Ameer Abbas:
In my experience, I think I talked to over 150 customers in 2024. What I realized is that enterprises, the enterprise culture is nothing. The software development culture and enterprises is not ready for vibe coding unless everybody starts doing it, and I think you’d have to completely change the way you even think about documenting writing codes, your intent, your prompts need to be logged, all of those things because people are going to read different things in the future. They’re [00:00:30] not going to read lines of code, they’re going to read maybe the providence of how that code came to be.
Simon Margolis:
But just so that the audience also understands, if they want to take advantage of the Gemini Code Assist bells and whistles that we’re sort of teasing here. Does that mean they need to be developing everything inside of Google Cloud platform? Do they need to be using Google IDEs? What does that look like? Can they bring their own tooling platforms?
Ameer Abbas:
Yeah, I didn’t answer that question. So if you think of the software delivery lifecycle, generally we talk about two sort [00:01:00] of loops, right? It looks like an infinity. So there are two loops. There’s the inner loop, which is the developer’s experience if you want to call it. This is how you actually write software, and then there’s the outer loop, which is what happens when the software is pushed into the world.
That’s the operator’s sort of side of things. From the inner loops perspective, we’re not very sort of GCP specific or Google Cloud specific. We work in a lot of different IDEs. By the time you hear this, we’re going to have a CLI tool that is open source, [00:01:30] which is way more ubiquitous and way more approachable and universal. You can run it anywhere on any operating system, and we’re in source code management system. So GitHub being a very large source code management that many enterprises use, so you can use things like code review agents. So now you have a Gemini Code Assist bot sitting in source code management. We’re also in some of our own surfaces like Android Studio, Firebase Studio, so some of the first party IDs, but also inside the console. So in BigQuery, in app integration, and those types of [00:02:00] things too. So from an inner loop perspective, it’s pretty ubiquitous.
Simon Margolis:
Sure.
Ameer Abbas:
The outer loop and I think this is a differentiator, the outer loop, which is the operator’s sort of domain, we have a complementary product called Gemini Cloud Assist. They’re very similarly named. So you have the code assist and cloud assist, and you can think of it as inner loop, outer loop, code for developers, cloud for operators. Gemini Cloud Assist is much more GCP aware and very specific to your environment. So for example, let’s say an [00:02:30] application is down, you can just ask in natural language a question is why is my application down? Oh, it can go inside of your GKE environment and say, “Oh, you’re in crash loop back off.” Oh, it’s because your GKE clusters don’t have enough nodes.
Simon Margolis:
Sure.
Ameer Abbas:
Do you want me to expand the nodes? And then you can push a button. So it’s a lot more practical. It’s not just trained on generic world’s data and it’s just giving you some generic advice. It’s actually looking inside of your GCP environment and saying things like, “You [00:03:00] can expand this GKE cluster.” Or saying, “Hey, you’re paying too much for your cloud SQL instance. You can go down from an E4 to an N2.” Stuff like that. So it’s giving you cost optimization, FinOps sort of a things. It’s helping you with incidents, helping you resolve incidents faster, and again, it kind of goes down to if you want to improve the assembly line, you have to think of the entire assembly line including the outer loop, because if I’m pushing let’s say twice as many features out into the world, I want to make sure that there are complimentary set of tools on the other side so [00:03:30] they can resolve incidents, they can design applications, they can do cost optimization.
That thing is very GCP specific since you asked. So I would say the outer loop is a little bit today at least, and in the cloud assist world is a lot more GCP specific because it needs to have access and your permissions and all those things to your environment. The inner loop is you can deploy anywhere.
Simon Margolis:
That makes good sense. So it sounds like if you’re a customer and you want to take advantage of some of these things, you’re able to without making big shifts to your technical [00:04:00] landscape, but it sounds like incremental, additional added value to adopting more and more of the Google related.
