Cisco Bets it All On AI At Cisco Live | Tech Field Day News Rundown: June 18, 2025

Google breaks the internet. Ion Q Bonds with Oxford Ionics Synopsis opens up for startups, unified threat naming, automating telemetry, arista's new COO. And we take a look at all of the AI announcements from Cisco live in this week's episode of the Tech Field Day Rundown.

Hello everyone and welcome to the Tech Field Day rundown for June 18th, 2025. My name is Tom Hollingsworth and I have successfully avoided the June gloom that is in San Diego, at least it was last week. And I am back to be happy to host the rundown with my good friend Mr.

Alistair Cook. Al, have you ever just wanted to get away? Well, some days I do just want to get away and today is of course, national Want to Getaway Day.

And uh, it conveniently is also national Go Fishing Day if you want to get away and go fishing. And that seems to be a pretty popular thing to do around where I live. People head outing, head out onto the, uh, Pacific Ocean from here and look for some fish to, uh, to make a meal outta, which is a great way of spending a day.

Exactly. Well, the good news is, is you don't have to splurge on anything expensive to catch the great news stories that we have coming up because we've already taken care of that for you. So we're gonna dive right in with one that probably affected you last week because a major Google Cloud outage on June the 12th disrupted large portions of the internet, which affected, you know, minor services like Spotify, discord, Google search and Google Meet.

The root cause was tied to an issue with Google Cloud's identity and access management platform, which also impacted things that depend on that, like, uh, CloudFlare, that, that was another big problem we ran into. And of course, as is usually the case engineers implemented the fixes, some of the services began recovering, but then everybody was talking about this, there was a lot of outrage highlighted the risks of relying on centralized cloud infrastructure and of course made a whole bunch of companies start saying that, Hey, at least we're not Google Cloud 'cause we don't go down. Uh, although none of those was CrowdStrike.

Al what's your take on Google Cloud's minor little curve level? Well, I think there's a, a couple of aspects to this. One of them is that every IT system goes down now and there and as they say, uh, eventually software works and eventually hardware fails.

And between those two, there's usually something going wrong somewhere. And so the fact that Google has had a a significant outage here, Google Cloud has had a significant outage is to me no more significant than previous outages of cloud providers. Every IT system goes down at times, but what it does do is highlight how much we've centralized our use of it.

And a relatively small number of providers, as we see have seen in AWS's outages of, uh, on occasion that large s swathes of our IT estates are dependent on these, these clouds. And this consolidation into a smaller number of clouds means that the blast radius, the impact of a single failure is much larger than it was when we had a much more distributed decentralized environment. There is of course, a large argument around designing your cloud application for failure.

And this is often where the cloud providers say, well, if you'd followed our design rules, you wouldn't have had a complete outage. And unfortunately, usually the IAM the identity and access management piece, the fundamental permissioning who's allowed to do what in your public cloud environment, uh, this is one of those services that you can't get around even. Uh, I'm far more familiar with this.

On A-W-S-I-A-M is a global service on AWS and, uh, if there's a fault in IAM, it's gonna affect all of your regions. Uh, it appears that the same is true on Google Cloud. I'd be surprised if it wasn't true on most of the other infrastructure as uh, service clouds as well.

But if the, uh, the identity and access management component breaks down, well, you're no longer authorized to access anything that's kind of better than when it breaks. You're allowed to access everything. 'cause that would make it much more of an attack target.

But it still brings back that central idea. We're we're putting a lot of reliance on a relatively small number of providers, and if one of those providers goes down, there's a pretty significant impact to that. Large numbers of websites and, and applications were offline while this particular, uh, incident was occurring.

Now, unlike some of the incidents we've seen on premises, the RESO resolution time was relatively fast. It was certainly not minutes to resolution, but it was hours rather than days to resolution. And some of the failures that we've seen on premises, uh, we've seen organizations take weeks to months to return from, uh, some of their failures.

