AI Implementation in Go-To-Market (GTM)

I am super excited to have Abraham Gomez today, on the podcast. He's a customer engineer at Google. He's worked with the the best of the best of the startups, creating 11 fireworks, increasing Orbitz as a VC as well. And I think he's one of the best person out there to give us a bit of a pulse on where AI is heading within companies. He's building the AI for the most ambitious companies and giving CEOs and CTOs those tools that they need to actually grow faster. And, hopefully, in the episode today, we're gonna try to unwrap how we can steal some of the stuff that he has done and how we can actually incorporate that into everyday companies so that they can hopefully grow at the same speed as those companies in in Valley. So, Abraham, welcome to the podcast, and and thanks for being here. Oh, appreciate it. I'm really excited to, you know, talk to you. Just to get going and set a little bit like the stage here on the podcast, I have one first question for you. Mhmm.

What does a customer engineer do, and what does it mean? It's a good before I would joke around with my family, they're not technical, they would sort of ask and poke around, like, what do you do? I would just tell my family, look. I I sell very nerdy things to very nerdy people. Right? That was sort of the best I could explain it. I tried to explain the clout to them. You know, it just was hard. Then I think in 2022 when ChatGBT came out and then 2023 sort of hit the masses, right, it was a lot easier to explain my role. It's like, oh, you know, ChatGPT? Well, Google has their own version. And, essentially, our goal is to get a set of customers to drive their adoption of Google Cloud services. And, obviously, in the last, you know, three years, a big part of that was using Gemini. Right? So walking startups, that's usually that's where I focus on is walking startups into adopting things like Gemini. One thing that I've been very fortunate, I've always I looked through my calendar to see how many founders I've met, and I found that it's way over 400 now. Right? And through that experience, like, if anyone just you know, if you were to meet 400 founders or 400 dentists or 400 lawyers, right, and you have a conversation centered around AI, you'll learn. You'll see patterns. Right? You'll you'll you'll keep in touch with them. And, you know, it's been really interesting, but I'm trying to share out more with with people.

I'm very excited to unwrap some of that on the podcast. Like you said, you joined five years ago where back then in 2020. And to what extent could you have predicted that today, would be pushing multi AI agent systems and companies automating entire functions? Can tell you me a little bit more about this journey and how fast it has accelerated

maybe the past three years? That's a good question. I I when I joined Google in 2020, and I sort of split my my time at Google from 2020 to 2022. Sort of pre chat GBT time. Or every now and then when we would try to position AI to startups, they're mainly focused on machine learning, was rare even to start ups between 2020 to 2022. When 2023 hit in 2023, for me, it was influx of start ups really just diving into the deep end. It means to actually implement AI because in the past, a lot of traditional companies, right, were terrified of AI because it's a black box. And a lot of these companies, they can't operate with a black box system. They need to understand what's happening under the hood. If you look back in 2024 versus 2026, it shifted so much in just those two years. Right? There is an expectation. And and, you know, some conversations I've had with maybe non startups, which is, like, you know, bigger enterprises, is that AI is now also seen and viewed as a productivity tool, which in 2023, it wasn't. Right? 2023 was a feature or a product that you could create. Now I think everyone sort of knows, oh, yeah. This is a tool to help me you know, just like when the spreadsheets came out on computer. Mhmm. Right? Now it's it's sort of table stakes that you need these tools, right, to just help you throughout your your day at work.

You, as a customer engineer, you are basically you you you need to be able to talk about both GPUs, TPUs, scalability of whatever solution that you're pushing. But at the same time, you're working very closely with CEOs, CMOs, CTOs, and so on on the business side. Yeah. So can you tell me a little bit more about how can you strike this balance, present Gemini products both from a technical standpoint and a business perspective?