Ameer Abbas:
And that’s what we’re seeing. I mean just going back to this tool, you and I kind of come both from the infrastructure side where it’s a lot more sticky. Once you’re on a Kubernetes, let’s say, I’m just using that as an example for your platform, whichever cloud vendor you’re in, you have to make that decision kind of upfront because it’s a lot of investment from skillset and also just it’s very [00:04:30] sticky versus if think about code assist, and if you think about the way most people think about code assist, which is just an IDE plugin, I’m just going to deploy it. It’s a very low cost of change, but what I would tell our audience that are listening is even though you can switch between, you can treat it as a tool, you can just say, “This is just an IDE plugin. I can go from plugin one to plugin two.” But again, you’re looking at a very myopic sort of one aspect of the SDLC. If you want to get the full sort of benefit, that’s [00:05:00] where we’re looking at the entire landscape.
Simon Margolis:
No, it makes good sense and again, personally speaking, that’s what I do is I’m all the way in the Google ecosystem using everything in one place, and it is a very magic feeling of when everything just works.
Ameer Abbas:
First party things just work together and it’s always going to be the case no matter what technology.
Simon Margolis:
Right. Okay, so zooming out a little bit, I want to address kind of some things that I hear I’m sure you hear from people all the time, which is, well, what does all this mean for the developer [00:05:30] ecosystem? Are developers going to lose their jobs? If I’m a student, should I even be pursuing a software development education? And obviously you’re not making any comments on behalf of Google or anything like that, but we’d love to just get, you’re so close to this, we’d really love to get your perspective on what are the next one, two, 10 years look like for this ecosystem and what might you say to that student that is pursuing a career in software development? What does that look like?
Ameer Abbas:
Right, first of all, I empathize [00:06:00] with them. The fear, I totally get it. Anytime something transformative comes in, it will raise questions. What do I want to do? And also I would compliment that by saying this is a very fast moving technology. So some people may even say well, by the time it gets there, I’m going to be done with my career. So you have to kind of take it with a grain of salt. I think learning what it can actually do, I think learning the bounds of what it does. So one of the things that doesn’t have is agency, and we can talk about agents too, but even agents [00:06:30] are prompted.
So you still need a human to use it. So what I tell people is this is the same concept as when we were going from, I don’t know, VMs to Kubernetes. There were people that were just stuck in their ways, or even when we were going from regular networking to software-defying networking. I started my career and networking, learning Cisco and CCIE, all those things and there was a lot of resistance doing something the new way. So to me, get rid of the resistance, lean into it, use [00:07:00] this tool. As a matter of fact, this is the only tool that I’ve worked on in my 25 plus years of working in this industry. It gets better the more you use it, which is a weird thing to say in a technology thing. You just get better at prompting, you get better at task management.
So to me, avoiding is not the answer. To me, this is going to take over and you can’t slow technology down or you can’t slow something down. So the way I think about it is I just kind of go with the flow, and so to me, if you’re a student [00:07:30] of software development, you still need to know software development by the way, you still need to know fundamentals. You still need to know designs and systems and those types of things. As a matter of fact, the better the software developers are, the better they are at prompting
And we can talk about the prompting portion as well. I’m also seeing kind of these two edges. So there is what I’m calling intentional prompting, and I actually stole that from one of my friends at Google, but I really like that term, which is I really want to spend time crafting the prompt so I don’t have to spend too much time looking at the output, because [00:08:00] that’s how these technologies work. You prompt it to do something and it generates an output. If you spend a lot of time and be very specific about what you want, it will give you something and you’ll know exactly what it’s doing, and then that becomes sort of your pair programmer, and then there’s the vibe coding on the other side, which I totally get the appeal for it.
Enterprises are not quite there yet and this is where enterprises still need smart individuals, smart men and women to come in and that know how systems designs, and that know how to use these tools to [00:08:30] the best of their ability and know when to not use these tools by the way, and know the wisdom to know the difference.
Simon Margolis:
Yeah, exactly.
Ameer Abbas:
So to me, we’re in the augmentation phase and even I think if we get into the automation phase, which we can talk about in the what’s next portion of this talk, it’s not like automation is going to replace augmentation. It is just going to compliment, in my opinion, augmentation at least for the foreseeable future. So to me, I would learn the way when I went through college, we didn’t even have computer engineering in my school at least. So [00:09:00] I did electrical engineering, which was adjacent to computer engineering because that’s what I wanted to work on. So now you have computer engineering and I’m hoping that you’re learning AI as part of that, but if you’re not, then get out there. It’s very approachable. It’s very easy.