So on, on the terms of scales of failure, yes, it was a significant failure, it had a lot of impacts and it was downfall longer than we would've liked. But this is the real world. Things go wrong, things get broken, and, uh, in the increasingly connected and centralized world, uh, connected to large numbers of, uh, locations from a small number of providers, yeah, that tends to be a high blast radius for these faults.

Can you design around it? Can you design your infrastructure to cope with one cloud provider having a failure and another cloud provider not having the same failure maybe, but the cost is gonna be outrageous to do it. Is it really worth it In order to avoid these relatively few relatively short failures, it's always a business decision to invest in higher availability or not.

It's very hard to get high availability across multiple clouds in the world of quantum ion Q is a US quantum computing company and they're buying the UK based Oxford. I strange, there's lots of ions going on. 08 billion first billion dollar deal in the quantum computing industry.

Uh, the goal is to combine ion q's powerful quantum systems with Oxford's, uh, chip based technology, just much easier to produce it scale the move helps iron q advance its plan to build large, reliable quantum computers by 2030. And the deal also boosts Iron Q's position as a leader in the growing quantum tech space. Tong, we talk quite a lot about quantum and particularly around when quantum computing is gonna be big enough and reliable enough.

Do you think 2030 is a realistic goal for INQ? I think it is. I just don't think that what they're gonna be selling is gonna be very cheap.

And the reason for that is because the amount of effort that's gonna go into building a quantum computer out of these components, there's a lot of inputs to it. So the short, short version for all of you out there that are wondering what the heck is an ion quantum computer, they essentially take ions, you know, just little elementary particles that have a charge and they stick them in a field. And effectively what they're doing is they're using electromagnetic spectrum to trap these ions in this field and then they super cool the whole thing to just a couple of degrees above absolute zero so far so good, right?

Well then you have to manipulate those, uh, ions to be able to get data out of them, right, like any, any traditional computer would. And you do that using this complicated thing known as a laser. And so the problem you run into is one, the inputs cost a small fortune.

'cause the amount of, you know, things like liquid helium that you have to use to keep these things super, super, super cold, but also the complexity of the way that the lasers need to interact with everything makes it a lot more difficult to just kind of spin one of these things up. And I mean, we've seen that with a lot of quantum computing, even I I say startups, but like Cisco started a quantum computer lab out in, uh, California. And, and one of the things that's a big problem is the fact the amount of power that they have to pump through those lasers is well frightening.

And you think about it. Uh, but ultimately what this means is, is that these two companies realize that in order to bring something to the market for them to be able to get a return on their investment, they're going to have to examine how to make this cheaper and more reliable with less precise inputs. And think all the way back to the late 18th century when Eli Whitney invented interchangeable parts, uh, back then anything you made was custom built.

And that meant that whoever built it needed to be very, very good at what they did, right? So if it was a hunting musket or any kind of like a steam engine or anything like that, it was all custom built and that meant whoever built it had to deliver it and had to assemble it. And it was a pain in the, but with interchangeable parts, I could just say, okay, take three of these and two of those and put them together.

Like quantum computers are kind of where we were back then where everything is kind of custom and bespoke. If they're going to make this more of a, you know, marketable thing, they have to make it so that maybe not you and I but somebody can build one of these without needing to have like all of this kind of crazy training to make it work. Can they do that in five years?

Well, I honestly, I, I don't know, five years ago if you've told me that people would've even had quantum computers capable of what we have today, I would've probably wagged my finger at you. But the point remains that there is a large gap between something that is available for sale and something that is available for sale cheaply enough where everybody can buy it. I think what we're going to see is that Ion Q and Oxford Ionics are going to figure out how to make a cheaper quantum computer.

That doesn't mean they're going to make a cheap quantum computer. I still expect this thing to cost multiple millions of dollars to put together, but even at multiple millions of dollars, it puts this kind of technology into a space where large organizations that potentially could need that can buy it and use it as opposed to having to farm time out on these large, large systems that are effectively backed up for months or years at this point because it's the only game in town if you wanna do those kinds of things. So we'll keep an eye on it.