Yeah. No. No. That's a that's a good question. I think well, one, Google, you know, it's funny with naming their things because the the first thing I ask when someone says, hey. We wanna use Gemini. I say which one because there's many, many ways Mhmm. At Google, right, to use Gemini if it's on AI Studio, Vertex AI, or consumer app or Gemini Enterprise. Right? There's so many entry points that it's it's really important to understand when someone says I wanna use AI. It's like, really understand what do they define. Right? Mhmm. Because I feel before in the past, the definition of implementing AI post, let's say, 2022 and and previous, implementing AI meant I have a research team. They're gonna training a model, and then they're gonna do inferencing in house. Right? Yeah. That AI could mean that plus so many other things. Right? Is it a productivity tool? Are you guys launching a a feature that involves AI? Is it an internal tool? And I I think, you know, going back to your question of of how we pitch it and how we sell it, it's really gonna be on on the use case. Right? Mhmm. I think the one thing that's really interesting about generative AI specifically is that it pushes the the feature or the product so close to the user experience that that's why I think everyone's excited. Right? I think it ties very, very well with the KPIs that certain companies, you know, working on. Right? With AI, you could literally see, you know, some people are mad and they're cursing at the AI agent that you made. You're like, okay. They probably aren't enjoying this experience. Right? So I think that is one big thing that has changed is that now developing in today's age. Right? It's so close to your KPIs. It's so close to your user experience. And I think that's a big reason why there's so much more adoption. Right? Because that feedback loop between what users care about and are willing to pay for is getting smaller and smaller.

I just wanna quickly touch on a message that you sent me a week ago Mhmm. Where you mentioned year by year, basically, the tech evolution. And you said, I think 2023, we had AI wrappers, then we had tools, then in 2025, we had multimodal, and in 2026 is the year agents. So can you maybe just, again, explain each of these terms and why today with agents, we can act improve those KPIs at the business level?

If I had to summarize what was happening in in generative AI with just two words for each year. 2023 was the year of the AI wrapper. Essentially, that's you're you have a foundational model, like, you know, OpenAI's ChatGPT was common back then. And you're essentially just giving it a prompt. You're changing the way that the response is behaving, and that was sort of it. That's sort of where it ended, and then you resurface that to your user. Right? And I'm not gonna be one to say that AI rappers are a bad thing. Right? If the market wants to pay for them, great. Right? But in 2022, really, it was the year of the AI rappers because everything was brand new. No one knew what was happening. Anything you created was novel. Right? So 2023, the year of the AI wrapper. In 2024, what we saw now was something called function calling. Essentially, cool. You made an AI wrapper, but now you wanted to access data sources outside of the knowledge cutoff of these foundational models. Right? Because all these foundational models, they only know what they're trained on. Right? And if they were trained, let's say, up to the point of 01/01/2026, they're not gonna be able to give you information about 01/02/2026, right, unless it does a tool call to maybe a search engine. Right? Mhmm. So 2024, that was a big, big part of what the tech industry was trying to figure out. Things like retrieval, augmented generation, or RAG, right, so that you could give it your documents or function calling where you could maybe call a weather API, maps API, or search API so that your user or or so that your use case, right, has a more fulfilled, like, experience for the end user. Mhmm. Then 2025, came around, and that's when we sort of saw multimodal. Right? So now the first two years, 2023, 2024, was really pretty much text focused. Right? But we know that AI could now do really well with vision. Right? Nano Banana from Google DeepMind is a good example, or video generation from, like, OpenAI Sora or Google's VO and all these different you know, Google even has some music ones as well, and they recently launched a new one that's, like, to generate worlds. And that's what we call as multimodal. Right? It's not just text based. It's vision. Right? It's audio as well. I don't know if Google will launch something in the smell category, but, you know, that's what I sort of see 2025 is, like, multimodal. Now Mhmm. 2026, right, is now multi agent. Again, right, I challenge myself to define each year with just two words. Mhmm. In a way that I see 2026 is really gonna double down on agents working more autonomous. Right? Not having that limit of it's not gonna do anything unless it's chatting with a human like a chatbot, right, for a consumer to use, but it still provides value to them in the background.

Coming back to your role and the companies that you are advising on agents pretty much today, Do you see any specific pattern, you know, across the 400 people that you talk to, maybe not on a daily basis, but but in terms of, like, use cases for agents and just GenAI that, yeah, people are super excited about?

