Site reliability engineers (SREs) have the tough assignment of resolving thorny performance and reliability issues. But their primary mission is to provide devops teams with operational insights and to suggest implementation improvements on business system performance, security, and overall robustness.
Google introduced its SRE playbook in 2003, but it took some time for the role’s definition, tools, and techniques to become mainstream. Startups were the first to adopt observability for cloud-native applications and create dedicated SRE positions. As tools matured and SRE responsibilities became more clearly defined, larger enterprises assigned SREs to work as a bridge between devops and IT ops teams to improve resilience across a wider range of applications, APIs, and data pipelines.
SRE is a career path for multidisciplinary engineers with strong investigative instincts, sharp data analytics skills, and the temperament to perform under pressure. It has become a critical responsibility as tech became mission-critical for enterprises, and it is a growing role in the genAI era as more businesses deploy AI agents.
But the critical need for resiliency and greater technological complexity brings new challenges for SREs. According to the 2026 State of Production Reliability and AI Adoption report, 44% of respondents experienced an outage linked to ignored or suppressed alerts in the past year, and 35% report their engineers occasionally ignore or dismiss alerts due to alert fatigue. More than 70% of alerts received are not actionable, according to 57% of organizations.
So, is AI making the SRE’s role easier and helping businesses run more reliable technology operations? On the other hand, AI is also driving complexity, as companies deploy genAI tools and AI agents across more business functions and seek to automate more decision-making across operations.
Over the past decade, SRE responsibilities have become somewhat easier through improvements in monitoring platforms, observability practices, tools for centralizing operational data, and AI applied in IT operations (AIops). But during the heat of resolving an outage or performance issue, it’s not easy to correctly identify what system triggered the issue versus other downstream systems impacted by it.
According to the Komodore 2025 Enterprise Kubernetes Report, 79% of production incidents originate from recent system changes, including deployments and changes to compute environments. But the other 21% of incidents stem from issues outside of the business’s control, including network failures, third-party changes, and cloud provider failures.
“SREs using AI capabilities succeed or fail in the moment an incident unfolds, when engineers are deciding what to investigate next,” says Itiel Shwartz, CTO at Komodor. “If the system streamlines root cause detection, connects signals to recent changes, and explains its reasoning in a way engineers recognize, it earns trust. If it adds uncertainty or demands extra validation, it gets sidelined, regardless of how bespoke the model behind it may be. What’s less obvious is what it takes to make AI for SREs work in production, and how different that reality is from prototypes, demos, or early internal builds.”
AIops is not a new capability, especially in using machine learning to correlate logs, metrics, and traces across monitoring and alerting systems. IT service management and SREs have been using AIops to reduce the mean time to resolve incidents and to perform accurate root cause analysis (RCA) efficiently. Agentic ops is the next wave of genAI operational capabilities, including tools for monitoring AI agents, managing their access rights, and detecting AI model accuracy drift.
“AI is useful during major incidents because it can pull together a lot of context into a few clear sentences, which is exactly what an SRE needs in the moment,” suggests Shani Shoham, chief revenue officer at OpenObserve. “The complexity of architecture and the different tooling make it easier for AI than for a human, but autonomous resolution is still a way off.”
The business pressure to keep systems up, secure, and performing well is a 24/7 stressful responsibility. According to The SRE Report 2025 from Catchpoint, 36% of SREs often or always experience elevated stress during an incident, and 28% said the stress persists even after the incident is resolved. AI capabilities may prove to be a game-changer in helping SREs avoid burnout and reduce stress.
“AI can improve RCA by taking in a much larger incident context than any engineer can hold at 3am, reasoning across traces, logs, metrics, deploys, config changes, alerts, ownership, and recent production behavior,” says Noam Levy, founding engineer and field CTO at Groundcover. “Beyond attempting a full RCA, its immediate value is distilling the signals that actually matter, reconstructing a clear timeline of cause and effect, and helping engineers separate correlation from likely causality. Once a fix is deployed, agents can also verify remediation by comparing pre- and post-fix behavior, but this depends on broad access to rich, correlated production signals and a cost model that does not discourage adoption or experimentation.”
