What is Agentic Security Delivery?
Reviewed by ContraForce team
Security isn’t a product you buy. It’s an outcome someone has to deliver.
This is the field guide to the operating model replacing how that work gets done.
Agentic Security Delivery
Agentic Security Delivery is an operating model in which Security Delivery Agents execute the work of security, onboarding, triage, investigation, response, ticketing, reporting, and tuning, under human-governed standard operating procedures (SOPs), across every environment a team protects. The tools stay; the manual delivery layer between them is replaced by agents that act, governed by humans who decide.
The delivery chain.
Security is delivered in seven steps, repeated for every environment you protect. Automate one and the work pools around it. Delivery means running the chain, and closing the loop: every classified incident tunes the detections that fire next.
Onboard
Connect and baseline a new environment
Triage
Separate signal from noise
Investigate
Correlate evidence, form conclusions
Respond
Contain and remediate, inside approval gates
Ticket
Sync every action to your systems of record
Report
Evidence and outcomes, written down
Tune
Classified incidents sharpen the next detection
Agentic Security Delivery
Detection engineering runs as CI/CD.
Tuning isn't a project you finish at onboarding, it's a pipeline that runs on every closed incident. The classification an analyst assigns is the commit that starts the build.
Classify
Every incident closes with a verdict, true positive, benign, or false positive
Diagnose
Verdict patterns name the defect: a coverage gap or a noise source
Engineer
The rule is authored or refined as code, versioned, reviewed
Validate
Replayed against history: fires on the threat, quiet on known-good
Deploy
Pushed through the pipeline to every environment at once
Measure
Fire rate and precision watched; efficacy feeds the next pass
Measure feeds Classify
Detection got automated. Delivery never did.
Twenty years of security investment went into seeing more: more telemetry, more detections, more dashboards. It worked. Detection is a solved problem for most organizations, solved so thoroughly that the alerts now outnumber the people by four orders of magnitude.
Delivery never got the same treatment. The work between the alert and the outcome, onboarding, triage, investigation, response, ticketing, reporting, tuning, stayed manual, held together by analysts pivoting between consoles. Every client added multiplies that work; hiring is the only lever, and it stopped scaling years ago.
Agents changed the equation. Not assistants that summarize, and not scripts that replay, agents that do the work under enforced SOPs, with humans owning every decision that matters. When the delivery layer becomes software, the operating model changes. That model is Agentic Security Delivery.
The three shifts
Adopting the model means changing three defaults at once.
From tools to outcomes
Stop measuring what was deployed. Start measuring what was delivered: incidents closed, environments covered, reports shipped. The stack is a means; the outcome is the product.
From scripts to agents
Scripts replay the known. Agents investigate, adapt, and act inside guardrails you set, so the case nobody scripted still gets handled, the same way, every time.
From one environment to all of them
One standard, enforced everywhere. SOPs deploy centrally, execute locally, and leave an audit trail per environment, whether you protect five or five hundred.
Read the field guide.
What is the Security Delivery Gap?
The problem, named: the distance between security bought and security delivered, and the three ways it compounds.
Read the concept
What is a Security Delivery Agent?
Not a chatbot, not a script. The worker that investigates, decides within your guardrails, executes, and documents.
Read the concept
AI SOC vs Agentic Delivery
Four eras of security operations, and why automating one step of the chain is not an operating model.
Read the concept
The vocabulary of delivery.
New categories earn their names by defining them precisely. These are ours.
- Agentic Security Delivery
- An operating model in which Security Delivery Agents execute the work of security, onboarding, triage, investigation, response, ticketing, reporting, and tuning, under human-governed SOPs, across every environment a team protects.
- Security Delivery Agent
- A worker that performs security operations end to end, following codified SOPs within approval gates a human sets.
- The Security Delivery Gap
- The distance between the security an organization has bought and the security it actually receives.
- The Scale Tax
- The compounding operational cost every additional protected environment adds. Tools price linearly; operations cost compounds.
- Delivery chain
- The seven-step loop from a connected environment to a delivered outcome: onboard, triage, investigate, respond, ticket, report, tune, with tuning feeding back into the detections that fire next.
- The Tuning Loop
- The continuous cycle that closes the delivery chain: classify the incident, diagnose the gap or noise, engineer the rule change, validate it against history, deploy it everywhere, measure the result. Detection engineering, run as CI/CD.
- Gamebook
- A codified SOP, response steps, confidence thresholds, approval gates, that agents follow identically in every environment.
- Approval gate
- The configured point at which an agent must stop and route a decision to a human before acting.
- Human-in-the-lead
- The governance stance of agentic delivery: agents act autonomously below the gates; humans own every consequential decision above them.
Frequently asked questions.
What is an Agentic Security Delivery Platform?
