Skip to content

AI transformation · AIOps

Attackers move at machine speed. Now operations must too.

Exploits arrive within hours of a patch, and routine tickets wait days in queues and hand-offs. Silex puts AI agents into your L1 and L2 queues. They investigate every alert and request, draft the fix, and apply it through Ansible Automation Platform. People approve and audit instead of executing each step.

Why now

The window between patch and exploit has closed.

Turning a patch into an attack once took weeks. In 2026 it takes hours, and a frontier model can write a working exploit from a patch in under an hour. Remediation still runs on a monthly cycle.

  1. 2018 to 201963 daysMandiant average
  2. 2021 to 202232 daysMandiant average
  3. 20235 daysMandiant average
  4. 2026Under 24 hoursIvanti Sentry, after public PoC
  5. 20268 hoursRails, after the patch
  6. 2026Under 1 hourAI-written exploit from a patch
Time from patch or proof of concept to first exploitation, log scale. Sources: Mandiant (2018 to 2023 averages); Dark Reading (Ivanti Sentry, June 2026); Rietta (Rails, July 2026); Anthropic (June 2026).
29%of vulnerabilities newly exploited in 2025 were attacked on or before the day they were disclosed.VulnCheck, 2026
43 daysmedian time to fully remediate a known exploited vulnerability. CISA now gives agencies 3 days for the most severe.Verizon DBIR 2026; CISA BOD 26-04
195 a daynew CVEs published in the first half of 2026, up 50 percent from a year earlier. Headcount does not grow with CVE volume.CVE data review, mid-year 2026

The L1 and L2 queue

Minutes of work, days in the queue.

Most operational tickets need an hour of work or less. They wait to be picked up, then wait again at every hand-off. With the platform, the agent does the work around the decision, and your people make the decision.

Today

An engineer does five of six steps.
  1. Pick up · EngineerWaits days for an engineer
  2. Gather evidence · EngineerSearches logs and tools
  3. Write the change · EngineerDrafts the plan
  4. Approve · Your approversApprove in your ITSM
  5. Execute · EngineerLaunches the AAP job
  6. Verify and close · EngineerChecks and writes it up

With the platform

Your people make one decision.
  1. Pick up · AgentStarts within minutes
  2. Gather evidence · AgentPosts findings to the ticket
  3. Write the change · AgentFiles the plan as an ITSM change
  4. Approve · Your approversApprove in your ITSM
  5. Execute · AAPRuns only the approved steps
  6. Verify and close · AgentVerifies, closes, and attaches evidence

The result

Your engineers approve and audit instead of gathering and typing. The approval step doesn't change: the same people approve in the same tool.

Machine speed

From a monthly cycle to a continuous loop.

The person does not disappear. Their role changes from executing each step to approving and auditing.

Human speed · the monthly cycleMachine speed · the continuous loop

Trigger

Human speedWaits for Patch Tuesday or the quarterly scan.

Machine speed

An advisory or known-exploited listing starts a run in minutes.

Playbook authoring

Human speedEngineers write and test scripts by hand, over days.

Machine speed

The agent drafts, lints, and canary-tests the playbook in the same run.

Baseline

Human speedRarely taken. Drift is found by the outage.

Machine speed

Every host is recorded before each change: packages, services, trust stores, SELinux settings.

Approval

Human speedA change board that meets weekly.

Machine speed

Pre-approved paths. A person approves each action in the ticket, with a full audit trail.

Verification

Human speedRescan next cycle. A failed fix waits a month.

Machine speed

Rescan in the same run. A failed fix is rolled back or reopened at once.

Exposure window

Human speedWeeks, from advisory to fix.

Machine speed

Hours, matching the attacker's clock.

One ticket, start to finish

A patch cycle in 14 minutes, for $7 of model time.

A ticket asks for an application tier to be patched. The agent learns the environment, files the change in ServiceNow, and waits. Once a person approves, Ansible Automation Platform patches the idle node first, the agent checks it, finishes the tier, and resolves the ticket.

Take-up to resolved14:3217:53 to 18:07 UTC
Patched2 hosts16 AAP jobs launched
Approvals given1 of 1in ServiceNow, before anything ran
Model cost$7.12against a $50.00 cap
FinishedTicket INC0010586Change CHG0030365
  1. Request filed
  2. Environment learned
  3. Fix reviewed
  4. Change filed
  5. Approval
  6. Idle node patched
  7. Checks
  8. Patch finished
  9. Ticket resolved

What happened

all times UTC
  1. 17:28Run started: patch-cycle workflow, act with approval
  2. 17:42Findings posted to INC0010586 as a work note
  3. 17:55CHG0030365 filed
  4. 17:56CHG0030365 approvedBy a person
  5. 18:06CHG0030365 closed successful
  6. 18:07INC0010586 resolved by the run
  7. 18:09Run finished

Records

ServiceNow
INC0010586
Resolved
CHG003036516 steps run
Closed successful

Every step, read, and job is in the run's record.