Yeah. I think I think now what what you are seeing is companies doubling down on on solving, like they they it's very centered around, in my opinion, very tangible problems. Right? And now they're centered around specific industries more than ever or even verticals. So we all know, you know, in The United States, there's a lot of paperwork. There's a lot of processing needed in in the legal industry as well as in the health care industry. Right? Mhmm. So you are seeing this strong sort of push of automating work that we know really well how to do at the human level, but it scales by how many humans you throw at the problem Mhmm. And replacing that, right, or providing tools for those people using AI agents. Right? Yeah. And I think that is something that's happening, and we're gonna see a lot more is is really doubling down on use cases that are almost in the background, right Mhmm. Of setting up these automations. Right? You could call it AI automation. You could call it an agent. But I think, again, it's very, very close to a very tangible, painful experience that, like, you know, a human has to do, but it just can't scale. Right? Because maybe of the domain knowledge that's needed or just physically, you know, human crawling process so many things. So, initially, we're talking about AI just

for what it is, just like a new technology, like kind of a feature. And now we are slowly moving towards business metrics. And within KPIs, we talk about OKRs. Mhmm. And let's say you want to increase traffic by 10% and you need to post 10 articles per week. This is your OKR to achieve a 10% increase in your traffic. So you're saying that, essentially, the agents, you're looking at the OKR of publishing 10 articles per week, which would be automated, and then eventually, we can just stack up those agents across each OKRs to then eventually be able to steer and and and decide how far we wanna go with that KPI. That's way to look at it. Yeah. Yeah. Exactly. And and it's you know, content creation is for sure, I think, the first one that you're seeing. Right? Everyone's sort of complaining of AI slob, but, you know, content creation is there. 90%

of the things you do with a computer, you could automate. You just have to make that decision if it's worth automating. We just touched on a on a bunch of super relevant

points around managing expectations with your clients on what can you do with AI, with off the shelf AI versus what you need to build and customize. But so how do you actually you know, when you meet, like, CEO and CTO who's telling you, okay. We wanna build this. How do you manage expectations of the complexity, and how early can you tell whether you're gonna be able to do it with off the shelf tools versus what you're gonna actually have to, like like, develop, like, fine tune models for or just more custom configurations?

That's a that's a good question. You know, it it always depends on on the use case, and I think the steps you wanna take are the steps that are low hanging fruit and not expensive. Right? The easiest way is just to start playing with one of the know, if if it's one of my customers, obviously, gonna be a Gemini model or or a model from DeepMind, but cool. So so we define the use case, what it is. The first thing I wanna understand is how would you guys evaluate the system, whatever it is. But knowing how you're gonna evaluate at scale is really important. If the evaluation currently is, oh, we have this one person that has this domain understanding and they have to see, right, the answer and digest it, that's not gonna scale well. So your problem now is build the AI system plus replicate your evaluation person. Mhmm. Right? Let's say it's you, Vlad, right, that you're really good at understanding and seeing, you know, if if there's an issue with with whatever is being processed. Right? Mhmm. Sitting down and actually documenting, like, your domain experience and teaching the Nawa LM by teaching, I mean, you sort of build the right constraints so that they behave in the way that you do. You know, that's a time consuming thing. Right? But it's very important regardless of the solution you create because if you're not thinking about how you're gonna evaluate the success, that's when you fall into many issues. Because the issue with building with LOMs that's very unique with any other way of building is that the likelihood or the different avenues that an LLM could take are essentially nearly infinite for our brains. Mhmm. Right? But we've all seen LLMs hallucinate. We've all seen how maybe users push things to a limit. So I think especially when you have something that's very important, like in medicine, in in health, in in law, things like that, you have to think evaluations first. Financial market, push the limits of whatever you produce. Right? You could think of a thousand ways that a user is gonna interact with your system.

Mhmm.

