What is SOAR, and is it still worth buying?
SOAR (security orchestration, automation, and response) connects security tools and runs playbooks on alerts. How it works, the main products, and where playbooks break.
Definition
SOAR
Also called Security orchestration, automation, and response, Security orchestration, automation and response, SOAR platform
SOAR (security orchestration, automation, and response) is security software that connects a SOC's tools through integrations and runs playbooks, predefined workflows that enrich, triage, and respond to alerts, alongside case management and threat intelligence handling. Gartner coined the term in 2017. Today most SOAR capabilities ship inside SIEM and XDR platforms, and AI agents are taking on the investigation work that playbooks had to script step by step.
How SOAR works
A SOAR platform sits between the tools that raise alerts and the tools that act on them. Gartner's market definition lists what a SOAR product has to provide: customizable workflows that turn repeatable tasks into playbooks, storage for incident data, manual and automated triggers, a way to operationalize threat intelligence, and support for a broad range of existing security tools. In practice that comes down to four parts.
- Integrations. Connectors to each product's API, so the platform can read from a SIEM or email gateway and act in an EDR, identity provider, or ticketing system. Splunk SOAR calls these apps: a MaxMind app provides a
geolocate ipaction, an Okta app providesset passwordandenable user. - Playbooks. Workflows that chain those actions with conditions, loops, and manual approval steps. A trigger starts them, usually a new alert or case.
- Case management. A record of each incident: the alerts, entities, evidence, analyst notes, and actions taken.
- Threat intelligence. Indicator lookups and reputation data that playbooks use to enrich alerts.
A typical phishing playbook shows how the pieces fit. A user reports an email. The playbook pulls the headers and URLs, checks each against threat intelligence, detonates attachments in a sandbox, searches mailboxes for copies, and then either closes the report as benign or deletes the messages, blocks the sender, and opens a case for an analyst. Every branch is one the author wrote in advance.
Common SOAR use cases
SOAR fits high-volume work with known steps. The SANS 2024 automation survey found phishing response is the most commonly automated security process, automated by 52% of organizations, followed by vulnerability management (43%) and data enrichment (42%). Forrester's advice to new SOAR buyers is to automate consistent, repeatable processes and start with enrichment.
Other common playbooks disable a compromised account, isolate an endpoint, block an indicator at the firewall, open and sync tickets in an IT service management system, and notify on-call staff. Most of them are short, well-defined response actions that an analyst has already decided to take.
SOAR products and how they build playbooks
Each major product describes its playbooks differently. These summaries come from each vendor's own documentation, including the part each design leaves to your team.
- Splunk SOAR. A visual editor links actions provided by apps, without writing code. The actions a playbook can use are determined by the apps integrated with your instance, so a tool without an app is out of reach until someone adds one.
- Cortex XSOAR. Playbooks are task-based graphical workflows that can be fully automated, fully manual, or in between, and content ships in packs. Automated tasks are scripts written in Python or JavaScript, so custom automation means writing and maintaining code.
- Microsoft Sentinel. Automation rules tag, assign, or close incidents without a playbook. Anything more involved is a playbook built in Azure Logic Apps, which creates separate Azure resources that can carry additional charges, and each Logic App authenticates separately to every resource it touches, including Sentinel itself.
- Google Security Operations SOAR. Alert playbooks run per alert. Case playbooks run once across a whole case, so a case with ten similar alerts doesn't enrich the same IP address ten times. Playbooks can act on verdicts from the Gemini Triage and Investigation Agent, which has a quota of "typically around 10 investigations per hour per tenant"; alerts above it need manual triage.
- Tines. Tines Stories, a visual no-code builder, builds workflows called stories from eight types of actions. Tines 3B is a newer code-first environment, and Tines says that when AI writes the workflow, the no-code abstractions Stories was built on are "unnecessary and cumbersome."
- Torq. Torq calls itself an AI SOC platform that runs AI agents alongside deterministic workflows. Each agent execution is capped at 25 tool calls, and Torq says agent outcomes may vary between runs.
The market's own labels show where this is heading. Gartner's Peer Insights page now titles the category "Security Orchestration, Automation and Response Solutions (Transitioning to Security Information and Event Management)."
Where deterministic playbooks fall short
A playbook only handles the cases its author anticipated, and every one of them has to be maintained. Three limits show up repeatedly in the research.
