Market Analysis

AI-SRE : Another AI tool nobody asked for?

November 10, 202610 min read
AI-SRE : Another AI tool nobody asked for?

Through the advent of technology, the software industry has always found a new branding or buzzword to put in front of every existing product. A simple example of this would be how the Salesforce flagship product pivoted from Salesforce Cloud CRM to Einstein ML to AgentForce. 

As vendors market a new category called AI-SRE, the important question is whether this represents a genuine shift in reliability engineering or simply another layer of LLM-powered summarization over existing observability tools

To answer that question, we need to look beyond the hype and examine what AI-SRE actually does, what problems it solves, and where it meaningfully differs from traditional observability and incident response tools.


What to expect from an AI-SRE tool?

There are many different ways I could describe an AI-SRE. What stands out to me is what I actually need in production versus what is easy to build using claude code in five minutes.

What I need is something that works like an actual on-call engineer. It checks the issue, goes through the logs, examines every piece of code one by one, finds the problem, determines the root cause, and fixes it to resolve the issue. What I do not need is something that simply summarizes all alerts with an LLM call and derives a root cause from that, which could be completely unrelated.

This is where AI comes in.

When a single person does this, it can take hours. But an agentic system that has full context of my infrastructure can perform the same process while running multiple failure modes and hypotheses simultaneously, going through all of my logging and metrics data to find the root cause and providing resolution within minutes, hence saving significant amounts of money that would otherwise be lost to incident downtime.


The real risks : Privacy / Skill Erosion

Before this vision can be adopted, many concerns must be addressed. 

Obviously, building an AI system that goes through all of your sensitive data comes with drawbacks. The main issue is that production logs often contain deployment keys, unmasked PII, and internal service endpoints. AI data privacy incidents jumped by 56% in 2024. As a result, the need for a vendor that queries your data on demand and never stores it is higher than ever.

Another concern is the learning curve that AI is potentially destroying.

The skill gap in many video games decreases because you keep losing to better players, learn what they are doing, and eventually improve. But if there were an AI that simply pushed you to the highest rank, you would never be able to compete on your own.

The same applies to junior SREs and developers trying to analyze incidents. You gain seniority only after seeing many incidents, making mistakes, facing setbacks, and sometimes coming close to losing your job because you were unable to identify the root cause quickly enough. These experiences are what build real expertise.

When you rely on an AI system that solves everything for you, you miss out on the technical depth and experience needed to understand and resolve issues independently. Instead, you risk blindly following what the AI says and gradually losing control over your codebase.


Market Validated : The category is now real.

Despite these concerns, the market has matured enough to justify its own category.

Gartner published its first dedicated AISRE market guide in January 2026. This was not an AIOps guide; it was specifically focused on AI-SRE. That is the clearest signal that the category has formally separated.


The Competitive Landscape : Incumbents vs Startups

With the category established, the next question is who is best positioned to build it.

Large companies such as Dynatrace, Datadog, New Relic, and Grafana Labs are the incumbents. They dominate the market, and their primary revenue comes from the sheer volume of data they process. The more logs, metrics, and traces they ingest, the stronger their observability and root cause analysis capabilities become, and the more revenue they generate.

In contrast, startups such as Resolve AI and NeuBird tend to excel in investigation depth. With limited resources, they perform very deep analysis across the codebase and produce highly focused outputs.

In my opinion, the ideal AI-SRE tool would combine both capabilities. It would be able to process large volumes of data while also conducting extremely deep investigations. In other words, it would combine breadth-first and depth-first search to identify the true root cause.


Engineering Challenges : Parallelization and Compute Costs

Delivering this vision at scale introduces significant engineering challenges.

The challenge is that running investigations across vast amounts of data with AI requires significant parallelization, with multiple hypotheses being tested simultaneously.

In 2026, the compute cost of using an LLM for this type of analysis is usually negligible compared to the cost of even a single hour of downtime. However, many teams fail to realize that these costs compound as usage scales.

An AI system that continues investigating a non-existent issue can generate a surprisingly large bill before anyone notices.

This concern becomes even more significant given recent findings that AI-generated code has increased the number of security and reliability issues tenfold over the past year.

We are shipping code to production at an unprecedented pace, and our monitoring systems are struggling to keep up. As a result, more things are breaking.


The biggest gap : Resolution instead of Prevention

Yet the most important limitation is not technical but conceptual.

The biggest gap I have found in the market is that nearly all AI-SRE tools focus on responding to incidents rather than preventing them.

Everyone is racing to respond to incidents faster. Nobody is asking why we keep having the same incidents.

If AI is writing your code, reviewing your pull requests, and monitoring your systems, then when something goes wrong, who actually understands what failed?

In my experience, when something breaks, we often end up searching through the entire codebase instead of knowing exactly where the issue originated. This is not a situation I want to face when multiple users are relying on my platform simultaneously.


Vision for what AI-SRE should become

The future of reliability is not AI that reacts faster. It is AI that helps systems fail less often and helps engineers understand why. The most valuable AI-SRE platforms will not replace engineering judgment; they will strengthen it by transforming incident response from reactive firefighting into continuous prevention

That is what a true AI-SRE tool should be.


Author

Tanay Agarwal

Tanay Agarwal

Software Engineer