But in reality, there's millions, if not, you know, many millions of ways that a user could actually interact, and you can't find all those constraints. Like, you just can't humanly possible think of all the edge cases. So you really want to provide that structure. Right? Mhmm. So I think, you know, at the end, you want to think about how you're gonna test any AI system first before you actually dive in if you're, you know, CFO, CEO, CTO, anything like that. Because if not, it just gets wild and gets very crazy very quickly. Yeah. You know, and I think that's also when a lot of people quit these projects. I know there was a article somewhere out there that said that 80% a high number of percent of AI projects fail, which I think is true, but I think also it's so easy to create an AI project. So I guess it depends how you define an AI project. Right? Yeah.

It's funny that you are raising the stat because I think the stat is from the MIT report saying that the like, 95% of the projects fail. Mhmm. But, actually, the author of this research just released a statement a couple of days ago saying that his words were completely transformed by the media and that basically completely misinterpreted his his analysis. And so it's crazy how this is those are stats that everyone had in you know, this is a narrative already. It's too late to, like, change people's opinion on it. It's been a I think the paper has been around for almost a year, and and it's completely I mean, it's false. It is factually false or at least this the the research paper isn't, like, scientifically sound enough to to to determine

that the Yeah. Are are right. Yeah. Because I I mean, I do see you know, I think what's interesting too is that there there's a lot of internal use case. Right? Use cases for leveraging AI, you know, at the at the founder level or the c suite level. And, you know, I remember I was in a conversation with with a pretty large bank, and what the CFO said there was, you know, the appetite to use AI for their day to day job is is so large and not just for, like, launching a new product or or shifting the company's trajectory, just literally having a tool to make a meeting more fun. Right? Because now you could do that. Like, now you could quickly build a vibe quoted little app, right Yeah. That is maybe just for your core team or just for a presentation. Like, that's so crazy to think, right, that a CFO of a large bank knows that they could vibe quote a tool for an internal meeting. Yeah. Right? And it's not like, five years ago, you would never think the CFO of a bank is gonna spend the time to build a tool just to make it more fun and engaging for his team. Right? That's Yeah. That seemed like a waste of time. Right? Because that would take probably a week. But now you could build that, you know, on the side as you're working. And there's a lot of these, like, few little moments that you could use to just make the work environment either more enjoyable or just run smoother. Right? That's a very low lift in today's age. Interesting.

I think the the distinction is, you know, all these internal quick productivity tools that people develop, and then you have, like, those most strategic AI projects that people need to implement at the at the company level. Mhmm. But I think both, like, both work streams, there there there's so much that can happen. There's so much that can be done that you need to prioritize somehow because you can only get so many resources to things. And even if it's fast, people are good. You still have to prioritize, and you have to have accountability. Too crazy. So, usually, when you, you know, when you work with your clients, how do you actually think about those road maps about, you know, what to build at first? Every company

is is different, but more or less, you know, you could think of it as stage one, you try to understand the problem. Right? Stage two, maybe you try to present a potential solution or an idea of a solution. Right? Stage three, you try to show them how to implement. And stage four, you decide if it worked out or not. Yeah. Now that's typical sales, like, sequential. In reality, right, it's more like you go stage two and then back to one and then two and then three and, you know, in real life, it's a little bit more dynamic than that. And the reason I I say that is because I think depending on the stage, right, you're trying to fully just understand the problem first because you want to understand what moves the needle. One of the things that startups really care about is getting the next round of funding. Right? So I try to understand in your journey to the next round of funding, what are the biggest blockers, or what are the things that you're going to launch with. Right? A lot of times, their previous funding will say in a news report or something that, oh, company x got this round of funding to scale out their sales team. I'm like, cool. Okay. So they probably need to scale out their sales team to get the next round of funding, then whatever AI initiative, right, has to be centered around that. Right? I could show them maybe cool things that are fun and maybe a productivity boost, but if we can't prove that that productivity boost, you know, isn't gonna have a meaningful impact to that end state for, in this case, let's say, getting the next round of funding, you know, and showing that their sales team is is is doing really strong and it's working, then it's not gonna really benefit them. Right? And maybe, you know, part of this initiative almost always that startups will do is insert the startup name, and then they'll add the term cloud. Right? So a good example is, like, Langchain. They they launched this product called LangGraph, and they made LangGraph Cloud. Not Elon Musk's, but with a q at the end, inference as a as a service. They created a cloud as well. Also, lovable.dev, they created lovable cloud. Right? So it's very common for a lot of startups to create this, like, managed offering. Mhmm. And almost always, that's around series a. So I sort of focus on on that type of initiatives. Right? Like, okay. This is meaningful to you. Mhmm. And let let's focus on this one thing. Because, again, for startups, it's gonna be almost always getting that next round of funding.