Not only are incidents resolved faster and with less stress, but AI can also free up SRE time to focus on proactive work and create a career path for junior developers into SRE roles. Quais Taraki, CTO at EDB Postgres AI, adds, “AI reduces toil by automating repetitive tasks while accelerating incident resolution through copilots that correlate signals across distributed systems, allowing SREs to focus more on resilience strategies like chaos engineering and failure analysis.”
AI can have long-lasting operational impacts, especially for organizations looking to deploy more mission-critical technology and AI capabilities. Two longer-term benefits of AI for SREs are reducing the number of bridge calls needed for incident response and the number of engineers required in “war rooms” to coordinate root cause analyses.
“When something goes wrong, AI that guides SREs can do the full analysis, get to the root cause, and perform the remediation,” says Spiros Xanthos, founder and CEO of Resolve AI. “AI also helps avoid many escalations, and when escalations are needed, it targets the right people from the network, infrastructure, and the application teams. AI for SREs centralizes operational intelligence, exposes tribal knowledge, and can guide more junior developers.”
While AI capabilities have been a net positive in helping SREs improve system reliability, the growth of AI code generators, vibe coding, and spec-driven development is adding to their workloads. According to one study, 41% of all global code is now AI-generated, and Gartner predicts that 40% of new enterprise production software will be created using vibe coding techniques by 2028. But coding velocity is creating new issues for SREs as AI pull requests have 1.4 times more critical issues and 1.7 times more major issues, according to CodeRabbit. “AI-assisted development has created an unprecedented velocity of code reaching production, expanding surface area, edge cases, and failure rates faster than traditional SRE practices can absorb,” says Vinod Jayaraman, cofounder and CTO at NeuBird AI. “The speed of shipping has far outpaced the speed of understanding what breaks in production. To close this loop, SREs need enterprise agents that can capture precise diagnostic context, including correlated traces, service dependencies, and anomaly timelines, and structure it as actionable input for the engineers and AI coding tools responsible for the fix.”
The growing number of AI agents deployed to production creates new challenges. AI agents are not just code; they have multiple failure points. They are built using language models, connect to proprietary sources for context, and integrate with Model Context Protocol servers to support more complex workflows. Changes are ongoing and not deployment events, so the SRE’s job of identifying the source of performance and accuracy drifts isn’t trivial.
“Traditional SRE was built for systems that fail in reproducible ways, but agents fail differently and drift when a model provider pushes an update, and behavior shifts silently with no baseline for comparison,” says Mohammed Aboul-Magd, vice president of product at SandboxAQ. “Most organizations can’t even answer the basics: how many agents are running, what they have access to, and whether they’re still doing what they were built to do.”
“Every time a senior engineer leaves, they take years of learned failure patterns with them, and the next outage starts from square one,” adds Ronak Desai, cofounder and CEO at Ciroos. “Using AI for compounding operational memory changes that, and every incident your system resolves, the AI learns it.”
SREs should take a leadership role in emerging best practices, including defining their standards for AI agent non-functional acceptance criteria, observability practices, and release-readiness criteria. SREs should update their service-level objectives (SLOs) and define error budgets for AI agents in production.
Ryan Downing, vice president and CIO of enterprise business solutions at Principal Financial Group, says, “Standard SLOs and error budgets give teams the guardrails, and AI helps interpret the telemetry against those targets, reducing noise so engineers can get to the real issue faster and automate parts of remediation before customers are impacted.”
The more dramatic shift in site reliability engineering is an evolution of its business scope. IT leaders focus on uptime, performance, and issue resolution, as well as understanding their impacts. Business leaders will look to IT and SREs to identify, determine root cause, and remediate a broader class of issues, including rogue AI agents and the impacts of rapidly deploying new agentic capabilities.
“AI agents are handing SREs categories of problems they’ve never had to solve before, specifically failures defined in business terms, not technical ones,” says Blake Sherwood, distinguished technologist for AI and platform strategy at Smarsh. “Traditional reliability engineering is built around latency, errors, and crashes, but agents now fail due to skipped compliance steps or outcomes that looked fine technically but were wrong contextually. Most SRE teams aren’t wired for that yet.”
The question is whether SREs with AI-augmented tools can keep up with the velocity, complexity, and business urgency of deploying new AI business capabilities.