Simon Margolis:
So that’s a great answer. I think that makes a lot of sense. I agree personally, for what it’s worth, I don’t see this. I tried a lot of these tools to just completely vibe code. I want an app that does this stuff. We’ll make it, and don’t get me wrong, there’s amazing technologies [00:09:30] out there right now and they keep getting crazier where sometimes it actually does an okay job at that type of thing, but I find those tools are so much more effective when I have an opinion as a developer, I’m just looking for some help and it does a really good job at that.
I almost like to equate it to one of the regrets I have is I was a Vim developer forever. I didn’t want to move into these crazy IDEs with other bells and whistles, but I do now and I’m glad that I do, and there’s a ton of stuff that makes my life way easier because [00:10:00] I use these software suites to do my development. I kind of view this like that, right? Could you develop code without using AI at all? Of course you could. Might you be significantly more efficient at developing code if you use these new tools at your disposal? Probably so.
Ameer Abbas:
And just to add to that vibe, I don’t think you can have a podcast about AI, especially software development and not talk about vibe coding. So vibe coding and every YouTube video that I’ve seen, it’s one person just doing something from scratch. [00:10:30] There’s a lot of problems with that. Enterprises don’t normally start from scratch. There’s always something that you’re building up on. So your starting point is not zero. Your starting point is 1,527, and then now you’re trying to get to 1600. So it doesn’t quite work. Secondly, it’s always one person. Enterprises work in teams and groups. There’s a lot of sort of cultural tribal knowledge that is not in vibe coding. Again, in vibe coding, you don’t even really look at the code sort of the point
And then thirdly, we don’t talk about the long-term [00:11:00] ramifications of vibe coding. If I write something by myself and then I leave the company and now you have to maintain it, how does that work? And so again, I’m not anti vibe coding by any means. I really enjoy watching those YouTube videos as much as the next person. In my experience, I think I talked to over 150 customers in 2024. What I realized is that enterprises, the enterprise culture is not the software development culture in enterprises is not ready for vibe coding unless everybody [00:11:30] starts doing it.
And I think you’d have to completely change the way you even think about documenting, writing codes, your intent, your prompts need to be logged, all of those things because people are going to read different things in the future. They’re not going to read lines of code. They’re going to read maybe the providence of how that code came to be, and so that’s kind of my view on it. I’m open-minded, but at the same time, today we’re not there yet. So that’s maybe another to folks that are in school or getting their careers started. I think this is actually a great time to get your career started [00:12:00] because there’s going to be a huge demand for software developers that know how to use AI to really 2X, 10X, 15X themselves.
Simon Margolis:
Absolutely. Yeah, because again, going back to what does the actual enterprise care about, it’s shipping features, shipping codes, and people that can use these tools well are going to do those things much more efficiently.
Ameer Abbas:
Those strategic and business imperatives have not changed. Everybody wants to up the stock price, make more money, make their business business more productive and [00:12:30] be competitively advantageous.
Simon Margolis:
Exactly.
Ameer Abbas:
And so those are not changing. There’s just basic human things no matter what technologies are going to come and go, and so if you can use this technology to further those things, have at it.
Simon Margolis:
Exactly. Yeah, no, I agree completely. So if you’re the enterprise and you have a bunch of code, you’re not looking to have a bunch of individuals do vibe coding, of course there’s a ton of YouTube videos, a ton of stuff out there for the individuals that do just want to get started individually to get started with these tools, [00:13:00] but I’m curious on your thoughts because in your role, you talk to lots of large businesses, you help them make sense of all this stuff and figure out where to get started. Do you have sort of blanket advice or guidance you’d give to an enterprise leader that’s listening to this conversation and saying, “Okay, I want my company, I want my team, I want to embrace these things.” How might they get started?
