Intent-Based Networking Explained in Practice

Intent-based networking explained for engineers: how declared outcomes become enforced policy, how drift is caught, and how to adopt it safely.

Intent-Based Networking Explained in Practice
Intent-Based Networking Explained in Practice

Last year one of our engineers spent most of a night tracking down a fault that should never have existed. After a late change window, a finance VLAN had quietly gained a route toward the guest Wi-Fi. Nobody typed a wrong command. The scripts ran exactly as written, and that was the real problem. Intent-based networking explained properly comes down to closing the gap between what you meant and what the devices actually do.

Intent-Based Networking Explained: The Core Idea

Intent-Based Networking Explained: The Core Idea
Intent-Based Networking Explained: The Core Idea

Put simply, you declare the outcome you need, and the system works out the device settings required to produce it. You describe what must be true, not the sequence of commands needed to get there. That difference changes who owns the problem when the network starts to drift.

Many people assume intent-based networking is just scripting with better marketing. Scripts do what they are told, once, and then they stop caring. An intent system keeps asking whether the outcome still holds. Faster scripts only repeat the same mistake faster.

The payoff is biggest on multi-site, multi-vendor estates. Every platform has its own quirks, and no single engineer carries the full picture anymore. We have seen teams with twenty years of institutional knowledge still get caught out because a firmware note changed one default.

From a Sentence to an Enforced Policy

From a Sentence to an Enforced Policy
From a Sentence to an Enforced Policy

An intent begins as a declarative statement such as “finance VLAN cannot reach guest Wi-Fi.” It describes a fact about the network, not a list of ACL lines. That keeps the statement readable even after the hardware underneath is replaced.

The translation layer comes next. It maps the statement onto device-level policy for each platform involved. The same intent might become an access list on one switch and a firewall zone rule on another. Most of the real engineering happens here.

Vague intents break down at this stage, because a sentence that cannot become testable policy gives the translator nothing to work with.

Activation is the final step. A solid implementation validates syntax, dependencies, and conflicts before any change touches production traffic.

Keeping the Loop Closed

Keeping the Loop Closed
Keeping the Loop Closed

Enforcing an intent once is not enough. Devices reboot, someone patches a ticket by hand at 2 a.m., and a firmware upgrade quietly changes a default. So the system keeps watching. Where the platform supports it, it collects streaming telemetry, with SNMP and syslog as older fallbacks.

It then compares observed state with intended state. When they disagree, you have drift. Depending on configuration, the system either remediates automatically or raises an alert for review. We recommend starting with alerting only. Automatic correction is only as reliable as the intent behind it, and a wrong intent corrected automatically can take a segment down within minutes.

Where Plain Automation Stops

Where Plain Automation Stops
Where Plain Automation Stops

Scripts execute steps, while intent defines the end state. A script that adds a VLAN to twelve switches finishes and exits. Whether that VLAN is still correct next Tuesday is not its problem. Automation without assurance can drift silently for months, and nobody notices until something fails.

Scripts surface drift only once users or an incident reveal it. An intent layer flags or fixes it earlier. The skill profile shifts too, moving from CLI knowledge toward policy modelling alongside scripting.

Neither replaces the other. Many teams keep scripts as the execution engine underneath an intent layer, a sensible arrangement for most mid-sized estates.

One Intent, Written and Verified

One Intent, Written and Verified
One Intent, Written and Verified

Consider a realistic example. The intent is simple: finance hosts must not reach guest Wi-Fi. A YAML-style expression might look like this:

intent: finance-isolation
source: vlan-finance
destination: vlan-guest
action: deny
verify_every: 5m

A translator could produce a generic CLI fragment like the one below on an access switch. Syntax differs between vendors, so treat it as illustrative only:

ip access-list extended FIN-NO-GUEST
 deny ip 10.20.0.0 0.0.255.255 10.30.0.0 0.0.255.255
 permit ip any any
interface vlan 20
 ip access-group FIN-NO-GUEST in

Verification is the part people skip. Run a test from a finance host, confirm the traffic is refused, then check the access list counters. A command such as show ip access-lists FIN-NO-GUEST should show matches on the deny line after your test. If the counter stays at zero, either the traffic never reaches that switch or the policy is not applied where you assumed. In one of our lab runs, that exact mismatch turned out to be a trunk that had been pruned months earlier.