I mean, it is big that it's a billion dollar acquisition, but billion dollar acquisitions also have to produce results if you wanna pay off in the long run. Semiconductor company Synopsis is partnering with plug and play to help chip startups get easier access to their powerful design tools. They're piloting an accelerator program at which selected startups will use synopsis software and intellectual property to speed up development and to cut costs.

Overall, the goal is to support faster chip innovation across industries and help grow synopsis's role in the deep tech space. And of course, plug and Play is a company that's kind of, uh, providing funding and is maybe a little more eye towards the results of those companies being able to use Synopsis intellectual property and tooling. Al what's the ultimate goal here?

Is it for Synopsis to make more money or is it for these companies to be able to get a jumpstart on the semiconductor design and manufacturing process? Well, there's not an or in there, it's an, and so it is a dual objective here, the objective for plugin players, for the startups that they're mentoring, that they're investing in to be more successful, to decrease their failure rate, uh, as they're building new hardware products. A lot of Silicon Valley innovation is based around software, and software has a very fast design cycle.

I can write some code today and be testing it tonight. Uh, hardware is a little tougher, particularly custom Silicon Hardware. Uh, and that's the sort of space where Synopsis is working in here.

They have some electronic design tools for designing actual custom silicon, and it's these tools that are being made available to the startups that are working with plug and play. Uh, the objective here is, is really for plug and play to be helping these companies to be more productive and therefore get better exits. Uh, and of course Synopsis gets their tools in front of a whole bunch of new startups who are doing innovative things.

And naturally, if this is the tool you started building your hardware with, you're probably gonna continue to build your hardware with this. Uh, I think the really significant thing in here is a recognition that hardware is not just the game for Intel and a MD to be building and Nvidia, that hardware solutions are a place for innovation for smaller companies to innovate and to build new solutions that are not just more of the same, more cause scaling out or, uh, faster. Um, Zion CPUs right there is a place for innovative new hardware to be built.

Um, and it's not all just software running on general purpose CPUs, but also a recognition that that change to designing hardware, particularly designing hardware. Uh, custom hardware for new solutions is very high risk and very long, uh, time for return. Now, Silicon Valley's, uh, venture capitalists, they like a return within 10 years of investment and they wanna see a 10 x return as their preferred kind of run rate because 90% of the people they back won't give them that 10 x return.

So any tool that's gonna make it easier for those startups to turn their idea into some real production silicon absolutely is gonna be really beneficial to getting that innovation, getting those startups to operate and therefore getting a return. And I think we will see another wave of innovation in custom hardware. We're already seeing it from the larger companies we're seeing, uh, NPUs, dpu, whatever offload you would like to consider.

Uh, as of a really significant part of the innovation at the moment, a lot of those innovations are around arm cord CPUs and maybe software just running on those arm cores. I think this synopsis, uh, deal with plug and play is oriented more towards developing truly custom hardware, not just new firm. We're running on embedded CPUs.

So it will be interesting to see in five years time whether we start seeing a, a large set of innovation in new hardware, new types of hardware being available, new expansion cards or possibly new device types entirely turning up on the market. That will be really be the proof of this one. Microsoft and CrowdStrike have teamed up to make it easier to identify cyber threat groups by combining the different names that various security companies use for the same attackers.

This helps reduce confusion from having too many names for one threat and one threat group so that security teams can respond faster and more effectively. While it's tough to create a single naming system, this partnership is a big step towards a clearer and more consistent threat. Information in cybersecurity, cybersecurity, less confusion over whether it's a salt thing or a, uh, typhoon thing that is coming to attack you.

Tom, do you get a clear idea of whether a salter is a typhoon or a typhoon is a type of salt? Uh, it depends really on which part of the world that you're in. Um, but, uh, I, I think that being assaulted by a typhoon is probably the worst thing that could happen to you, especially if you don't even know who it is or where it's coming from.