Ameer Abbas:
And I think that’s the first step is don’t wait, start now. Actually, I started saying that in the middle of 2024 is that the first step is don’t [00:13:30] just keep studying because this thing is moving a lot quicker. Just get your hands dirty. Look for low hanging fruits and stuff like that. Look for things that are maybe even not as I don’t know, business critical. Let’s just say that. That’s where a lot of people start. I would also say look for very specific use cases that have an end goal, that you have some sort of a metric, some sort of an exit criteria that you’re trying to achieve and know the limits and the bounds of the product, and so I think that really determines what you’re going to do with it.
[00:14:00] So actually my job is to really sort of let people’s expectations down to reality because I think there is this sort of a hype around this technology or any technology is that it’s going to just automate the world, which is basically where that fear comes from, it’s not. It has its limitations. Again, you and I are very close to this thing. We see the rough edges. It’s getting much better. Gemini 2.5 is markedly better than 2.0 and it’s only been six months or eight months. It’s getting better, but at the same time, and like I said, two people [00:14:30] can use the same product. You and I could be on the same IDE sitting right next to each other producing different quality code.
And so to me, having a very clear-cut use case is very important. Having a time-boundedness is very important. We also do a mixture of qualitative and quantitative surveys and we can publish some stuff in the show notes in terms of qualitative one, also the quantitative one, what are you actually measuring? Are you measuring adoption? Are you measuring trust metrics like acceptance rates, lines of code? Because not only do you want [00:15:00] to get started fast, but you want to prove that this tool is worth the juice is worth the squeeze so to speak. So I work with partners like yourself and enterprises to have almost like a journey in terms of how you should think about adopting this particular tool, starting very small and then zooming out.
And then I would say culture plays a big part and that you cannot influence in enterprises. So what I tend to ask the enterprises to find some champions or find some, maybe Gen [00:15:30] Z calls them influencers. So it’s very hard for you or myself as a vendor to go in and try to hype them, but it’s very easy if you already have these influencing hotspots in your software development. I like to talk to them and then I use them as a catapult to say, “Now, go ahead.” So it’s almost like a train the trainer sort of a situation.
In terms of using the tools, I mean that’s just a matter of just doing some workshops and you and I have worked on this together, that [00:16:00] becomes very tactical, but those are some of the big things is you can start today, which is where Gemini Code Assist to me is the best product to start with instead of starting with Vertex or Agentspace where you would still require some skillset and you still need to know how to build things and maintain it yourself. Code Assist, fully SaaS, we manage it, we maintain it, we support it. You just simply pay and play.
Simon Margolis:
Just go.
Ameer Abbas:
You just go and you can start today. I do a two-hour workshop with customers. I think you and I have done it together where we onboard 50, a hundred [00:16:30] people and set up their rag database and just show them some sort of good practices to kind of get them going because it’s not something that you can learn the chords, but then until you actually play it, you’re going to just get better and better and better.
Simon Margolis:
Exactly. Yeah, and I think that’s what you said it a couple of times, whether it’s for the individual, the student, the enterprise developer or the enterprise at large. I think the theme I’m hearing is just go, just get started.
Ameer Abbas:
And you’re measuring different things by the way at these different sorts of taxonomy or resolution. So [00:17:00] you’re measuring different things at the individual level as well, and you’re measuring different KPIs or different sort of metrics, and that’s something that we don’t need to get into the minutia of those, but we have a very nice blog that actually talks about that adoption journey, what to measure at each of those stages and how to both quantitative and qualitative metrics. Lots of our friends from Dora helped us with quite a bit. God bless them. They’ve been doing some amazing things in this space too. So we’re trying to really make this enterprise ready. [00:17:30] This is not just some experimental play toy thing that you can do. We’re looking at it from business KPIs down to security, privacy, trust, indemnification, those types of things, and that’s very important for companies that are in the FSI space, the financial services or any industry, healthcare industry. So those are the kind of things I would-
Simon Margolis:
No, it makes great sense. We definitely should link to that somewhere in the notes here because that’s what we hear from our customers all the time, which is how [00:18:00] do you know if you’re successful? What are those metrics? So I love that that’s already something you have put a lot of thought into. Ameer, I think we got to do a part two at some point because this is such a good topic and there’s so much going on and I think that if we blink, there’s probably a whole new paradigm to talk about as time goes on, but I think we can wrap it up here for today. In a few brief moments, can you sort of give your thoughts on where this is heading?
