There's a pattern in security that's easy to miss if you only look at product features. The biggest jumps in this industry haven't come from better versions of an existing category. They've come from a new underlying capability arriving and resetting what the category even means. CrowdStrike is the clearest example we have from the cloud era, and we think the same pattern is happening again right now, one layer up the stack.
Big data didn't make antivirus better. It replaced the methodology.
Legacy AV, McAfee-era, was built on a per-endpoint model. Signature match, heuristic score, decision made locally, on the box, largely forgotten once made. Every improvement inside that model was still bound by its baseline: each endpoint reasoning alone, with whatever local intelligence it shipped with.
CrowdStrike's move wasn't a better version of that model. It was a different one entirely, made possible by cloud-scale data infrastructure becoming viable: route every endpoint's telemetry into one centralized graph, so detection stops being a per-machine decision and becomes a population-level one. An anomaly in one tenant's environment becomes a queryable data point across millions of endpoints. That's not local detection tuned harder. It's a new baseline for what a detection engine is.
Big data made that baseline possible. Without cloud-scale storage and query infrastructure, the centralized graph doesn't exist, and neither does the company built on top of it.
AI isn't making security service management faster. It's making it a different category: agentic case management.
We want to be precise about what we mean here, because it's easy to hear "AI in the SOC" and assume the pitch is a faster SOAR. It isn't. SOAR's model, however automated the execution, is still built on rules written in advance and cases resolved one at a time by whoever picks up the ticket. That model's ceiling was never execution speed. It was context: a playbook only knows what its author anticipated, and a ticket only carries what the analyst has time to gather before they act.
Agentic case management starts from a different question. Not "how fast can we execute a predefined action," but "how much relevant context can we bring into this specific decision, and can we make that same quality of decision every single time, regardless of who's on shift or how deep the queue is." Consistency and context are the actual product. Speed is a side effect.
A concrete case: the "impossible travel" identity and cloud anomaly
A senior DevOps engineer logs into Okta from San Francisco. Twelve minutes later, an OAuth token under their account authenticates from an AWS IP in Frankfurt and provisions a high-powered EC2 instance.
Ticket-and-rule methodology | Agentic case management | |
Triage | Evaluates the alert against a static rule set. If the IP isn't on an approved list, it fires a severity flag to the tier-1 queue. | Pulls context across domains as part of forming the decision: GitHub deployment logs, HR travel approvals, AWS CloudTrail, all queried together, not waiting on a rule that anticipated this exact pattern. |
Edge case | The engineer is on a flight using a corporate VPN router with a subnet the rule set doesn't recognize. The playbook can't parse it, fails silently or throws an exception, and the case lands back on a human with none of the legwork done. | Recognizes the network as a known corporate VPN node, verifies a Terraform deployment was triggered from the user's active session, and reconciles the identity match as part of the same decision. |
Resolution | Either a blunt company-wide session revocation that interrupts production, or a multi-hour wait in queue for an analyst to manually check AWS logs. | Identifies the new EC2 instance is missing standard security tags, applies the correct security group rules, and files a full contextual summary for human review, the same depth of judgment whether it's 2pm or 3am. |
The difference isn't that one path is automated and the other isn't. It's that one path can only act on what was anticipated in advance, and the other is reasoning across live context to reach a judgment call, the same kind of judgment a strong analyst would reach, applied with a consistency no queue-based process can sustain across every case, every shift.
Autonomy here doesn't mean unattended action. The system operates inside defined boundaries, weighing blast radius, proposing context-aware remediation, and keeping human approval where the decision genuinely warrants it. What changes isn't who's accountable. It's how informed and how consistent the judgment is before it reaches a human.
The pattern, stated plainly
Big data didn't upgrade antivirus. It created the conditions for a new detection methodology, and CrowdStrike built on top of that baseline before the rest of the market recognized the shift had happened.
We think AI is doing the same thing to security service management right now, not producing a faster ticketing tool, but establishing agentic case management as the next baseline: decisions made with more context, and made the same well every time. That's the bet we're building Treat on.