And that's one of the biggest problems that we've run into as of late, is this idea that everything needs a catchy name. Right? And look, I get it, like CV 2025, alpha Baker, purple Monkey.

Dishwasher is not the greatest way to refer to a thing unless you're a machine. However, trying to come up with a creative name for a thing that actually Desi like talks about what it does is also kind of maddening. Like, do you know what Heartbleed does right off the top of your head?

Probably not, because you have to dig into the details. And it doesn't help that if a group that did something is detected by say, Microsoft and then a different group say, I don't know, Mandiant looks and sees the same group but isn't aware of what Microsoft has discovered, they might give them a same name, which is why we get up, you know, weird things like Apt 31 or Fancy Bear or this or that, and it creates confusion. And I'm gonna pull out my nerd card here for a second to, uh, illustrate this point.

Uh, if you've ever played Battle Tech, you are familiar with the 10 meter tall, uh, robots that run around and shoot people and, and do all kinds of fun stuff. And at a certain point in the lore, uh, a group of these things showed up out of nowhere and the people who were fighting them were fighting over who got to name them the first time. So in one particular case, one group of from one country called it the Vulture and Theano another group from a different, uh, called, called it the Taka Haka.

And even the people who piloted the machine called it the Mad Dogs, you had three different names for the same thing. And if you're reading the books and you're like, okay, which one was that? Is that this one or that one?

I don't know. And some of you out there who are familiar with Japanese may say, well, wait, haka is the Japanese word for vulture. Yeah, but it's still Japanese.

And I'm not sure which one you're talking about there, especially when you consider that there's a masica and a Dai shi and a whole bunch of other things. The point is, is that I'm throwing all of these words at you and I could be referring to the same thing, but I could be referring to different things. And unless we know how to deal with that specific thing, we can't fight countermeasures against it.

And if we have some kind of a unified model where one group or one group of people will say, okay, this is the name for this group that produced these things, and this is how you defend against them, then we can start spending more of our time solving these problems and combating these attackers than trying to give credit for who deserves to be able to name something. I mean, Lord knows we've got that problem right now. Is that mountain in Alaska Mountain McKinley or Denali?

The ultimate answer is, who cares? It's a mountain, it should be there. And no matter what we wanna call it, uh, I tend to lean more towards the Denali crowd because that's what the people who've been there for thousands of years call it.

But the point is, is that we need to have one authority that says, this is how it's gonna be, and this is the name we're going to use. I applaud Microsoft, I applaud CrowdStrike for trying this. Uh, but I will also tell them that there is another common problem in the industry, which is if you wanna make a standard naming convention for something or standard anything, the easiest way to do it is to decide what you want it to be called and get enough people behind your, your movement so that everybody else in the industry aligns against you and creates a standard directly opposite of the one that you wanted to.

You can ask Cisco how that worked out. I think ultimately though, this is a good step forward, and here's hoping that this one can stick. Who knows?

Cribble has introduced the new copilot editor, which is an AI powered feature built into its cripple stream platform, which is of course available to all their current users at no extra cost. And it automatically maps and normalizes telemetry data, which understands log structure and semantics to translate schemas standardized formats, build pipelines and route data analytics tools and minutes rather than hours or potentially days. Uh, this human in the Loop tool not only accelerates onboarding and reduces vendor lock in by enabling streama schema agnostic transformations, but also optimizes cost by filtering noise and enriching critical fields for it and security use cases.

And as we saw last week at Cisco Live, uh, when you feed a bunch of log data or packet captures into a standard LLM, it tends to barf everywhere. So I guess the question that I have for you, Al, is does an automated telemetry system really provide value for the people who have to comb through logs every day? And I think the simple answer is yes.

So maybe we'll move on to something else. Now. The more complex answer in it is, is also yes.

Um, you know, what we are seeing is that general purpose LMS are designed for for handling human to human conversations. That's what they were trained on. Logs are not human to human, they're machine to machine.