Yeah. And do you think that for the companies that are profitable and are not necessarily, like, looking for funding

Mhmm.

What would be, do you think, the the mental exercise that people would have to go to? They go from the board meeting and Yeah. What KPI they have in mind, and then you would translate that into specific initiatives and discard the rest, or should you try to find overlaps? How do you because, like you said, it's a lot more dynamic than just figuring out one problem.

Yeah. I think what you know, maybe even startups included, but every company now is dealing with is they know they could build a lot more things. Right? They see a lot more potential options that they could do, but there is this sort of hesitation of are we on the right path? Right? Are we focusing on the thing that provides the most ROI? And I think that does come in in two things as one culturally. I think a lot of companies have to sort of really sit down and think is what is our culture when it comes to leveraging AI tools and building AI tools? Because a big part of it, sadly, to this day is self education or staying up to date. Right? Because products that were made in 2023, that's just three years, you know, four years ago. Mhmm. Today, they you'll you'll laugh at them. Right? And, also, you don't wanna build something that in a few years sort of becomes irrelevant. Right? So I think a big part of today is you do have to adopt a mindset of some time has to be allocated and just staying up to date and understanding what's happening in your domain, right, specifically and the things that matters. Right? If you don't care about training and inferencing, don't focus on that. But in your domain, in your expertise, in your world, try to always keep a foot, like, in the other room of where AI is going because I've seen it so many times that companies will invest a lot of effort in building a product or a feature. And if they just waited six months later, it would have been a lot easier. Right? Mhmm. So I think you have to just adopt the AI. You have to culturally change a little bit because developing with AI is different. It's it's very unique because it it truly advances at such a quick scale that you wanna just make sure that you guys aren't diving into the deep end with your eyes closed, so to speak, and you sort of miss out on other opportunities or other things. Also, I think when it comes to figuring out, like, what to focus on, I think the the thing is if you're new into leveraging AI or building AI products or or leveraging LLMs in general, always focus on the low hanging fruit. Like, don't aim for something very ambitious because it it it culturally does change. Right? When if you were use of developing or using tools that are very predictable, You know, welcome to LLMs. They are not. And that changes behavior, right, in in a company. You have to realize that a lot of it is gonna be creating that constraint collectively as a company, you know, your product managers, your engineers, everyone sort of has to work together to make sure that you know how to create that constraint. Right? Because if not, the LLM's gonna just sort of hallucinate, then you're gonna sort of feel like it was a waste of time, and then you're gonna feel that you're sort of missing out even more. Right? Because you're like, oh, my team can't figure this out. So I think culturally, it's more than anything. It's gonna be this cultural shift of being confident that your team could sort of educate themselves. Mhmm. Two, that you guys could collectively create sort of constraints around an LLM. And three is, like, how do you go about if the LLM does have issues and does hallucinate that maybe, historically, you just didn't deal with because all the code you wrote, right, input and the output, you guys knew about.

Now you may not know the input, and now you may not know the output. So you really wanna constraint both sides, right, of the Do you see, like, specific patterns or, like, structures that seem to work for fast decision making and companies quickly decide, okay. We need to make these bets this year because this is where the the AI is heading. How do you actually build something and be in line with, you know, the not falling behind, but not also working on something that is technically still not possible? So

the the true answer is it depends, but I'll give you maybe a mental framework that I've seen Mhmm. Sort of help. So so I think there's two things when it comes to making a prediction in anything. In in AI specifically is accuracy and then precision. So an accurate like, accuracy could be that eventually inferencing is gonna be more popular than training. Right? I think we all know that's an accurate statement. Precision would say would be saying, I think in the year twenty twenty seven, q four, inferencing is gonna be more important than training. Right? Because I'm being very precise with my with my hypothesis there. And the way that I see it is you as a company need to think of a hypothesis of whatever your domain is. Mhmm. And, you know, obvious more than likely, it's easy to be accurate. It's hard to be precise. But I feel the stronger you are in a domain, the more forgiving it is to be precise. Right? So the stronger you are in that market in understanding the pain points your customers are truly having and understanding what your competitors are doing, right, the more forgiving it is to be precise. It's

