Details
AI SRE readiness begins with the quality and completeness of an organization’s observability data—not with deploying an autonomous agent. In this 10-Minute Take, Elastic’s Frank Swain presents a maturity model that progresses from manual queries to natural-language interaction, generative AI, agentic workflows, and eventually trusted autonomous operations.
The session explains how consistent telemetry, composable skills, tool calling, and Model Context Protocol servers can give agents increasingly useful capabilities while allowing teams to expand autonomy gradually.
Agents can only correlate signals effectively when logs, metrics, and traces use compatible schemas and structures. This consistency enables end-to-end reasoning across the environment.
Data-retention decisions also matter. Telemetry sampled or discarded to reduce storage costs could become the missing evidence needed during a future investigation.
Elastic’s maturity model moves through five broad stages:
Manual queries and human-led investigations
Natural-language interaction with observability systems
Conversational investigations powered by generative AI
Agentic workflows involving tools, memory, and skills
Autonomous AI SREs that investigate, learn, and take approved actions
Teams should assess their current capabilities and pursue the next practical stage instead of attempting full autonomy immediately.
Skills package repeatable workflows into composable units that an agent can invoke when an alert occurs.
Rather than inventing an investigation process each time, an agent can use these predefined workflows to gather evidence, analyze conditions, or begin a controlled response.
Tools and callable functions let agents move beyond conversation and perform limited operational tasks. Depending on their permissions, agents may pull telemetry, query logs, execute scripts, trigger runbooks, or initiate remediation steps.
These mechanisms are essential for autonomous operations, but access should expand only as reliability and trust are established.
Model Context Protocol servers provide a standardized way for agents to discover and invoke available capabilities across systems.
Within an observability workflow, MCP can help connect agents to telemetry sources, hosts, scripts, and runbooks without requiring engineers to leave their development or operational environment.
Fully autonomous operation is the final stage of the maturity curve, not the starting point. Teams can begin with read-only access and composed workflows before granting agents permission to execute scripts or make changes.
This staged approach allows organizations to validate performance and build confidence before increasing an agent’s operational authority.
Data quality determines agent quality
Practical implication: Standardize schemas across logs, metrics, and traces before expecting agents to conduct reliable cross-system investigations.
Retention policies affect future investigations
Practical implication: Evaluate whether sampling and storage policies could remove signals an AI SRE would need for root-cause analysis.
Skills make workflows repeatable
Practical implication: Package common investigative and remediation procedures as composable skills that agents can invoke consistently.
Tool access creates operational value—and risk
Practical implication: Begin with narrowly scoped, read-only tools and expand permissions only after the agent has demonstrated dependable behavior.
MCP can reduce integration friction
Practical implication: Use MCP servers to expose approved observability and operational capabilities through a standardized interface.
AI SRE adoption is a progression
Practical implication: Identify the organization’s current maturity stage and define the technical and governance requirements for advancing one stage at a time.
Q: What is machine-speed observability?
A: Machine-speed observability uses AI agents to analyze telemetry, investigate conditions, and potentially initiate remediation faster than a human-led operational process can.
Q: What foundation does an AI SRE need?
A: An AI SRE needs consistent and sufficiently complete logs, metrics, and traces. Compatible schemas and structures allow the agent to correlate evidence across the environment.
Q: Why can telemetry sampling create problems for AI agents?
A: Sampling can remove the evidence needed to reconstruct an incident. An agent cannot investigate a signal that was discarded before the investigation began.
Q: What is the observability AI maturity curve?
A: It is a progression from manual investigation to natural-language interaction, generative AI assistance, agentic workflows, and ultimately autonomous AI SREs with configured authority to act.
Q: How do skills support AI SREs?
A: Skills package operational workflows into reusable capabilities. Agents can invoke them to conduct consistent investigations or perform approved response procedures.
Q: What role does MCP play in observability?
A: MCP provides a standardized way for agents to discover and invoke approved capabilities, including querying telemetry, accessing hosts, and triggering runbooks.
Q: How are tools different from skills?
A: Skills generally represent packaged workflows or procedures. Tools and functions provide the specific interfaces an agent uses to query systems or perform actions.
Q: Should an AI SRE immediately receive permission to change production systems?
A: No. The session recommends progressing gradually by introducing skills and limited tools before granting broader permissions or autonomous control.
Frank Swain: Hello, my name is Frank Swain. I’m a product marketing manager at Elastic, and this is the second video in a three-part series.
Today, we’re going to talk about how to prepare for AI SREs and machine-speed observability.
In the last video, we explained what AI SREs are. But how do you actually get started with them?
Like most things, you start with the foundation. But what is that foundation?
The most important thing is quality, consistent data. Agents can only correlate telemetry when it lines up. They need consistent schemas and structures across logs, metrics, and traces. That is what makes end-to-end reasoning possible.
Another important consideration is the ability to keep every signal. Use a data store that doesn’t force you to drop telemetry to control costs. If you sample data to save money, that could create a gap in tomorrow’s investigations.
The next step is to sequence your investment. Decide where to start by assessing where you are on the maturity curve and what makes sense to pursue next.
With that in mind, we’ve created this observability AI maturity curve. It looks at the degree of automation compared with the risk of inaccuracy.
If you’re starting this journey today, you’re probably still performing manual queries. Human SREs manually log in to systems to retrieve analytics, dashboards, and visualizations, make decisions, and conduct their investigations.
The next step is to introduce a natural-language system. Instead of writing queries in a potentially obscure language, you can use natural language to interact with the system.
Beyond that, you can connect the system to an LLM or generative AI platform. This creates a chat experience in which you can conduct a conversational investigation of your systems.
The next stage is agentic AI. At this stage, AI agents can use tools, build memory, apply skills, and potentially involve multiple agents in a workflow.
The final stage is a fully autonomous AI SRE. Agentic SREs can conduct independent investigations, learn, and operate with a configured level of autonomy that allows them to take action in the environment.
Here are a few tips for advancing along the maturity curve.
When you’re starting with agents, the most straightforward approach is to give them composed or composable skills. These package workflows that can be activated when an alert is triggered. The agent can then respond using those defined sets of skills.
The next step is to add tools and specific functions that agents can call to perform limited actions in the environment and begin resolving issues more autonomously.
Model Context Protocol servers, or MCP servers, are also part of this progression. They give agents access to capabilities such as triggering runbooks and taking action in the environment.
Beyond that, you can provide agents with increasing capabilities and permissions. This can include access to hosts, the ability to retrieve telemetry and query logs, and even permission to execute scripts that take action in the environment.
If you take one thing away from this video, make it this: The agent is only as good as the ground it stands on.
Machine-speed observability doesn’t start with the AI SRE. It starts with data that is correlated end to end, consistent in structure, and complete. An agent cannot investigate a signal that you dropped to save on storage.
Get that foundation right first, and everything else on the maturity curve opens up: smarter root-cause analysis, proactive investigation, and eventually autonomous action that you can actually trust.
If you want a practical map for sequencing that investment, download our guide, Transforming Observability With Agentic AI and the Autonomous AI SRE. There is a QR code on the screen and a link in the description.
Thanks again for watching. See you next time.
Presenters:
Frank Swain
Principal Competitive Intelligence Manager, Elastic
Join Now for More Content & Events
For event and sponsorship inquiries, please email: [email protected]
