Tier 1 is where SOC capacity runs out first. Working an alert by hand takes 30 minutes to two hours or more, and 62% of alerts never get investigated at all. Alert fatigue follows from that math. Analysts triage by severity, low-priority alerts wait indefinitely, and the one that mattered sits in the pile with the rest.
In Strike48's H2 2026 survey of 534 security and IT leaders, 82% said agents should handle Tier 1 work, but only 22% were ready to fully automate it. Asked which SOC task they would automate first, 24% chose alert triage and 21% chose investigation and root cause analysis.
Agents earn that trust by starting with a narrow set of cases and showing their evidence for checking.
Why Tier 1 breaks first
Tier 1 breaks first because every obstacle lands there. Outsourced analysts who hit a roadblock mid-investigation often stop investigating alerts and move on to the next one in the queue. "If they hit a roadblock, they often just won't fix that roadblock," one director of security operations says. "They'll just say I can't do this alert. And then it sits there."
Prioritizing by severity compounds the problem, since low-priority alerts wait longest and some go unread. Each one is a bet that it was noise.
What a Tier-1 agent does
A Tier-1 agent takes the first pass on every alert. It gathers the surrounding context, correlates related alerts into a single case, filters false positives, and decides whether anything needs to escalate to an analyst. Humans keep authority over containment, isolation and remediation, and each step leaves an audit trail. As one example, Strike48's SOC Level 1 Agent handles initial triage and investigation, and the company says its specialized agents correlate hundreds of alerts into unified cases, filter false positives and produce escalation documentation in minutes, with 200+ alerts becoming one correlated case in under eight minutes.
Webflow's triage workflow shows what analysts receive. "If you have to open a ticket with full context, perfect. That's how we do it today," says Mohit Bansal, Senior Manager, Security Engineering at Webflow. The AI collects the context 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."
Log coverage decides accuracy
An agent can only triage what it can see. Most agentic SOC products sit on top of an existing SIEM, so a team that ingests only 60% of its logs hands its agents the same blind spots, and ingestion costs often keep the rest out. Closing that gap means ingesting everything or querying logs where they already live, in Splunk, Elastic or S3.
AWS's published case study of Strike48 and its Amazon Bedrock-based platform draws the same line between coverage and trust, noting that without visibility into the entire log surface, security leaders can't trust agents to act on the complete data picture. With virtually 100% log coverage, the case study reports remediation more than 90% faster than traditional systems and about $3 saved for every $1 spent.
Where humans stay in the loop
Trust grows by limiting what agents can do alone. John Sapp, VP of Information Security and CISO at Texas Mutual Insurance Company, sees AI helping most at the first two tiers through a managed security service model. "I think where it best helps is Level One and Level Two," he says. "If you're using a managed security service such as we are, they're always going to be your Level One and Level Two, and because they're using AI in their services to decrease the number of alerts and to increase their positive finding rate and decrease their false positive rate."
Sunil Varkey, CISO at Hexaware Technologies, draws the line by blast radius. After an unusual-travel alert, Tier-1 analysts reset the password and revoke session keys, a routine action he expects AI to take over. Blocking a revenue-generating application exposed to the internet is a different call, and he's less sure about it. "Am I going to immediately block it completely out? I may not do that," he says.
What happens to the Tier-1 role
As agents take over first-pass triage, the Tier-1 role moves from working a queue to reviewing what the agents closed and picking up the escalations that need judgment.
Tronox, a titanium products producer, had an alert backlog from identity platforms, EDR, firewalls and threat logs, and its analysts built every investigation by hand. The Tronox team let Strike48 agents handle a small set of L1 cases first, then widened the scope. L2 agents now verify L1 closures, and a 25% random sample of closed cases is re-analyzed as QA.
"At first, we were thrilled when we got agents to autonomously close 42% of cases in a week," says Adam Chapman, Director of Cybersecurity at Tronox. "But after we saw we could fully trust their work, we decided to let them do more." Within a month, agents closed more than 75% of cases on their own, returning roughly 670 analyst hours, and overall MTTR fell 80%.
Strike48 puts automation at 98% of L1 work and 85% of L2 work. "Analysts should open a case and find the investigation already done, with the evidence laid out," says Tim Leehealey, Co-Founder and President at Strike48. "Their time goes to the decision, which is the part that needs a person."
Frequently asked questions
What does a Tier-1 agent need before analysts can trust its verdicts?
Grounding in evidence, baselines and confidence scores, plus the team's own playbooks. Showing the evidence behind each closure is what lets analysts check the work.
Can the agent triage alerts from logs I don't ingest today?
Yes. Search-in-place connectors read data in S3, Splunk, Elastic and other stores without migration, and agents query them in their native languages, such as SPL and KQL.
How long does deployment take?
Strike48's pre-built agents deploy in days. Shared SaaS goes live in minutes, isolated compute also goes live in minutes at a higher cost, and on-prem or air-gapped deployments take weeks. Teams choose the isolation and control level they need without changing the agent experience.
Do we need to replace our SIEM?
No. Strike48 runs as a layer on top of the existing stack and connects to hot and cold data across Splunk, CrowdStrike, Microsoft, Datadog, Snowflake and AWS.
How do teams limit what an agent can do on its own?
Through permissions the team sets, an approved list of tools each agent can call, and human-in-the-loop controls that define where an agent acts autonomously and where a human reviews first. Every action is logged, which gives analysts a record to audit.
Does an agent train on our alert data?
No. Strike48 says customer data never trains its models and keeps each tenant isolated. The platform is LLM-agnostic, so teams can swap models without re-architecting.