And so specific training for handling the kinds of data that are handed off machine to machine is gonna yield a, a large language model that can handle that. It can understand the, the way that communication is formatted and structured can do things like recognizing that probably the first block of data is going to be a date, but that depending upon where you are, that formatting of the date or what the source is that formatting the date might be different. So things like simply being able to recognize that it's a date, what the source format is, and then output that in a standardized format.

And that's what data normalization's about. Uh, absolutely there's value in that. Certainly a human can look at a date and say that's what the date actually reads at, and maybe write some logic to transform it into a standardized way of writing the date.

Uh, although Google Docs doesn't seem to be able to cope with the fact that there isn't just a single standardized way of describing a date, uh, ask me how I know that, uh, US state format is almost all that, uh, Google Docs will work with beyond just that simple, simple example of the first field is probably a date and type. There are sort of commonalities to how log data is sent, but there isn't a standard for this is the order for this set of fields. And so again, having a tool that can look at the set of fields and look across a, a reasonable sample of the, the data coming in through a log stream and say, well, we can divide it up, we can reformat it, we can actually re represent it maybe as a series of key value pairs that are more identifiable to our application than the straight stream of text.

So yeah, I think there's some real value to be had, and this is where AI that understand specific data types are gonna be very beneficial to us. I like that this is still human in the loop because I still don't trust completely AI generated content in even the most specific use cases. Um, I still, particularly as we're we're looking at new types of data coming in, want to do some validation.

Uh, as always it's around those edge cases. Uh, what about an elite? Yeah, is it gonna, does this transformation going to cope with maybe a, a number that's usually a, uh, degrees centigrade temperature, which shouldn't get above well a hundred, um, but maybe, uh, one of the sensors is gonna send me in Fahrenheit or maybe, uh, when I go in and change the air conditioning in my hotel room to read out in far, uh, in centigrade rather than Fahrenheit when I wanna change the temperature.

And is that gonna change the results that come through in this, this log screening? So yeah, it's, it's absolutely awesome. The more we can have AI tools that deal with this routine, um, drudge work that is part of it, building applications or building in this case, um, log data pipelines, the more we can use AI to deal with the drudgery piece and have the humans deal with the more creative and, uh, innovative piece, the, the better.

Uh, that's absolutely what we want AI to be doing for us. And so this ai co-pilot that we often see an assistant that runs alongside you and deals with just the minor stuff that might even be called the, the AI executive Assistant, um, would, is absolutely a great use of ai. And I hope we've seen more of this taking the drudge work away from people.

Arista has a new chief operating officer. Uh, the company announced that they were hiring Todd Nightingale for the role. Nightingale was most recently, uh, the GM of Fastly.

But most people in our audience know him as the VP and GM of Enterprise Networking and Cloud at Cisco and the SVP of Cloud Meraki. Uh, he ran the unit from 2015 until 2022, uh, when he departed to be the CEO at Fastly, he was highly regarded leader by the customers in the industry. But some insiders say he was overly protective of the Meraki culture and resisted the efforts to unify the brand with Cisco that unification happened after he departed.

But if he's a will like leader and he is now inside Arista, that's a whole different set of culture to be working with and protecting Tom. Is he gonna be a great steward for Arista? I think he will.

And I'll tell you that back in, I think it was mid 2022, so this is just before his departure. I watched him give a speech at Cisco Live and the amount of passion that this man displayed while he was on stage, I turned and looked at the people in the room and I said, that man is gonna be the CEO of a company one day. It just won't be this one because we all know how things go if the CEO chase in a large organization like Cisco and moving to Fastly was probably the best thing he could have done because it gave him the opportunity to be the head honcho for a while.

And so now he's stepping into the COO role at Arista. And some may say, well, you know, that's a step down right from CEO to COO. I would argue that this is maybe more a step in the right direction because Todd has a lot of experience running networking, and I think that he is seen as a passionate person who understands the technology and is ready to help people do things right.