it's funny you say that because I immediately, when you when you mentioned the, you know, being correct about the direction but not being very precise, I immediately thought about DeepMind Yeah. Where, essentially, you guys were very much ahead of everyone else Yeah. Developing those LLMs. And then there was this one year where you were maybe falling a bit behind and underappreciated possibly the extent to which consumers would have access to these tools and the adoption levels that would happen. Now you caught up. So you do have this thing that happened actually Yeah. Where it's directionally

right, but about the time Yeah. And that's that's a really good example too because, you know, Google DeepMind has been was acquired by Alphabet or Google in 2014. Right? 2017, the famous paper came out. AlphaGo was a thing as well. You know, Google, I think Google DeepMind was accurate that the world needed AI. Yeah. They had so much domain experience that they could be wrong on the precision of when Yeah. To go to market. Right? Because I think 2023 before Gemini was barred. Mhmm. 2024, you had Gemini 1.5. At the end of 2024, you had Gemini two. And what's crazy is to see their growth in 2025. Right? You had Gemini three, Gemini 2.5, you had v o two, you had nano, nano one and two. All these models, you know oh, and also audio models and so much that they launched. But, yeah, I think Google was able to maybe not be precise, right, but still benefit because they, you know, are a leader in in AI domain specifically.

Yeah. So a a living example of your mental framework,

which Yeah. Turned out to be good. Yeah. And and I know you asked the second question. What what was the follow-up?

The second question was about how can you make sure that you are directionally correct? So there is the thing about the business decision. Yeah. But then you also mentioned, like, the tech evolves so fast, and it's probably the worst nightmare for some c a CEO or, like, a CMO to, like, unlock a budget to invest in a specific thing, but then realize that in two years' time, basically, it's gonna become irrelevant.

Maybe another mental framework that could work is, you know, we all know, like, one thing that tech founders preach, especially y y Combinator and accelerator, is shorten that feedback loop with your customer. Mhmm. Right? And I'll take it a step further. I think for companies, it's two ways that you want to shorten the feedback loop, obviously, with the product. Right? So if you are looking at the AI space of we're gonna design and launch a product in a six six month window Mhmm. That's not a bad way, but I would urge and and see, can you get those insights from the users sooner? Can you launch something with smaller amount of features sooner so that you get that feedback? And then second that I think a lot of founders, but I think specifically founders, but I think also companies fell is getting feedback on your traction. Right? Mhmm. You could call this marketing. You could call this sales. But the way that you guys are actually getting traction and interest, that feedback has to get smaller as well. Mhmm. Right? Mhmm. So I I try to recommend that you want both sides.

And if so, if you apply so this is the mental model to, like, distributing product. But if you apply this model to internal productivity operations, you know, where if we take, you know, your example of the sales team of earlier where it raises series a, they have $25,000,000 to deploy, and 80% of that is gonna go towards sales initiatives. Yeah. How do you usually go about this?

