An Artificially Clever Peer Reviewer Arrives

Nicely whats up everybody, it’s been a minute (or many) since my final weblog publish, however I’m excited to say that the key venture I’ve been engaged on for the final 18+ months has come to a conclusion (properly, for me no less than 😉 ) and I’m wanting ahead to investing extra time in blogs, movies, and getting again on the market.

I’ve a number of matters I’m trying to dive into, however getting back from Cisco Reside in Vegas final week, I figured I would as properly leap on the AI hype prepare, chugga chugga chugga, and share a little bit of my very own ideas on the subject.

Let me begin by saying that I’m simply starting to actually dip my toe into the AI pool. There are a lot of different engineers already manner out within the deep finish, and I hope to swim out and be part of them quickly.

Cisco Reside 2024 in Vegas was simply the compelling occasion I wanted to step down on step one (to maintain up the metaphor), and I included AI in two of my talks throughout the week. There was be a Community Engineer in an Synthetic Age within the Cisco U. Theater, an replace to my discuss, “ be a Community Engineer in a Programmable Age,” that I first delivered at Cisco Reside in 2017. After eight years, it felt just a little bittersweet to depart the programmable age behind, however thrilling to be getting into a brand new age. Synthetic Intelligence additionally made a last-minute “bonus” look in my breakout “Programmability Expertise for Engineers Who Don’t Wish to Code.”

For these talks, I spent a while pondering by means of how community engineers must be fascinated with and utilizing AI proper now, and what to probably count on for the long run. I’m not going to undergo all the main points on this weblog publish, however relatively, I wish to focus in on one particular use case that I discovered shocking and a bit thrilling: Utilizing AI to see assessment your work. Buckle in, and let’s test it out.

Bridget takes Hank to Community Engineering College

One of many examples in “Programmability Expertise for Engineers Who Don’t Wish to Code” revolves round creating IPSEC VPN configurations for routers based mostly on a configuration template and spreadsheet of information.  I first present the right way to use formulation inside the spreadsheet to create the configurations, and later use Ansible and Jinja2 templates to do the identical factor.

As I used to be rehearsing the session, I puzzled… Might AI do the identical work?

If I’m trustworthy, I figured it could get shut, however have to be mounted/completed to be carried out.  So I crafted this immediate and despatched it to Cisco’s ChatGPT implementation often called “BridgeIT”.

Utilizing the information from the beneath CSV information,

vpn_name,tunnel_interface,side_1_public_ip,source_network,source_mask,side_2_outside,tunnel_destination,destination_network,destination_mask,preshared_key
VPN_101,GigabitEthernet0/1,203.0.113.101,10.98.128.0,0.0.3.255,GigabitEthernet1/1,198.51.100.33,172.28.11.0,0.0.0.255,ZA788I
VPN_102,GigabitEthernet0/1,203.0.113.102,10.98.128.0,0.0.3.255,GigabitEthernet1/1,198.51.100.34,172.28.12.0,0.0.0.255,AM285I

generate a configuration from every row based mostly on the beneath jinja2 template

! VPN {{ vpn_name }}
crypto isakmp coverage 10
 encryption aes
 hash sha256
 authentication pre-share
 group 14 

crypto ipsec transform-set vpn_transform esp-aes esp-sha256-hmac

access-list 100 allow ip {{ source_network }} {{ source_mask }} {{ destination_network }} {{ destination_mask }}

crypto isakmp key {{ preshared_key }} tackle {{ tunnel_destination }}

crypto map {{ vpn_name }} 10 ipsec-isakmp
 set peer {{ tunnel_destination }} 
 set transform-setvpn_transform 
 match tackle 100

interface {{ tunnel_interface }} 
 crypto map {{ vpn_name }}

I despatched it away and waited… and waited… and waited.  However then this reply got here again:

Primarily based on the supplied Jinja2 template and the CSV information, listed below are the generated configurations for every VPN:

For VPN_101:

! VPN VPN_101
crypto isakmp coverage 10
    encryption aes
    hash sha256
    authentication pre-share
    group 14

crypto ipsec transform-set vpn_transform esp-aes esp-sha256-hmac

access-list 100 allow ip 10.98.128.0 0.0.3.255 172.28.11.0 0.0.0.255

crypto isakmp key ZA788I tackle 198.51.100.33

crypto map VPN_101 10 ipsec-isakmp
    set peer 198.51.100.33
    set transform-set vpn_transform
    match tackle 100

interface GigabitEthernet0/1
    crypto map VPN_101

For VPN_102:

! VPN VPN_102
crypto isakmp coverage 10
    encryption aes
    hash sha256
    authentication pre-share
    group 14

crypto ipsec transform-set vpn_transform esp-aes esp-sha256-hmac

access-list 100 allow ip 10.98.128.0 0.0.3.255 172.28.12.0 0.0.0.255

crypto isakmp key AM285I tackle 198.51.100.34

crypto map VPN_102 10 ipsec-isakmp
    set peer 198.51.100.34
    set transform-set vpn_transform
    match tackle 100

interface GigabitEthernet0/1
    crypto map VPN_102

I used to be duly impressed. Bridget (as BridgeIT is affectionately often called) understood precisely what I wished and was in a position to create the configurations for me completely. Rating one for our artificially clever robotic overlords.