Like, like that's always been the drive that he's been behind. Uh, yes, there were few insiders that I heard as soon as the news broke that said that he was more of an obstacle, at least in their, uh, experience because he very much resisted the integration of Meraki proper into Cisco. And anyone, you know, from that acquisition up through 2022 knew that they were very bright lines.

You know, this is Meraki, this is who we are. And that's different than these people over here. Arista has a very strong culture.

Anyone who's ever watched a Tech Build day presentation that is involved, someone like Ken Doda knows that they're very passionate about what they do. Ken notoriously is a vehement advocate for code quality. And I think that Todd and Ken are matched horses.

I also think they're gonna work well with Jri because Arista has a drive, they have a vision for what they want to do, and they've been executing on it. And, and we've talked in the rundown in recent weeks about, um, making acquisitions to possibly complete some of that vision. We've seen presentations from them as of late to kind of making a more holistic, uh, you know, c campus to data center type, uh, offering.

And I think that Todd Nightingale brings on board a seasoned leader who knows how to manage those transitions, who knows how to position those products so that people understand that value. And it cannot be understated enough that the company that he worked for Meraki has a very strong cloud management background, and Arista has cloud vision. And I feel like that is something that they're going to be developing even more deeply now to allow people to kind of build out networks that are technically proficient, but also easier to manage, which as we've seen from a lot of companies over the years, is the way that a lot of folks want to go.

Um, and of course, you know, we're, I'm sure we're gonna be talking about it very shortly about how ease of management is, is a driving factor for a lot of people. So I, I hope Todd settles well into his role over there. Honestly, I hope to see Todd at a future tech field day event.

I'd love to get him in front of our crowd and have him display some of the passion that I've seen him with in the past, and I think it will work very well for the people there. Well, it's now time for us to take a closer look and Cisco Live, uh, was last week Cisco live us and the opening keynote was dripping with AI as all good tech keynotes are these days. Cisco's product lineup was the, for the rest of the year and into the future, is going to embrace AI agents and information for networking operations and security as well.

The CEO, Chuck Robinson, chief product Officer Jett Patel, talked about all the integrations of AI into their offerings from dashboards to NVIDIA network technology. Cisco's pushing AI into infrastructure. One of the big discussions was around agentic ops, the Cisco term for agent based networking operations.

They also discussed how they would help build the AI data centers, uh, for the future with pod based architectures and more. Tom, you were there, apparently it was a little gray for, uh, San Diego. What did you see at Cisco live?

Well, hold on just a second. Let me consult my LLM. What were the big announcements from?

Yeah, it's gonna take too long. Cisco's going all in on this, but we know they had to because that's what the market is telling every company they need to do. We, I I, the way that I described it to people was it's the look busy, the boss is coming mentality.

You need to be doing something with way ai, what I don't know, but I need to see that you're doing something. And, and I get the perspective that they're coming from. And honestly, they're, they're kind of taking the right approach.

They have years, decades at this point of institutional knowledge that is built around the idea of building better networks, right? Like you just something as simple as a Cisco solution reference network design. That is a way that you're supposed to build a network with Cisco gear, makes total sense.

Now, implement all of that into an AI model and have that AI m model consult on the way that you're building things. Seems pretty easy, right? Until you start hitting hiccups.

And that I think is where this is going to matter more than anything. What happens when my brownfield doesn't look like your perfect ivory tower design? How can I adjust for these things?

How can I ensure that this piece is gonna work if that piece fails? Uh, you know, even stupid questions like which routing protocols should I use to tie these things together is not always a straightforward answer depending on what kind of technical debt you're dealing with. The other thing is, is that there's a lot of talk about AI security, right?

Like using AI to do threat hunting, to notice patterns, integrate that with Splunk, be able to offer that to more people. The idea, and this was on every petty cab in, uh, San Diego that was right around the convention center, you know, ask us about agent ops. The idea is, is that they wanna build these agent-based AI systems that are going to help operations teams run their networks.