You know, if it's internal, hopefully, the the end user is very vocal in what they like and don't like. I find employees tend to be very vocal. Mhmm. Change management isn't easy. Right? Everyone may have a way of doing a tool, making a tool, or, I mean, using a tool specifically. And I think what's worked is really having for internal use cases, even if they're not the ones building it, being part of the idea and scope. Right? Because everyone sort of I think of it as the IKEA model. You know how IKEA benefits a lot from you feeling like you built the furniture. Right? So I think, again, because it it's so novel that the tool you're building is so close to the user experience. Right? It it I can't stress that enough. It's 10 times more closer to the user experience than other ways of building that you wanna take advantage of that insight. Mhmm. Right? Your end user could probably read the code a lot more than they could have five, ten years ago. Right? A big part of building with LLMs is that the system prompt, right, or the instructions that you give it, and it's just a block of English. Right? Having actually them be part of that discussion, I think, helps so much in finding what is that low hanging fruit. And I think when you're first starting out, you just wanna focus on that muscle of building with AI more than the end result because sometimes, you know, you focus so much on the end result, but learning how to build with AI should be an OKR in my opinion, for most companies because it's it's new, it's novel, and it's changing so much that the fact that you guys even implemented something and push it out there, it should be applauded, especially if it's your first one. Right? Because there's even with me, there's a lot of startups that they aren't an AI company. Mhmm. But for them, you know, jumping into agents, the fact that they even were able to launch something, they've learned so much, right, as a company. So I think that's also something to consider is that it is novel. It is different, and it's something that you should applaud internally and then take the learnings. Right? Because we've all if you're a developer that's been working with AI code bases or or coding with, like, Clog code or or Vibe code in Unlovable, the first two, three, four, five projects you built probably aren't great. But then you learn how to work with these. Right? Yeah. Just how everyone sort of learned how to communicate with ChatGBT, right, in 2023. As a company, you have to learn how to build with these tools. And it's not something that happens overnight, but it's more of a cultural shift. And once it does, right, it's a lot more approachable, and it's not so intimidating to just try these new projects or initiatives that you have.

Yeah. And so so this is more related to if we bring this back to, like, the prediction of, like, having having people to, like, dedicate some of their time to, like, building out, essentially, you're suggesting is that these companies just build up the muscle internally to, like, build out, reduce that feedback loop so that as soon as they see that they are going, not necessarily off tangent, but that whatever they have built does not really Mhmm. Work or bring any KPI uplift, they should just iterate, take a step back, steer in a different direction.

Exactly. You know, you you build a harness for the LLM. Right? But I think what sometimes happens too is you don't build a harness for your company. Mhmm. Right? And you want that same structure. Right? A lot of the great vibe coders that I know is they'll spend so much more time on planning than actually vibe coding. Right? They'll consolidate, okay. What do I wanna build? What's the purpose? How do I evaluate it? What are some good samples that I could borrow? And then they'll attack the problem. Right? Mhmm. And they doing that process constraints what the LLM is gonna produce. Yeah. And similar to to your team, right, is you want to surround them with the right resources, surround them with all the information they need so that, you know, it doesn't go too much on a tangent, but then monitor, right, the success of Yeah. Either building traction or the feedback from the product itself.

So so if we take us, like, an example of a company Mhmm. You know, like a b to b company trying to encourage people internally to adopt AI tools, what do you think are, you know, the top three resources they should have access to that, you know, their bosses should unlock for them to play around and to really build out their own productivity tools so that they can Yeah. Work better. You

know, one thing that that I don't know if there's a so so the the two tools that probably make the most sense right away is some type of if your company is big enough, some way of searching internal tribal knowledge quickly. Right? So if you have usually, once a team's bigger than 50, maybe 70, there's a lot of documents internally, right, that are very hard to find, that sometimes aren't shared around. So I think an internal search is something to really consider, right, especially as you grow. Right? I there's sort of a magic number after a 100. Once you double to 200, the company is too big for, like, everyone to know each other and just message each other. Right? Mhmm. So having and there's many tools out there like Lean, you know, Gemini Enterprise as well, there's probably more. But something that helps your team quickly find and identify all of the work that they're doing is always beneficial. Second is gonna be obviously some type of copilot or or tool like ChatGPT or Gemini is, you know, so many I I think it's it's now like the spreadsheet of today. Right? It's it's like to not give your team OfficeSuite, you know, a word processor or an Excel or, like, Google Workspace. It seems like you're making it much more difficult for your team for for not a good reason. Right? And, obviously, I know security and and all that stuff is important, so I know a lot of these tools now, you know, are are much more friendly with customers' data. But I think having a tool like Jaiji BT or Gemini, like, in there so that people could just articulate, explain, maybe, you know, generate graphs, all these things, those two are if you're now becoming commonplace. The new one that that I haven't found a solution yet, and maybe there's a company out there that that does, but a way for your internal teams to build apps and share it internally only and secure. I think that's some use cases that I'm starting to hear, like, both companies big and small is, hey. I build this tool for my team. I wanna share it out. What's the best way to do that? Right? And the Internet or at least right now, a lot of tech mainly focus on Internet, not intranet. Right? So I think there is a strong appetite of enabling your your team to actually build apps or little tools that maybe go out of the scope because, you know, I think most companies build these tools within Excel or Google Sheet. Right? But now you could generate a lot more interesting, useful applications. And, again, a lot of times these come not from your engineers. They come from your, you know, sales team, from your marketing team. Right? Mhmm. I think empowering those type of people is gonna be more commonplace this year. Yeah. One, because and and they want to. You know, the appetite of nontechnical people to build tools is going up for sure. And I find that exciting, but I think those are sort of the three tools I would really look into. Some type of enterprise search, some type of, you know, Gemini or ChatGPT solution so that you could just talk with. And then third is enabling your team to build and share tools that make their lives better with the rest of the the team or organization.