Branches get brittle. A Forrester analysis of SOC automation notes that each added step in an automation chain narrows the cases a playbook applies to. That works "for consistent inputs," but SOC inputs change with every new attack technique, tool, and business process. When an alert arrives with a field missing or a pattern nobody scripted, the playbook stops or takes the wrong branch.
Maintenance never ends. In the SANS 2024 survey, about 68% of respondents named the engineering effort to deploy and maintain automation as the most challenging attribute of SOAR, and 48% named software costs. Forrester found teams often implement "a maximum of five to 10 playbooks in total over the first several years," and that "often, a lone security analyst maintains SOAR." Gorka Sadowski, who helped define SOAR at Gartner in 2017, describes organizations with hundreds of playbooks to write, test, tune, and retire, where every new tool or exception sent the playbook team back to work: "a never ending story and ROI was getting progressively worse."
Novel alerts fall through. Torq's documentation puts it directly: "Traditional deterministic workflows require defining every condition and edge case, a rigid, time-consuming process." An alert type nobody wrote a playbook for goes to the analyst queue, which is where alert fatigue builds. Routing alerts to an agent instead changes where that alert ends up: in Cotool, a response agent investigates each alert from any source as it arrives and closes or escalates it with the evidence attached, whether or not anyone scripted that alert type (Cotool).
The category label has suffered too. Gartner's 2024 Security Operations Hype Cycle marked SOAR "obsolete before plateau," though Gartner analyst Eric Ahlm told Dark Reading the call targeted vendors that sell only dedicated automation platforms, not automation itself, which SIEM and XDR products have absorbed.
How AI SOC agents differ from SOAR
An AI agent decides its next step at run time instead of following a branch someone wrote in advance. Given an alert, it chooses which logs to query and which indicators to check, reads the results, and keeps going until it can explain a verdict. That covers the investigation work playbooks struggle to script, while the response actions themselves can stay deterministic and gated by approval.
SOAR vendors are adding agents on top of the playbook model rather than replacing it. Palo Alto Networks' Cortex AgentiX turns existing playbooks, scripts, and commands into actions that agents plan with, so that automation content stays part of what agents run and what someone maintains. Google Security Operations' agentic automation passes the Gemini agent's verdict into predefined playbooks, within the quota above. Torq says its AI agents suit cases that would otherwise need complex branching, and recommends deterministic workflows when results must be identical every time. Cotool starts from the agent: a response agent is written in plain English, such as an OAuth grant responder that lists the grant's scopes in Google Workspace, checks the publisher's domain in VirusTotal, traces what the token read and changed, and asks #security in Slack before revoking the app everywhere (Cotool). The investigation steps are the agent's choice; the revocation is the step a person approves.
A playbook still fits when a few well-understood alert types, such as reported phishing or account lockouts, make up most of your volume and you need the same audited action every time. It runs the same steps on every alert, each one written and reviewed in advance, and it carries the maintenance load described above. For the broader architecture, see what an AI SOC is.
Playbooks as agent skills
Cotool is built agent-first: agents pick up every alert, gather context across your tools, and close or escalate it with the evidence attached.
Take the reported-phishing playbook from earlier, and a report it didn't anticipate: the link points to a file-sharing service the company uses, so the reputation check comes back clean and the playbook takes its benign branch. In Cotool, the triage agent picks up the same report and chooses what to check next: who else received the message, whether the recipient signed in from somewhere new after clicking, and whether the endpoint saw a download. It closes or escalates with those results attached. Deleting the messages or suspending the account waits for approval in the app or Slack.
Instead of rebuilding playbooks as branches, you attach your runbooks to agents as skills, versioned packages of runbooks, reference docs, and scripts, and Cotool can import existing Tines stories as agents. Agents and skills live in Git as YAML with pull-request review and rollback, and analyst feedback becomes instruction changes you review. In Cotool's documented format, the phishing playbook becomes one agent file that references a skill directory, cotool/skills/phishing/, holding a SKILL.md and the old playbook as references/playbook.md (Cotool docs):
apiVersion: cotool.ai/v1
kind: ResponseAgent
metadata:
sync_key: triage-reported-phishing
name: Triage Reported Phishing
spec:
modelAlias: anthropic:chat:sonnet-4.6
systemPrompt:
file: triage-reported-phishing.md
skills:
- phishing
acceptanceCriteria:
- Checks every recipient of the message, not only the reporter.
triggers:
- source: webhook
name: Reported phishing
Cotool polls the repository about every 5 minutes, records each synced change as a version tagged with its commit, and keeps the last good version when a file fails validation. That moves upkeep into code review rather than removing it: someone still edits the skill when the process changes, but there are no branches to rewire.
Related terms
- SIEM: security information and event management; collects logs and runs the detections that feed SOAR. See legacy SIEM vs modern SIEM.
- Playbook: a predefined workflow of actions and conditions that runs on an alert or case.
- XDR: extended detection and response; vendor platforms that correlate endpoint, identity, email, and cloud alerts, and increasingly include automated response.
- AI SOC: a security operations model where AI agents investigate alerts. See what an AI SOC is.
- Alert fatigue: analysts tuning out alerts because most aren't worth acting on. See what alert fatigue is.
- Defense factory: a continuous, agent-first operation that finds, validates, and fixes security weaknesses on a loop. See what a defense factory is.
FAQ
- What is the difference between SIEM and SOAR?
- A SIEM collects and stores logs and runs detections that produce alerts; a SOAR platform takes alerts and runs playbooks that enrich them, open cases, and act in other tools. The SIEM decides what is suspicious, and SOAR decides what happens next. Most SIEMs now include SOAR features, such as playbooks in Microsoft Sentinel and Google Security Operations.
- Does SOAR use AI?
- Classic SOAR does not need AI: a playbook follows the conditions and steps its author defined. Most SOAR products now add AI, for example AI prompts as a step inside a Cortex AgentiX playbook, or AI agents that choose their own steps in Torq and Cortex AgentiX. Torq's own documentation still recommends deterministic workflows when a use case needs identical results every time.
- What does SOAR stand for?
- SOAR stands for security orchestration, automation, and response. Orchestration means connecting separate security tools so one workflow can use all of them, automation means running those steps without a person, and response means acting on an incident, such as isolating a host or disabling an account.
- Is SOAR dead?
- SOAR as a standalone product category is fading, but security automation is not. Gartner's 2024 Hype Cycle marked SOAR "obsolete before plateau," and a Gartner analyst told Dark Reading that the call was about vendors selling only dedicated automation platforms, since SIEM and XDR products have absorbed SOAR features.
- What are examples of SOAR tools?
- SOAR tools include Splunk SOAR, Palo Alto Networks Cortex XSOAR, the automation rules and Logic Apps playbooks in Microsoft Sentinel, Google Security Operations SOAR, Tines, and Torq. Swimlane and Fortinet's FortiSOAR are also listed in Gartner's SOAR market.
Sources
- Security Orchestration, Automation and Response Solutions Reviews and Ratings - Gartner Peer Insights · accessed November 16, 2026
- SOAR Is Dead, Long Live SOAR - Dark Reading · accessed November 16, 2026
- SOAR is dead, long live the SOAR - Gorka Sadowski · accessed November 16, 2026
- The State of Automation in Security Operations: A SANS Survey · accessed November 16, 2026
- Planning Is Paramount When Adopting SOAR - Forrester · accessed November 16, 2026
- The "Autonomous SOC" Is A Pipe Dream - Forrester · accessed November 16, 2026
- Use playbooks to automate analyst workflows in Splunk SOAR (Cloud) · accessed November 16, 2026
- Cortex XSOAR Concepts · accessed November 16, 2026
- Get Started with Cortex AgentiX · accessed November 16, 2026
- Automation in Microsoft Sentinel · accessed November 16, 2026
- Azure Logic Apps for Microsoft Sentinel playbooks · accessed November 16, 2026
- Case playbooks overview - Google Security Operations · accessed November 16, 2026
- Respond to alerts and cases - Google Security Operations · accessed November 16, 2026
- Actions - Tines Stories docs · accessed November 16, 2026
- Introducing Tines 3B, the next generation of Tines · accessed November 16, 2026
- Torq AI SOC Platform · accessed November 16, 2026
- AI Agents FAQs - Torq Knowledge Base · accessed November 16, 2026
- AI in Torq - Torq Knowledge Base · accessed November 16, 2026
- AI SOC Without the Black Box - Cotool · accessed November 16, 2026
- AI Alert Triage and Incident Response Agents - Cotool · accessed November 16, 2026
- Response Agents as Code - Cotool Documentation · accessed November 16, 2026