Will it work? That is a big question for people because as I've discussed with people, even just today, there is an issue between operations and engineering personnel that will ask AI a question and then they will immediately question the results unless it's something they have deep knowledge about and trusted implicitly. And that trust takes time to build Cisco's agent stuff may come back with a little bit better of an option because it's like, well, in a perfect world we would do this, but we also still have to implement those solutions.

And not everybody lives. In a perfect world, I would hope that this will turn out for the better because everybody seems to need AI in everything that they do. And not to be outdone, because this is the other thing you mentioned, they're also building the infrastructure that will run those massive AI engines.

Cisco's one of the first companies that was allowed to build Spectrum X network technology that isn't named Nvidia. Now I've said in the past, I think I mentioned this, uh, after we heard the initial announcement of this just after Cisco Live Europe, that one of the reasons why I think this is happening is because Nvidia realizes that Cisco has a very tight hold on some accounts in the market. And the only way that they're going to be able to get in there is if they partner with a company that has a proven track record.

Not saying Nvidia doesn't have a track record, but when's the last time you went shopping for an NVIDIA switch that wasn't built into an Nvidia AI pod? The answer's probably not. Um, it, it goes back to like when I used to deploy blade centers in, uh, in school districts, um, most people would just use the cheapest switch option that was in there, even if it didn't run anything even remotely related to an operating system that I could work on, which is where a lot of my frustrations came from trying to figure out how to make these switches work.

And more than once I told people, I'm like, why didn't you just order the Cisco switch? They're like, well, it was too expensive. So I think that that Cisco's trying to position themselves as being a, a key component of AI infrastructure in the future, while also using AI to help their infrastructure, you know, be easier to use, be easier to manage.

I don't necessarily know that that means that you're gonna be reducing operations personnel because that's usually the first thing that people say, right? It's like, oh my God, they're gonna get rid of, of me and I'm not gonna have a job because my job is, is this, this and this. And like, either your job is doing things that should be automated already, or they're gonna keep you on and have you tackle the hard problems, in which case you get to learn and grow as a network engineer.

And that's how all of us started out. Like I wasn't birthed knowing everything there is to know about the CCIE, that that came later, but it was when I got into those problems that were not just a quick fix, a quick typo that I was able to say, okay, this is how these interactions happen. I used it for the experience that it was, and I'll tell you that making mistakes is something, one that people need to do to learn.

And two, AI is surprisingly good at and surprisingly bad from learning from. I mean, Al you've, you've worked a lot with AI and honestly, you being the event lead for AI infrastructure Field Day have seen a lot of companies that are doing AI infrastructure. Do you think Cisco's gonna be able to keep up with this?

I think there's, there's some really cool things in what Cisco has announced. So definitely that, um, integration with Spectrum X using, uh, certified connectivity between the Nvidia nick that might be in your GPU pod and your Cisco switches, that seems like a pretty big deal. And the whole idea of having this secure AI factory and having some AI defense within that secure AI factory notice, there seems to be an awful lot of security being talked about in here in AI because, so I was, uh, you writing yesterday in an article, uh, there's a whole new attack vector into your organization, which has poisoned, uh, models turning up on hugging face and that these, this is a, a great way of getting a, a new type of executable file inside your nice squishy soft network.

And so, yeah, I like that there's a big focus in on here on governance of these models and guardrails and a lot of the things that are important when you come to running these models in production, because running in production is where you're getting business value outta your investment in ai. And so having a, a design here that focuses on what's important when we're running in production governance guardrails, that's making sure that your AI doesn't go rogue and start sending out all of your internal HR information at a request from somebody out on the internet. These pieces are really significant and important as you're building out an infrastructure.

Um, Tom, you talked about trust and Cisco is pretty significantly trusted in most of the large enterprise organizations around the world. Um, um, the cost of Cisco may not be the most popular thing, but then when we're talking about AI infrastructure, your network infrastructure is not particularly expensive. Next, suppose $20,000 a piece, GPUs that you are buying 20, 40, 50, a thousand of, uh, you know, the network infrastructure certainly is a significant cost normally, but not when there's GPUs come into play.