However wait… she wasn’t carried out.  The message continued:

Please observe that there’s a typo within the Jinja2 template, particularly within the line the place it says:

set transform-setvpn_transform

There must be an area earlier than “vpn_transform”. Right here’s the corrected line that ought to seem within the template:

set transform-set vpn_transform

Oh man… I had a typo in my template? Inconceivable!!! I shortly went and checked my immediate and the supply Jinja2 template that I copied for the immediate—a template I’ve utilized in my Ansible demo on this presentation for 2 years.

Positive sufficient, I had a typo. Whoops… boy, did I really feel embarrassed. And when you return and take a look at the configurations that Bridget supplied to my immediate, you’ll see that she truly mounted the error and despatched the right configs. Very, very properly carried out. Rating two and three.

However she nonetheless wasn’t carried out…

Moreover, you would possibly face a problem when making use of two totally different crypto maps to the identical interface. Often, a bodily interface solely helps one crypto map. If you must assist a number of VPNs on the identical interface, you’d sometimes use the identical crypto map title with totally different sequence numbers for every peer.

Right here’s how one can modify the second crypto map to make use of the identical crypto map title with a special sequence quantity (assuming you wish to use the crypto map title VPN_MAP for each):

For VPN_102 (with a revised crypto map title and sequence quantity):

! VPN VPN_102
crypto isakmp coverage 10
    encryption aes
    hash sha256
    authentication pre-share
    group 14

crypto ipsec transform-set vpn_transform esp-aes esp-sha256-hmac

access-list 102 allow ip 10.98.128.0 0.0.3.255 172.28.12.0 0.0.0.255

crypto isakmp key AM285I tackle 198.51.100.34

crypto map VPN_MAP 20 ipsec-isakmp
    set peer 198.51.100.34
    set transform-set vpn_transform
    match tackle 102

interface GigabitEthernet0/1
    crypto map VPN_MAP

Discover that the entry checklist quantity for VPN_102 has been modified to 102 to keep away from battle with the entry checklist for VPN_101. The crypto map sequence quantity for VPN_102 has been set to twenty, differentiating it from the sequence quantity 10 used for VPN_101. When you comply with this method, guarantee that all of your VPNs use the identical crypto map title (VPN_MAP on this case) with totally different sequence numbers.

Okay… this exhibits some critical community understanding. Combining a CSV file and a Jinja2 template collectively isn’t that removed from the code help and co-pilot use instances which have been in place for awhile—probably not that a lot of a stretch. And discovering a typo in a configuration template that’s most likely current in tons of of public Web documentation, weblog posts, and coaching examples? Additionally probably not that shocking.

However making this ultimate leap associated to community structure and design… this was actually fairly shocking. I didn’t inform Bridget that every of those configurations can be deployed on particular person routers, so she fairly “frightened” that I used to be trying to deploy the configs on the identical gadget. And if I did that, it WOULD be a problem. Her recommended resolution is affordable, although admittedly follows examples which can be additionally properly documented on the Web.

So what does this imply for us community engineers?

I used to be discussing this instance throughout Cisco Reside with some attendees, and the analogy of a “peer assessment” was introduced up. This was a use case for AI in community engineering that I hadn’t formalized in my head but, however had been dancing round with different examples.

There’s a very actual threat of engineers utilizing AI-generated configurations “as is” of their community.  It is a unhealthy concept for a similar cause as copying and pasting code from Stackoverflow or one other untrusted Web supply. Earlier than utilizing configurations generated by Bridget or one other generative AI device, it’s vital to confirm the accuracy.  In any other case, you threat changing into one other sufferer to hallucinations or simply inaccuracies from overconfident AI assistants.

Nonetheless, I feel there’s actual potential in utilizing AI as a peer reviewer for configurations and different artifacts we create. Right here, WE, the community engineer, use our data, expertise, and expertise to construct an answer. Then, we ask AI to have a look and supply some suggestions, much like sending it to a co-worker, associate, or Cisco TAC.

This second-set-of-eyes use case is one which I feel engineers ought to take into account attempting out for themselves—with a few essential caveats:

1. Verify together with your firm and group’s AI insurance policies. Lots of the publicly obtainable AI instruments take any immediate info despatched to them and use it for future coaching and fine-tuning. Community configurations are thought of confidential for a lot of corporations as there’s a critical safety threat in sharing them publicly.
2. I might nonetheless suggest having one other “natural intelligence” assessment any vital configurations. I’m excited by the potential of AI, and we probably will get to a degree the place AI is so correct we will depend on it for a lot of use instances, however let’s take the gradual method to get there.

Nicely, what are your ideas on this use of AI for community engineering? What different use instances are you exploring in your day by day work as an engineer? Let me know within the feedback. And keep tuned for extra from me as I proceed to swim into the Synthetic Age!

PS… This weblog publish was written by me, Hank Preston, and never by Bridget or one other AI device 🙂

Join Cisco U. | Be a part of the  Cisco Studying Community as we speak without cost.

Comply with Cisco Studying & Certifications

X | Threads | Fb | LinkedIn | Instagram | YouTube

Use  #CiscoU and #CiscoCert to hitch the dialog.

Cisco Helps Construct AI Workforce With New Expertise Certification

Navigating the Multicloud Journey with Cisco’s New Certifications [Infographic]

Share: