Get started with the Deep Network Model AI Assistant in Cisco U.

At Cisco Live in San Diego, D.J. Sampath, Senior Vice President of Cisco’s AI Software and Platform group, wowed the crowd with a demo of AI Canvas. That’s a multi-data, multi-agent system, integrated with Cisco’s AI Assistant and powered by Cisco’s Deep Network Model. In that demo, we could all see AI Canvas’s ability to speed troubleshooting, bring siloed teams together, and enable automation across the entire stack.

AI Canvas won’t be available until October. However, we wanted to offer our CCIEs, CCDEs, and Cisco Certified DevNet Experts the opportunity to work with the Deep Network Model as soon as possible. So we’re making the model available to CCIEs and other experts through an AI Learning Assistant available in Cisco U.

We think CCIEs (and soon, other network engineers) will find a wealth of ways that the Deep Network Model can help them learn more and become more efficient. But we realize that agentic ops is brand new, and that you might be wondering how you can immediately start experimenting with the Deep Network Model. So I thought I’d offer some sample use cases to help you get started.

Tailored scenarios and training paths

As a CCIE, you’ve got years—sometimes decades—of experience in networking, and you’re fully up to speed on your organization’s IT infrastructure. But what about your team members, especially more junior network engineers? The Deep Network Model AI Assistant can be used to build tailored scenarios and training ideas so that everyone on your team can learn the skills needed for the network you currently have, as well as any new technologies your organization plans to roll out.

The Deep Network Model understands a wide range of networking technologies, but it’s trained explicitly on a depth and breadth of Cisco-specific material. It’s also trained on the materials and coursework available in Cisco U. You might try a prompt such as this one:

I am the tech lead for a small team of network engineers. I need to quickly get them up to speed on the networking technology we use in our environment, including BGP, MPLS, and OSPF. Could you build me a custom study plan?

When I asked this question of the Deep Network Model AI Assistant, I got a very nice syllabus in outline form, with links to courses in Cisco U.

Here’s a sample:

Design validation and optimization

Cisco Validated Designs (CVDs) are essentially blueprints, and IT professionals are accustomed to working through them. But sometimes you need more guidance. The Deep Network Model AI Assistant can help make CVDs more navigable. It can access other sources to help flesh out CVDs and offer suggestions for improving or optimizing designs.

It can also summarize the CVD, giving you a high-level overview before reading the whole thing. You can ask it questions such as:

Considering the CVD for FlexPod, provide a getting-started document that I can use to configure my initial UCS manager.
I am beginning to implement the CVD for FlexPod. Could you give me a high-level overview of what I’ll be doing and the pieces I’ll be working with?

The Deep Network Model AI Assistant can help validate an existing design with respect to a CVD and offer suggestions for improving or optimizing designs.

What kind of storage technology should I consider for booting my blades in a UCS B chassis?

If you’re having issues with a CVD, you can ask the Deep Network Model AI Assistant where you should start looking.

Automation assistant

The Deep Network Model AI Assistant can also help with automation. You could ask it questions such as:

I am an expert in network architecture and need some help automating our branch SD-WAN deployment. What would be a well-supported, easy-to-learn tool that would help me support this? My team doesn’t have a great deal of coding experience. Could you provide examples and links to relevant documentation and training?

Troubleshooting

The Deep Network Model AI Assistant can help analyze network diagnostics, such as syslog messages and debug output, and examine problem symptoms to provide insight that might be missed by human eyes. Although generative AI is still a young technology that can make mistakes, expert-level IT professionals are well-equipped to evaluate the output for accuracy and detect hallucinations.

For example, the Deep Network Model AI Assistant could help interpret a syslog message. You could simply input the message into the assistant and say you need advice or a place to start. Because it’s trained on Cisco’s syslog formats, it can give guidance and cross-reference other data.

If you’re working with multiple data sources, the analysis becomes more complex. With the Deep Network Model AI Assistant, you can describe the symptom you see and then ask, “What should I look for?” (Of course, you should always be careful about pasting raw output into AI.) In this way, you can use the assistant to guide you to the point where you’re comfortable taking over.

A lot of debugging is essentially looking at different types of diagnostic data and searching for the needle in a haystack that will help you know what to do next. The Deep Network Model AI Assistant can help with that process. For example, if you need to troubleshoot routing adjacencies, you’ll likely need to gather data from multiple devices and correlate the data to identify a root cause.

You are a network troubleshooting assistant. Help me diagnose why my OSPFv3 session is not establishing with one neighbor. This is the output from ‘show ospfv3 neighbor’:

          OSPFv3 1 address-family ipv6 (router-id 192.0.2.1)

Neighbor ID     Pri   State           Dead Time   Interface ID    Interface

192.0.2.2    128   EXCHANGE/BDR    00:00:38    13              Vlan300

192.0.2.6    128   FULL/DR         00:00:37    5               Vlan300

And this is the relevant config from Vlan300: 

ipv6 address FE80::300:241 link-local

ipv6 address 2001:DB8::241/64

ipv6 enable

ipv6 mtu 1500

ipv6 nd dad attempts 0

ipv6 nd ra suppress all

no ipv6 redirects

ipv6 ospf 1 area 0

bfd interval 1000 min_rx 1000 multiplier 5

Here’s the response I got:

At some point, many of us end up troubleshooting at the protocol level (packet capture or it didn’t happen, right?), where things get complex very quickly. In this case, you can paste the decoded output of a packet capture (such as that from Wireshark or Tshark) to the Deep Network Model AI Assistant, which can break down the frame details for you. It can identify hard-to-spot issues and dramatically increase the efficacy of deep networking troubleshooting.

The AI assistant can give you more meaning and context than you might get with other tools. I tried this with a problematic SNMPv3 packet. The AI assistant looked at the value of the fields and explained them to me. While Wireshark showed me the field names, the AI assistant explained that one field, the msgAuthoritativeEngineTime, represented the number of seconds a device had been online, which was 61411 (roughly seven weeks). The thing is, I just booted that device. So my SNMP manager was confused, and the SNMPv3 trap wasn’t being trusted. Bug found!

While most of us are pretty familiar with a wide range of network technologies, we may not be experts in every one of the protocols we run on our network. Therefore, consider how useful this can be for a protocol you’re not highly knowledgeable about at the field level. The AI assistant is excellent at analyzing those fields and explaining their network-relevant context. While the assistant won’t solve the problem for you, when used properly, it can give you some nice hints. Once you understand more about those fields, applying some reasoning and fixing the bug is much easier.

These are just some of the ways that the Deep Network Model AI Assistant could be helpful to experienced network engineers. I hope they’re a useful springboard for your thinking. If you try them out, I’d be excited to hear about the results you’re getting.

But I’d be even more excited to hear about use cases you’ve come up with that I might never think of. AI is an incredibly powerful tool that can make us more efficient and, frankly, less stressed. But we must figure out the best ways to use them, and we’re all on that journey together.