DoneThe application tier is patched and the ticket is resolved.

The run summary as the Silex AIOps Platform recorded it: a two-host application tier patched under one ServiceNow change, through Ansible Automation Platform.

Where agents help first

Start where the queue is deepest.

Each workflow covers the alert and request types agreed for it, and anything else goes back to the queue. Each card shows the autonomy a workflow can reach; every one starts read-only.

Assist

Infrastructure triage

Alerts from your monitoring tools, correlated with logs and the CMDB and diagnosed read-only. The lowest-risk place to start.

Up to self-heal

Disk and operational fixes

Disk cleanup and expansion, service restarts, certificates, and reboots, planned, approved, and verified.

Up to self-heal

Server patching

A baseline before each change, staged execution with a health check, verification, and a diagnosis of anything that fails.

Act with approval

Server builds

Builds through your pipeline and AAP, drafted as pull requests, with a separate approval for anything that spends money.

Act with approval

Database requests

Databases, schemas, grants, and space additions from your service catalog.

Assist

Network change compliance

Device changes checked against the approved change, its window, and your CMDB.

Act with approval

Server decommission

Dependency checks, a staged power-off, and a hold before anything destructive.

Act with approval

Security remediation

Per-host fixes and CVE campaigns from your vulnerability scanners, with the evidence drafted alongside.

How it works

One loop, with a person at the approval step.

The same loop drives triage, remediation, patching, and provisioning. Every read and every change passes the platform's gateways, and every change is held to the plan a person approved.

1Detect2Investigate3Draft4Approve5Execute6VerifyOne loopminutes, not weeks
  1. 1

    Detect

    A ticket, alert, or advisory lands in your ITSM queue, and the platform starts a run for it.

  2. 2

    Investigate

    The agent reads monitoring, logs, and the CMDB through the enforcement gateway, from a sandbox with no real credentials.

  3. 3

    Draft

    It posts its findings to the ticket and proposes a plan. A playbook it writes goes to review as a pull request.

  4. 4

    Approve

    A person approves the change in your ITSM. Nothing runs while it waits.

  5. 5

    Execute

    Ansible Automation Platform runs exactly the approved steps, in the window.

  6. 6

    Verify

    The agent checks the result, attaches the evidence, and closes the ticket.

Works with ServiceNow · Red Hat Ansible Automation Platform · OpenShift · GitHub · OpenTofu · Grafana · Prometheus · Loki · Tenable · Qualys · Rapid7 · Splunk · Datadog · Dynatrace

Zero agent trust

We trust the controls, never the agent.

Tickets, alerts, and tool output can carry an attacker's words, so the Silex AIOps Platform never relies on what the agent decides. The agent holds no real credentials, and six layers stand between it and production.

How the platform works
  1. Layer 1Your signed workflow definition
  2. Layer 2A sandbox with no real credentials
  3. Layer 3The enforcement gateway
  4. Layer 4Launch policy as code
  5. Layer 5The change gate and a person's approval
  6. Layer 6Ansible Automation Platform and your hosts

Autonomy ladder

Human-in-the-loop by default.

A workflow moves from Assist to Act with approval, and then to Self-heal, only when its diagnoses match how your engineers resolved the same tickets and it passes its failure tests.

  • Agents act only through approved Ansible job templates.
  • Agents never log in to hosts and never hold secrets.
  • Every change needs an approved change record until the workflow is promoted.
  • A console records every run, and each workflow has a stop control.
  • Every model call passes the platform's model gateway, which holds the key, enforces budgets, and records the call.
  1. Level 1

    Assist

    Agents investigate with read-only access and draft a diagnosis and a fix.

  2. Level 2

    Act with approval

    A person approves each change in one step; Ansible executes and verifies it.

  3. Level 3

    Self-heal

    Fixes you pre-approved as standard changes are applied, verified, and closed without waiting for a person. People audit.

Self-healing infrastructure

Routine fixes that fix themselves, under the same controls.

At Self-heal, a workflow applies, verifies, and closes the fixes you pre-approved as standard changes, like disk cleanup, service restarts, certificate renewals, and approved patch campaigns, without waiting for a person. Your engineers audit instead of approving each one.

Standard changes, on your terms

You choose which fixes count as standard changes. Each run still files and closes its change in your ITSM, so the record matches one a person approved.

The same gate and the same AAP

The gate holds every launch to the standard change's job templates, hosts, and window, and Ansible Automation Platform runs it. Nothing outside the template runs.

Verified, or backed out

Every fix is checked before the ticket closes. A failed check backs out what it broke, and can open a problem, and an incident if service was left reduced.

