Configuration-management tools
These tools describe the desired state of your devices in files you keep under version control, then make the devices match. The key idea is idempotency: running the same playbook again changes nothing if the device is already correct.
| Tool | Agent on the device? | Push or pull | Language | Note |
|---|---|---|---|---|
| Ansible | No (agentless, over SSH) | Push | YAML playbooks | Most common for network devices |
| Terraform | No | Push | HCL (declarative) | Infrastructure as code, great for cloud |
| Puppet | Yes (agent) | Pull | Puppet DSL | Agent checks in with a server |
| Chef | Yes (agent) | Pull | Ruby | "Recipes" and "cookbooks" |
An Ansible playbook to make a VLAN exist looks like this (YAML). Run it on a fresh switch and it creates VLAN 10; run it again and it does nothing, because the VLAN is already there - that's idempotency:
- name: Configure access VLAN
hosts: switches
tasks:
- name: Create VLAN 10
cisco.ios.ios_vlans:
config:
- name: SALES
vlan_id: 10
state: mergedAI in networking
AI is moving from a buzzword to a daily tool for network teams:
- AIOps / assurance: machine learning spots a problem (or predicts one) from mountains of telemetry a person could never read, and points at the likely cause.
- Generative AI: a large language model writes a config, explains an error, or drafts a script from a plain-English request - a tireless assistant, but one you must check.
- Agentic AI (new in the exam): an AI that doesn't just answer but takes steps on its own - gathering data, making a change, checking the result - within limits you set.
Prompting, briefly
You get better answers from a generative AI by being specific: say the role ("you are a network engineer"), give context (the device, the IOS version, the goal), and ask for a clear output ("give the exact IOS commands"). Always review what it produces before you paste it into a live device: AI can be confidently wrong.
Remember: Ansible: agentless, push, YAML, most common for network kit. Idempotent = re-running changes nothing already correct. Generative AI drafts; you review. Agentic AI also acts.