Network Engineering Automation Tools: Building a CI/CD Pipeline for Network Changes
Zak
8th August 2026
Most network teams already have some automation tooling, a handful of Python scripts, maybe some Ansible playbooks. What most teams don't have is those tools connected into an actual pipeline, which is the difference between automation that saves time occasionally and automation that's trustworthy enough to run against production without an engineer watching over it.
Code editor screen showing HTML, CSS, and JavaScript source code
The Core Toolchain
Device-level scripting: Netmiko and NAPALM. These libraries provide a consistent Python interface across different network vendors, so engineers write automation logic once rather than relearning each vendor's specific CLI and API quirks.
Configuration management: Ansible or Nornir. Ansible's declarative, agentless model is the standard choice for most teams, playbooks describe the desired state, and Ansible handles getting devices there. Nornir is a Python-native alternative favoured by teams who want full programmatic control rather than YAML-based playbooks.
Source of truth: NetBox. NetBox provides RESTful API access to device inventory, IP address management, and topology data, and integrates directly with Ansible for dynamic inventory generation, meaning automation always operates against current, accurate data rather than a spreadsheet someone forgot to update.
Templating: Jinja2. Configuration templates let engineers define a device configuration pattern once and generate consistent, correct configuration across an entire fleet, rather than hand-editing each device's configuration individually.
Building the CI/CD Pipeline
Having the tools above doesn't automatically give you a pipeline, it gives you a toolbox. A genuine CI/CD pipeline for network changes runs through five connected stages:
1. Lint
Static analysis of the proposed configuration change, tools like ansible-lint for playbooks or YANG-model validation for structured configuration, catching syntax errors and standards violations before anything executes against real infrastructure.
2. Validate
Confirm the proposed change conforms to organisational policy: does it violate a compliance requirement, does it conflict with an existing rule, does it match the expected schema.
3. Test
Deploy the validated change to a sandboxed environment (tools like GNS3 or Cisco VIRL) and run functional tests confirming connectivity, routing behaviour, and security policy work as intended, before the change goes anywhere near production.
Blue equipment cabinets densely packed with wired electronic components
4. Deploy
Apply the validated, tested configuration to production via the automation tooling itself, Ansible, Terraform, or equivalent, never via manual CLI entry, so the exact change that was tested is the exact change that gets applied.
5. Verify
Automated post-deployment health checks, confirming routing tables, security policies, and connectivity match what was intended, close the loop, rather than assuming a deployment succeeded just because it didn't throw an error.
Where Most Teams Get Stuck
Skipping the source of truth. Automation built directly against manually-maintained documentation or institutional memory inherits every inconsistency in that documentation. Get NetBox (or equivalent) genuinely accurate before building extensively on top of it.
Treating "deploy" as the only stage that matters. Teams that jump straight to automated deployment without linting, validation, and testing stages are automating the risk, not removing it, a bad change now propagates at automation speed instead of manual speed.
No rollback path. Every stage needs a clear rollback mechanism defined up front, if the only way to undo an automated change is to manually reconstruct the previous state, the pipeline hasn't actually reduced risk, it's just moved where the risk sits.
Starting Small
A full five-stage pipeline is the destination, not the starting point. A reasonable build-out sequence:
- Start with linting on proposed changes, low effort, immediately useful, no risk to production.
- Add a source-of-truth platform and get device inventory genuinely accurate.
- Add sandboxed testing for a narrow, well-understood class of changes.
- Automate deployment for that same narrow class of changes, with manual approval as a gate.
- Expand scope and remove the manual gate only once the pipeline has demonstrated it catches problems reliably.
Where Conxiea Fits
Conxiea's AI-driven InfraOps platform provides the monitoring and verification layer this pipeline depends on, continuously confirming deployed configuration matches intended state, and correlating any drift back to its cause, so the "verify" stage doesn't stop the moment a deployment completes.
Book a free demo to see how it fits your automation stack →