Earned, and easy to stop

A workflow reaches Self-heal only with a clean record at Act with approval, after passing our failure scenarios, and only when you sign the change. The off switch stops it at once.

Every self-healed change is in the audit and the weekly activity report, and the monthly service review covers how each self-healing workflow performed.

Evaluation

Evaluate before you trust.

An agent that closes tickets has to be trusted with production. Silex tests models and agent harnesses in a lab built like an enterprise environment, with Ansible Automation Platform, ServiceNow change control, a runbook repository, and a blue-green three-tier application with UAT gates.

We test models from Anthropic, OpenAI, Google, and open-weight families, pick the model, reasoning level, and harness for each workflow, and re-test whenever a model changes. In production, every ticket a workflow works is compared with how your engineer resolved it.

Why we built SRE-bench

What every run is graded on

  • Did the agent find the real cause?
  • Did it stay inside approved job templates and the maintenance window?
  • Did it cite evidence for each conclusion?
  • Did it reach the right answer without causing an outage on the way?

Runs are graded from platform records, not from what the agent reports about itself.

Foundation

Automation is how AI acts safely.

Silex is a Red Hat partner, and Red Hat Ansible Automation Platform is the platform's trusted execution layer. Agents never log in to hosts: every change runs as an approved job template under AAP's role-based access and policy as code. Silex has run enterprise Ansible programs since 2017, so the automation library the agents use is one our engineers build and maintain.

Red Hat Ansible Automation Platform · Event-Driven Ansible · AAP MCP server · OpenShift · OpenTofu

Automation and Platform Engineering

How we engage

From assessment to self-healing.

  1. 1

    Assess

    A Patch Window Assessment or discovery workshop measures your advisory-to-fix time and maps the queue.

  2. 2

    Pilot

    A triage agent works a test queue, then a share of live tickets, with read-only access.

  3. 3

    Act with approval

    Remediation, patching, and provisioning execute through job templates on one-step approval.

  4. 4

    Self-heal

    Workflows with a clean record run without per-change approval.

  5. 5

    Operate

    Managed operations: runs reviewed, models re-tested, and a monthly service review.

Day 2 with Silex

We run the agents with you.

Go-live is one day. Managed operations keep every workflow tested, measured, and current for every week after it, with a named Silex coordinator and the engineers who run your workflows.

  1. 1

    Operate

    Review flagged runs and refusals, investigate failures, and pause or stop a workflow.

  2. 2

    Test

    Prove every new model or agent version on the same tickets before it ships.

  3. 3

    Measure

    Track each workflow against its autonomy criteria and your ticket baseline.

  4. 4

    Report

    Weekly activity, a monthly service review, and a quarterly governance review.

  5. 5

    Improve

    Keep integrations current as your platforms change, and add the work you rank next.

The management console's runs for the day, including a run refused because its workflow was switched off and a run waiting for approval in the change system.
Every run of the day in one list, including a run refused because its workflow was switched off and one waiting on its approver.

Questions buyers ask

Questions

Will agents change production systems without approval?

No. Every workflow starts at Assist with read-only access. Changes need one-step human approval until a workflow has a clean record on your own tickets and passes its failure tests. You can stop any workflow at any time.

What does self-healing mean here?

A workflow at Self-heal applies the fixes you pre-approved as standard changes, such as disk cleanup, service restarts, and certificate renewals, without waiting for a person. Each change is still filed in your ITSM, held to its job templates and window by the gate, run by Ansible Automation Platform, and verified or backed out. People audit the results instead of approving each one.

Do we need Ansible Automation Platform?

Agents act only through approved automation, so an automation platform is required. Silex builds on Red Hat Ansible Automation Platform and can stand it up or extend an existing installation.

Which models do you use?

We test models from Anthropic, OpenAI, Google, and open-weight families in our lab, then beside the current model on your own tickets, and pick the model for each workflow. Every model call passes the platform's model gateway on the route you choose, so it is budgeted and recorded.

Where does the platform run?

In Silex's cloud, in your environment, or split between the two: full SaaS, hybrid with the data plane in your network, or self-hosted on your own OpenShift or Kubernetes. The controls and the audit are the same in every option.

How do you keep agents from being misused?

The agent is treated as untrusted. It holds no real credentials, reaches your systems only through the platform's enforcement gateway, and changes a system only through a plan a person approved in your ITSM. The platform page explains the six layers of control.

Which tools do the agents work with?

Your ITSM and CMDB (for example ServiceNow), your monitoring and observability tools, and your vulnerability scanners. Vendor agents such as Tanium can be called through their MCP servers.

How do we measure whether it works?

The assessment records your current advisory-to-fix time and queue wait times. Each phase reports time to resolve and the share of tickets the agents handle against that baseline.

Next step

Measure your advisory-to-fix time.

The Patch Window Assessment measures how long a fix takes to reach production in your environment today and identifies the workflows where agents will shorten it first.