Multi-Agent Based Automation of Zero-Touch Network Provisioning (ZTP)

This agent listens on the DHCP/PXE server for new device boot requests. It authenticates the device via certificate or unique identifier, retrieves its target configuration from the CMDB, and triggers the staging workflow. It eliminates the manual step of logging into a console server to begin provisioning.

Upon trigger, this agent fetches the approved OS image and software bundle from a secure, version-controlled repository. It validates cryptographic hashes against a golden source, stages the image on a local TFTP server, and instructs the device to load it. This automates a high-risk manual process prone to versioning errors.

Using a templating engine (Jinja, Ansible), this agent renders the final device configuration. It incorporates site-specific variables (IPs, VLANs, BGP peers) and pushes the config via NETCONF/SSH. It then validates the running config against intent, rolling back on mismatch. This replaces hours of manual, error-prone CLI work per device.

After config push, this agent executes a series of health checks: ping tests to upstream peers, SNMP polling for critical interfaces, and service verification. It logs results to the ITSM tool (ServiceNow) and, if failures exceed a threshold, escalates to a human operator. This ensures operational readiness before the device is handed over to the NOC.

This agent closes the loop by updating all downstream systems. It registers the new device in the network monitoring platform (SolarWinds, Prometheus), adds it to the IPAM (Infoblox), and creates a corresponding CI in the CMDB. It ensures the asset is immediately visible for lifecycle management, automating a fragmented, post-deployment administrative tail.

The central controller built on a framework like LangGraph or Prefect that sequences the specialist agents, manages state, and handles exceptions. If any agent fails, it routes the workflow to a human-in-the-loop queue in Slack or Teams with full context, ensuring no device is left in a broken state. This provides the necessary governance and audit trail for enterprise-scale automation.

Owns the end-state network design, template logic, and policy definitions. Responsible for defining the golden image, configuration templates (Jinja2, Ansible), and the target state for each device role (access switch, router, firewall). Their success is measured by provisioning accuracy and adherence to security/compliance baselines.

Owns the operational outcome: reduced truck rolls, faster mean-time-to-provision (MTTP), and post-deployment monitoring. They define the validation criteria for agents and integrate the ZTP pipeline with ITSM tools (ServiceNow) for automated ticket creation and closure. Responsible for handling exceptions routed from automated agents.

Owns the build and reliability of the automation platform itself. Builds and maintains the agent orchestration layer (LangGraph, custom Python), manages the DHCP/PXE server, image repository, and API integrations with network controllers (Cisco DNA Center, Arista CloudVision). Ensures the system is scalable, observable, and has rollback capabilities.

Owns the security posture of the automated workflow. Mandates image signing, secure boot validation, and least-privilege access for provisioning agents. Defines the policy for automated compliance checks (CIS benchmarks) post-provisioning and ensures the entire chain of custody is auditable. Must approve any agent with write-access to network devices.

Owns the commercial outcome: accelerated branch rollout, data center expansion velocity, and CapEx/OpEx savings. Funds the build and measures ROI based on reduced deployment cycles, lower remote hands costs, and avoided revenue delay from faster site activation. Decides the rollout sequence and pilot scope.

Owns the physical logistics that enable automation. Coordinates with hardware vendors to ensure devices support ZTP (PXE boot, vendor-specific extensions). Manages serial number ingestion into the CMDB and ensures staged devices are shipped to the correct locations with minimal pre-configuration. Critical for aligning physical and digital workflows.