Ameer Abbas:
Yeah, and I think where we are right now is a very sort of interactive [00:18:30] sort of a mode with any tool. By the way, I’m just kind of talking generically and this is just Ameer’s opinion. This is not Google’s opinion. So no matter what tool you use, it’s a very sort of call-in response, I prompt you to do something and then you respond. I have to think about the plan. I have to think about what I’m trying to do, and basically what I’m calling, if we want to get a little technical is I’m doing one-step inferencing today. I ask you to do one thing, you do this, and then I decide what to do with that output. Then I ask you to do the next thing. I think where we’re headed [00:19:00] next is the world of agents, and you can think of agents, I know it’s a very broad word and we don’t need to go into that is something that has the capability of reasoning, planning, and it can do multistep inferencing.
Maybe that’s the easiest way to think about it. Going from this interactive single-step sort of inferencing to multistep inferencing, which by definition means that you need to understand what steps you have already taken, what step there are there in the future, and also for each of those steps, you may need access to different tools [00:19:30] and different systems and those types of things. So by the time this podcast comes out as I mentioned, I teased earlier that we’re releasing a CLI tool. So in addition to that CLI tool, we’re also releasing an agent mode inside of our IDE chat.
So instead of just saying, “Create this thing or create some unit tests or review this.” You would just give it a general prompt that would be much more generic, much bigger context. So we take the full context there, it reasons [00:20:00] first, it gives you a plan step by step, one, two, three, this is what I’m going to do. You’re constantly in the loop. So you can say, “Well, between steps four and five, I want you to inject this, or don’t worry about step seven.” Or something to that effect. So you’re constantly talking to it. So I think of it as it’s a true pair programmer at that point.
Simon Margolis:
Yeah.
Ameer Abbas:
Currently, I just give you a command and then you just respond with some sort of an answer versus we’re talking where you’re helping me plan things that maybe as a human I don’t even know, and then I’m letting you sort of giving you the [00:20:30] freedom to go and do those things step by step. We really think that’ll up the developer productivity. So that’s the true augmentation in my opinion, going from this sort of a tool-based single step to a pair programmer base. That’s where we are just about to head into, and then I think in the future, not that that’s going to really replace it, is this idea of automation. I think there’s two types of use cases. There are use cases that can be fully automated. Maybe you want to translate a bunch of things, maybe you want to upgrade a bunch of libraries, maybe you want to convert from language A to language B, those [00:21:00] types of things, where it’s maybe not the best use of human resources.
And this is where tools like Jools or whatever the version of that going to be, we call them just kind of background agents, and that’s one of the keys about having these tools in CLI that right now you have to go into the tool and then do something. So imagine in the future you can invoke the tool from other things. Right now the tool is the one doing the invoking, but imagine that you could be inside of your PRD in a Google Doc and say, “I’m going to invoke [00:21:30] an agent and pick off some asynchronous things.” So I think that’s coming in the very near and in my opinion, it’s the next 12 to 18 months sort of a horizon. I know it sounds very sci-fi-ish, but that’s where we’re kind of headed.
Simon Margolis:
I believe it again, based on everything that I’ve seen, I know what we’ve talked about and I absolutely see it and I can’t wait to see what customers start to do when these KPs are in their hands. So again, like I said, we could talk about this forever and I hope that we do a part two or a part three or a part 11 of this [00:22:00] because there’s so much good stuff here, but thank you so much for your time and thank you everybody for checking us out and watching this. We’re going to link to everything that Ameer was referencing in the notes here, and please let us know what your thoughts are.
Ameer Abbas:
Simon.
Simon Margolis:
Thank you again.
Ameer Abbas:
Thank you for having me.
Simon Margolis:
Thank you all so much for taking the time to check out this episode of Cloud and Clear. Make sure to subscribe, make sure to light this video and check out the next episode of Cloud and Clear as well. Thank you all so much.
