures, so they should be read as reported observations, not universal industry statistics.
If engineers receive more alerts, faster investigation could reduce the time spent finding relevant evidence and help teams learn from recurring failures. But the report also warns of a deskilling risk: if staff rely on agents for diagnosis, they may have fewer chances to practise incident response. It suggests simulation exercises as one way to maintain those skills. The incidents that still reach a human may also be the more difficult ones, rather than a representative sample of routine problems.
Popular right now
Samsung 42-Inch Class OLED S90F 4K Smart TV (2025)
KitchenAid Artisan Series 5-Quart Tilt-Head Stand Mixer, Dried Rose (KSM150PS)
Vitamix Propel Series 750 Professional-Grade Blender, 4 Programs, Black
As an affiliate, we earn on qualifying purchases.
From Demo to Incident Workflow
The term “AI SRE” is being used for AI-augmented on-call tooling, though the report notes that the companies it spoke with do not find the label especially accurate. Site reliability engineering covers broader work than responding to alerts. Alternative descriptions, such as automated incident investigation or remediation agents, have not become as widely used.
The report frames the technology as one response to higher change volume, not a replacement for reliability practices. It identifies testing before deployment and learning from incidents as ways to reduce failures. AI incident tools may support the second approach by helping teams investigate causes and improve system resilience. Some technology companies are also building internal agents rather than relying entirely on vendors.
Accuracy is not a single score that transfers cleanly between organizations. The report says performance depends on a company’s systems and even on the engineering team. Suggested methods include testing agents against past incidents, running evaluations, and using separate agents or different models to challenge an initial hypothesis. These methods can help surface mistakes, but they do not establish that an agent is safe to make production changes on its own.
“The fully autonomous agents demoed on conference stages are still not a reality.”
— The Pragmatic Engineer report
Safety and Results Still Vary
The supplied report does not provide a controlled, cross-company measure of whether AI SRE tools reduce incident duration, prevent outages or lower the overall number of pages. It says accuracy is difficult to measure and varies by organization, making broad comparisons unreliable. The Rootly page figures are also not accompanied here by a precise period, sample size or baseline methodology.
It remains unclear how often engineers accept an agent’s diagnosis, how frequently its suggestions are wrong, and which organizations permit agents to take actions that change production systems. The report’s observations indicate increasing interest and some active deployments, but they do not establish that the tools work equally well across teams or that autonomous fixes are safe.
More Testing Before Autonomy
As teams adopt these tools, the next test is whether evaluations and backtesting on prior incidents translate into better results during live events. Engineering groups will also need to decide which actions agents may take automatically and which require human review. The report points to model selection, prompt design and cache use as ways vendors manage trade-offs among speed, cost and accuracy.
For on-call engineers, the near-term change is more likely to be AI-assisted investigation and alert filtering than removal from incident response. The report suggests practising with incident simulations to guard against loss of response skills. It does not identify a specific launch date or industry-wide milestone for unsupervised remediation.
Key Questions
What does “AI SRE” mean?
It generally refers to AI-assisted on-call tools that help investigate and triage incidents. The label is broader than the systems’ current focus, since site reliability engineering includes work beyond responding to alerts.
Are AI SRE agents fixing production outages by themselves?
The report says fully autonomous production fixes are not yet routine. Current use is more commonly focused on triage, investigating likely causes and helping engineers work through incidents.
Why are companies interested in these tools now?
The report links the interest to rising deployment activity and more on-call pages. It cites company pull-request frequency increases and Rootly page data, though the supplied material does not give complete measurement details for those figures.
Do AI SRE tools work reliably?
There is no single result in the report that proves reliability across companies. Accuracy depends on the organization and team, and the report says measurement is difficult; it describes evaluation and backtesting as useful practices, not guarantees.
Could AI on-call tools affect engineers’ skills?
The report raises a possible deskilling risk if engineers have fewer opportunities to diagnose incidents themselves. It suggests incident-response simulations as a way to keep those skills practised.
Source: newsletter.pragmaticengineer.com
