Security teams adding agentic AI are asking whether an agent can work across the tools they already run, and the SIEM is rarely the part they want to rip out. The friction sits in the spaces between tools, where data lives in separate consoles and analysts stitch it together by hand.
In Strike48's H2 2026 survey of 534 security and IT leaders, 74% said they had agents in production. Data infrastructure was the most common barrier to broader deployment: 31% said they lacked the infrastructure to support agents.
Practitioners who've had to make that work describe the same short list of requirements. The agent can't become another console, it needs data that's been collected and normalized, and it earns its place by gathering context an analyst would otherwise chase across tools.
The pain sits between the tools
Most of the friction in a SOC comes from tools that don't share data, and the SIEM is only one of them. Most organizations run vulnerability, asset and configuration management as separate tools with no single place that maps one to another, and disconnected tools produce duplicate alerts that connected ones could merge.
For an agent, that fragmentation shows up as legwork. An investigation that should start from a single alert instead starts with pulling a host's vulnerability status from one console, its owner from another and its configuration history from a third, and an agent that can read only one of those sees only one slice of the picture. The integration question is which of those sources the agent can reach, and replacing the SIEM answers none of it.
Amelia Hewitt, Co-Founder and Director of Cyber Consulting at Principle Defence, sees the same pattern across her clients' business units. "Often those business units are siloed. They are not necessarily using the same tooling because it might not be operationally possible," Hewitt says.
Buyers screen out anything that adds a console
A tool that adds another portal to check loses before the evaluation starts, because someone on the team has to staff it. Benjamin Glover, Manager of Information Security at Cumulus Media, makes that his first screen. He asks whether a new tool becomes another management plane his team has to check every day.
"I want to have something that's integrating into my current stack or that I can migrate my current stack to, because adding more work to the people within the team is not going to make anyone's lives easier," Glover says.
In practice the screen comes down to where the agent's verdict lands. A verdict that arrives in the ticket queue and chat channels the team already works counts as part of the stack, and one that lives in a new dashboard counts as one more place to look.
Coverage and normalization come before the agent
An agent can only reason over the logs it can read in a consistent format, so coverage and normalization are the first integration tasks.
Md. Abul Kalam Azad, CISO at Eastern Bank PLC, starts his answer on agentic SOC readiness by saying the agent has to be integrated with the SIEM, SOAR and the other log sources. He calls deploying agents "quite difficult" because they need "lots of complexity, log normalization." A team that maps its log sources, formats and gaps before the pilot gives the agent the same view an analyst has.
The mapping itself is plain work. List every source analysts open during triage, such as the SIEM, endpoint and identity tools, cloud logs and the ticketing system, then note how each one stores timestamps, hostnames and user identities, and mark the sources the agent can't yet reach.
The agent's job is gathering context across tools
An agent earns its place on an existing stack by collecting context from across it, which spares analysts from moving between consoles to assemble an alert's story.
Mohit Bansal, Senior Manager of Security Engineering at Webflow, describes how his own team gets the context first when an alert warrants a ticket.
"AI get all the context for us and shoots us a ticket on an alert with all the data so that we are not hopping different tools and we are making that decision very fast," Bansal says.
That output keeps the SIEM central. Alerts still originate there, and the agent's work returns to the ticketing system analysts already use, so the SIEM holds its job as the place where events are collected and correlated while the agent takes over the legwork around each alert.
Start with one workflow and widen from there
The lowest-risk way to add an agent is to give it one workflow on the existing stack, compare its work with the analysts' own and widen its reach as the results hold up. A risk-driven SOC strategy gives a team a way to pick that first workflow, and because the agent reads from the SIEM and the tools around it, nothing has to be replaced for the pilot to run.
A useful pilot measures the agent against the work analysts already do. Compare the agent's verdicts and evidence with an analyst's on the same alerts, track how long each takes to reach a decision, and note every alert where the agent missed a source it couldn't reach. Those misses feed straight back into the coverage work described earlier, and the pilot widens only when they stop appearing.
"We build our agents to read from the SIEM, the ticketing system and the security tools a team already runs, so the first workflow can go live without a migration, and analysts keep working in the systems they already know," says Tim Leehealey, Co-Founder and President at Strike48.
Frequently asked questions
What does SIEM-agnostic mean for an AI SOC?
A SIEM-agnostic AI SOC works with whichever SIEM a team already runs. It reads the SIEM's data through integrations and doesn't require a particular platform.
How does an AI SOC connect to existing security tools?
Through APIs, prebuilt connectors or MCP servers that expose a tool's data to the agent. Applications without standard connectors, such as custom enterprise systems, need integration work.
What happens to existing SIEM detection rules when a team adds an AI SOC?
They stay in place. The SIEM keeps detecting and correlating events into alerts, and the agent investigates the alerts those rules raise.
Can an AI SOC work across more than one SIEM or log source?
Yes, if each source is integrated and its data is normalized into a form the agent can read. Coverage across sources determines the quality of the agent's verdicts.
When does replacing a SIEM make sense?
When the SIEM can't ingest the data the team needs or no longer fits its budget. Adding an AI SOC doesn't force that decision.