Even the X 86 server cost and your SSD cost is small compared to the GPU cost, let alone the licensing that comes around every month for the software that you've got on top. Uh, where was I going with that? Oh, yeah, Cisco trusted, liked, um, good reputation for a liability.

One of the non-AI things that I thought was really significant in here was getting this Nexus dashboard that is unified management for both NEX os as well as the, uh, the, uh, a CI, the, uh, software defined networking and bringing hyper fabric in as well. This has long been a criticism for Cisco of we've got so many different consoles for managing the different pieces, bring them all together. So I like that coming along as well.

And this idea of an AI factory is really important for a large enterprises getting started, getting to the point where you can actually run some AI applications. The first use case is difficult, but scaling to hit 20, 30, 50, 200 different use cases within your organization is incredibly difficult. And so the more prescriptive guidance, the more of us scalable and particularly scale out architecture that you can build for hosting these applications over time is gonna be really critical.

I think Cisco's really well placed here. And of course, Tom, you've been a Cisco fan for a while. Uh, do you see a Gentech ops being a thing that's going to help you as, as now the CCIA?

I think it's gonna move the needle because it's another tool in the toolbox that Cisco certified individuals can use. Uh, I don't know what the adoption rate's gonna look like though, because we've seen this a lot from the past and honestly, talking to a bunch of people around Cisco Live there was chuckling. It's like, oh, look, it's Prime again.

Oh, look it Cisco works again. I I, I used those for a while and then they put them out to pasture because the next thing came along. This one's gonna have because it's being fed and, and and growing, you know, kind of like an AI algorithm normally does.

But ultimately, you know, you get to a point where the system's gonna be running and then someone's gonna go, well, if it's not really providing us UpToDate information on a regular basis, why are we paying for this? Because we all know that the way, reason why to do this is because it involves a yearly licensing fee, right? This is, this is the ultimate completion of Chuck turning that big ship, turning them from the John Chambers box pushers r us into the sleek and subscription based Cisco of the future that makes the investors very happy.

But I also think that there are challenges that are gonna be faced in by all tech companies with some headwinds and policy decisions made above the CEO level by people in Washington DC that are going to have to be overcome in order to make sure that these things do not be, are not regulated into oblivion or worse yet unleashed on an unsuspecting populace to do things that could create problems down the road. But I don't wanna naysay too much about that. I just wanna sit back and see if people adopt this or if not, uh, maybe I'll just need to start studying for the CCIE ai.

Who knows. One thing you can do though is stay tuned for Tech Fuel Day because we have great things coming up. The first one is actually happening right now.

You know that we are a part of the Futurum group and every June the Futurum group has the six five Media Summit this year of course it is AI Unleashed and if you missed the keynote from Michael Dell, you can go back and watch it along with Watts of great coverage, including GTU Patel c Chief Product Officer of Cisco talking about some security related things. com and check out six five Summit. We would love to see you there.

I will be back in July after US independent day to be bringing you networking field day 38. We have a great lineup of presenters and some brand new delegates who will be joining us for the very first time. com very shortly.

So make sure that you have all of those sites bookmarked. Thank you very much for tuning in for this episode of The Tech Field Day rundown. Don't forget that we do publish new episodes every Wednesday.

We love to bring you the news on Hump Day, and if you wanna subscribe to us on YouTube or possibly in your favorite podcast application of choice, we would appreciate that. Also, don't forget to leave a, uh, comment below. Let us know what you thought of our stories, whether or not we were snarky enough.

'cause Lord knows we can always turn the snark up to 12. Uh, that would make it a lot more fun. And check out Al and I and Steven as well as many other great people on other tech strong TV and future and group properties, we would love to, uh, hear you there.

We'll be back next week to talk about all the things that happened in the Enterprise IT community for the last week. Until then, for myself, Tom Hollingsworth, for Al Cook, Steven Fossett, and everybody else here at Tech Field Day, thanks for tuning in and we'll see you next week.