An Agentic Security Delivery Platform runs the security work between a detection and a closed customer report. ContraForce is an Agentic Security Delivery Platform: Security Delivery Agents execute triage, investigation, response, tuning, ticket, and report across customer tenants, inside Gamebooks that define what each agent may do and when a human decides.
What is the difference between MDR, an AI SOC, and agentic security delivery?
MDR outsources the work to someone else. AI SOC tooling automates the investigation and stops at a verdict. ContraForce runs the whole delivery loop, triage through the customer-ready report, under your standard operating procedures (SOPs). The distinction is where the work stops, and who owns the customer relationship when it does.
What is the difference between SOAR and agentic security delivery?
SOAR executes a branch tree you build and maintain in advance, so an incident it has no branch for stops. ContraForce compiles a Gamebook per incident from the entities actually present, so a novel incident still gets an investigation. SOAR automates steps you already scripted. ContraForce performs the work under controls you set.
Should an MSP use MDR or run its own AI SOC?
It depends on whether security delivery is a cost to outsource or a service to own. ContraForce exists for providers who want to own it: you keep the customer relationship, set the procedures, and price the service. MDR is the right answer when a provider does not intend to build a security practice at all.
What are the alternatives to outsourcing SOC delivery to an MDR?
A provider can hire and build a SOC, buy investigation tooling and keep the remaining work manual, or run delivery on an Agentic Security Delivery Platform. ContraForce is the third option: agents perform the eligible work under Gamebook-defined procedures, and analysts review outcomes and approve the actions policy reserves for a human.
How can an MSP scale SOC delivery without hiring more analysts?
ContraForce puts Security Delivery Agents on the repeatable work: triage, investigation, response, tuning, ticket, and report. Analysts review outcomes instead of producing them, and one multitenant control plane covers every customer workspace, so the book of customers can grow without a matching analyst headcount.
What makes an AI SOC platform a fit for a Microsoft-native MSP?
Four things decide it, and ContraForce is built to all four: whether the platform runs on the customer's existing Microsoft Sentinel and Defender XDR without duplicating their logs, whether it enforces your standard operating procedures (SOPs) rather than its own, whether it operates every tenant from one control plane, and whether it closes the loop into your PSA.
How do I add AI to an existing SOC?
In phases, and ContraForce is built to be adopted that way. Start with manual agent runs on individual incidents to evaluate behaviour, then let agents run automatically on the incident statuses and severities you choose, then allow them to execute Gamebooks once a confidence threshold is met.
What does AI triage do in a managed security operations center?
ContraForce Security Delivery Agents take each incoming incident, gather its entities, logs, and detection context, and reach one canonical verdict: True Positive, False Positive, Benign Positive, or Undetermined. That verdict is written back to the source system, so the queue an analyst opens is already sorted rather than raw.
How do MSSPs handle alert fatigue at scale?
Alert fatigue is a volume problem, so ContraForce removes the volume rather than the alerts. Security Delivery Agents work every incident to a verdict instead of sampling, the tuning stage acts on the detections generating the noise, and analysts spend attention on what an agent escalated.
What does an AI SOC analyst actually do?
A ContraForce Security Delivery Agent runs the loop a tier-one analyst runs: triage the incident, investigate it, take or recommend a response, tune the detection behind it, update the ticket, and produce the report. What it does not do is decide unilaterally. Approval gates hold whatever you gate.
What is the best AI security tool for Microsoft Defender XDR?
It depends whether you need analysis or delivery. ContraForce runs the whole loop on Defender XDR incidents: triage, investigation, response, tuning, ticket, and report, writing the verdict back to Defender. Tools that stop at an investigation summary leave the ticket, the report, and the closure with your analysts.
How does AI investigate a Microsoft Defender incident?
A ContraForce Security Delivery Agent takes the Defender XDR incident, gathers its entities, logs, and detection context, and reaches one canonical verdict: True Positive, False Positive, Benign Positive, or Undetermined. That verdict is written back to Defender XDR, and the evidence stays on the incident under Timeline, Comments, and Audit.
What is a Microsoft MXDR delivery platform for service providers?
A platform a provider uses to deliver managed extended detection and response on the customer's own Microsoft stack rather than its own. ContraForce is one: it operates Microsoft Sentinel and Defender XDR inside each customer tenant from a single multitenant control plane, with no duplicated log store and no endpoint agents.
Is there an alternative to building my own security agent on Azure?
Yes. Building your own means provisioning and maintaining Azure AI Foundry infrastructure, then writing the investigation logic, the multitenant isolation, and the audit trail yourself. ContraForce deploys the Foundry infrastructure through Agent Center and runs Security Delivery Agents on top, so the part you keep is your own standard operating procedures (SOPs).
Security is delivered,
not installed.
Connect one workspace and watch the whole chain run, agents acting, humans deciding.