Adopting It Without Breaking Production

Adopting It Without Breaking Production
Adopting It Without Breaking Production

Start by enforcing nothing. Build a read-only assurance baseline and let the system report how the network currently differs from what you believe it should be. You will almost certainly find surprises, and dealing with them is a legitimate first project in its own right.

Then pilot a single intent on a non-critical segment, such as a lab or a low-impact branch, and watch it for several weeks. Expand only once the drift metrics settle, so that the alerts firing are ones your team actually wants to act on. Alert fatigue kills more of these programmes than any technical limitation does.

Where Intent Fits in Your Architecture

Where Intent Fits in Your Architecture
Where Intent Fits in Your Architecture

Intent-based networking sits on top of your SDN controllers and fabrics rather than replacing them. For the longer view on where this is heading, our article on AI-native networking and autonomous operations covers that direction in more depth. For background on the underlying overlay design, our notes on VXLAN EVPN and when it is worth using make a useful companion read. Teams planning broader modernisation can also browse our automation articles for related practical guides.

Where a wider integration project also covers payment or billing flows, a cổng thanh toán miễn phí may be evaluated separately, outside the network design itself.

Feature availability depends heavily on hardware and software versions, and some capabilities only appear in licensed tiers. Check the official vendor documentation for your exact models before committing to a plan. Your team will also need policy-modelling skills, version control for intent definitions, and enough scripting to handle translation edge cases.

Pitfalls to Avoid

The most common one is the vague intent. “Keep the network secure” cannot be translated into testable policy. Good intents are narrow and checkable, and if you cannot write a test that passes or fails, the intent is not ready yet.

Vendor lock-in is another trap. Some platforms use proprietary intent models that are painful to leave. Ask how intents are stored and whether they can be exported in an open format before committing.

Version dependencies also quietly limit what you can do. A feature shown in a vendor demo may need a firmware level your switches cannot run. Verify everything against your own equipment, not the demo rack.

Costs, Skills and Rollout in Vietnam

Most enterprise networks in Vietnam are multi-site and built from several vendors, accumulated across years of projects. That makes the translation problem harder, because there are more platforms to map against.

The skills gap is real. Many local teams are strong on daily operations but have less exposure to policy modelling and telemetry pipelines. The fastest route is usually a pilot alongside an experienced partner on a live but contained segment, so staff learn by doing. Budget should be confirmed with your vendor and integrator, and typically covers hardware refresh for switches that cannot stream telemetry, licensing for the assurance module, staff time for the baseline phase, and training in policy modelling and scripting.

Frequently Asked Questions

Is intent-based networking the same as SDN?

No. SDN separates the control plane from the forwarding plane, which is an architectural decision. Intent-based networking concerns how you express and enforce desired outcomes. You can have either without the other, although in practice they often arrive together.

Do I need to replace all my switches to use it?

Not necessarily. Many platforms support intent workflows through existing management APIs. The real constraint is telemetry. Older hardware may not stream the data that good assurance depends on, so check your models against vendor compatibility lists first.

Does it remove the need for network engineers?

It changes the work rather than removing it. Engineers spend less time typing repetitive configuration and more time defining intents, reviewing drift alerts, and deciding what the network should be doing. Teams that treat it purely as a headcount cut usually end up running a network nobody fully understands.

How do I measure whether intent is actually being met?

Track the share of intents currently compliant, how long drift takes to detect, and how often drift is corrected correctly without a human stepping in. Pair those metrics with active tests, like the access list counter check shown earlier, so you are not relying only on the system reporting on itself.

Related topics