Industry executives and experts share their predictions for 2026.  Read them in this 18th annual VMblog.com series exclusive.

By JJ Tang, Co-founder & CEO at Rootly

Most of the attention over the last two years has focused on how engineering teams can leverage GenAI to build software faster. That’s real, and it’s happening at scale. But in 2026, I predict we’ll see greater attention going toward the other half of software development that still consumes an enormous share of time and budget: fixing.

Up to 84% of developers now use AI tools, and Cursor alone was seeing 1 billion lines of code accepted per day in April 2025, at a time when a few billion lines of code are generated worldwide every day. In other words, it is reasonable to assume that much of the code being written today is produced by, or at least alongside, GenAI.

But writing code is only half the work. Developers still spend more than a quarter of their time fixing bugs, and incident response has grown into an entire industry. As engineers figure out how to build at scale with AI, they will soon want to learn how to fix at scale with AI. And the good news is, according to open source benchmarks, AI is getting better at SRE-related tasks.

Prediction #1: AI SRE will become the default first responder for incident troubleshooting

A category is emerging to address this directly: tools specifically designed to help incident responders troubleshoot production issues in real time. You’ll see this described as AI SRE, and it will become a standard part of the modern on-call stack.

Generally, AI SRE tools connect to the systems engineers already rely on: infrastructure, monitoring, logging, deployments and code versioning, chat, knowledge bases, and incident management platforms. Their goal is to accelerate the part of incident response that consumes the most time: gathering evidence, correlating signals, and forming a hypothesis quickly enough to act.

When an incident is triggered, AI SRE systems ingest the incident payload and generate an investigation plan: what to inspect, what changed recently, which services are likely involved, and which signals are correlated. From there, they will compare the event to historical incidents and look across current telemetry to identify likely causes.

For example:

• By analyzing recent GitHub changes, it may spot a breaking change correlated with symptom onset;
• By inspecting Chronosphere observability signals, it might detect a resource bottleneck or saturation event in a Kubernetes cluster; and
• By searching Rootly’s past incident data, it can surface a similar failure mode and a previously successful mitigation.

The output is a summary of what changed, the likely culprit, and why that conclusion is plausible. The most advanced AI SREs will suggest a mitigation that responders can validate and implement quickly.

To be clear, I don’t expect AI SRE to fully resolve complex incidents in 2026. Things like multi-system failures, ambiguous symptoms, or cascading infrastructure events will still require human judgment. But it will dramatically compress time-to-triage and reduce “search time”. For low-complexity incidents, you can expect full resolution without human intervention.

Prediction #2: Agents embedded in CI/CD will become a standard layer for operational risk detection

Today, most GenAI use is synchronous: engineers prompt coding assistants as they work. In 2026, that model will feel incomplete. The next wave will be asynchronous: agents operating continuously in the background, especially inside CI/CD and infrastructure change management.

A practical example is an agent embedded in the CI pipeline to assess risk in a pull request. For example, if an engineer submits a change to a Terraform file, the agent can flag risks such as resource deletion, privilege changes, capacity shrinkage, and configuration drift.

Another example is “consistency agents“: tools that ensure code and infrastructure align with internal style guidelines, architectural standards, and compliance expectations. Because documentation is almost always stale, some teams adopt scheduled agents that continuously verify that runbooks and service documentation still match reality.

Prediction #3: GenAI can help us address new issue types

Another interesting trend I see emerging is engineering teams tackling new types of issues. Ones of them is responder wellbeing, which could become an operational metric and not just a conversation topic.

If you’ve ever been on-call, you’ve seen it: responders who are stressed, sleep-deprived, and drifting toward burnout. In many organizations, this is still treated as “part of the job,” even though operational fatigue directly impacts reliability, response quality, and retention.

The challenge is that managers have historically had limited visibility. It’s difficult to track how a team or individual is doing without waiting for a clear signal: missed alerts, declining performance, or attrition. Check-ins help, but employees may underreport stress, normalize overload, or fail to recognize how unhealthy the baseline has become.

But there is a way for systems that aggregate signals from incident management, paging frequency, alert volume, Slack activity, workload distribution, and after-hours engagement. That data can then be fed into models that flag patterns a manager should investigate: repeat escalations on the same individuals, disproportionate paging load, sustained after-hours work, increasing mean time to resolve, or “high churn” on low-quality alerts. The Rootly AI Labs built an open-source implementation of that, the project is still in beta but v1 is planned for 2026!

Closing

In 2026, building with AI will remain the headline, but fixing with AI will become the next frontier. As engineers figure out how to build at scale with AI, they will soon want to learn how to fix at scale with AI. The organizations that treat reliability, risk, and on-call sustainability as first-class AI problems will be the ones that turn GenAI into durable operational leverage.

##

ABOUT THE AUTHOR

JJ Tang is the Co-founder and CEO of Rootly, an AI-native platform for on-call and incident response.

Rootly powers incident operations for teams at NVIDIA, Dropbox, LinkedIn, and Canva.