Mhmm. Yeah. That's I I know that Palantir, they have a tool called Foundry Mhmm. Where it's very powerful because you can connect a bunch of that data sources, like Yeah. Secure, obviously, like pipelines and so on. And then you have this feature module called workshop, which is a no code built, like a no code platform, which you can build applications that you can then, like, share around people, and you have all the user permissions set up. But I still find it not probably suitable for nontechnical people because Yeah. You just have to, like, pick the components that you wanna add it. So it's basically like Webflow, essentially, where you just decide on the padding and the components and how to displace certain things. So you still need, like still, like, the old way, you know, even though it's powerful. Yeah. And I saying it should be prompting, basically. People are used to prompt now. Yeah. And I think also the like, for internal use, right, people don't really care if it looks nice. It's very focused on utility, which I think is different from maybe the use cases of Lovable Webflow.

You know, those are very focused on stunning, like, user experiences. Right? Yeah. So I think that that is a shift, and I think, you know, I think empowering your team at the end of the day to be able to build in sort of the AI era that we're in. Right? Yeah. It it it's it's a big win. Because I've been very surprised with some of you know, my my my team, the way we work is we have a customer engineer, the sales engineer, and then a sales account manager. Right? I'm the technical side, and they're more of the business. But I'm always surprised in what our business side of the team they'll build. Right? And I'm like, oh, that's actually very useful, or, oh, I didn't know you had that pain point that you know, just I just had no clue. Right? So it is, I think, a a shift we're seeing internally at Google too. But, also, I think, you know, many companies that I've worked with, especially the larger ones, you know, 500 plus, 200 plus people, they're seeing that too. It's like they're getting really cool novel ideas to make their business better from the nontechnical side, right, the business analysts

who's who just have a different perspective in what the business needs. So now we've talked about what companies should have, like, kind of resources. Yeah. What do you think is a distraction today? Like, do you see companies just wasting too much time on certain topics, certain tools that are maybe not robust enough to work in production? Or Yeah. I think well, one, think a big

problem too is analysis paralysis. Right? Mhmm. Because I I've seen it so much is is there'll be a big effort in understanding everything, and nothing happens. Right? Maybe they don't launch the product. They don't start that internal initiative. You know, they just sort of pause, and then they wait and wait. When in that time, right, you could have just done a thing and measured. And, again, it's it's back to that cultural thing of just create that idea and bring it to life internally or if it's a new product as quick as possible and measure, right, the result right away. I think that's the one biggest thing I see is is analysis paralysis for sure. Right? Startups and enterprises. It it happens everywhere even even with me. I think it's just human nature. Right? In terms of analysis paralysis,

I see this a lot also with my clients. They are very excited about building stuff, and then nothing happens Mhmm. Because they wanna go from zero to a 100 immediately. What do you think are, like, reasonable timelines or expectations to get things moving? You know, like, the very first project, do you want something out in two weeks, three weeks, two months? I mean How do you go about this?

You know, the way that I like to do it is I'll open either Vertex AI or AI Studio, and Mhmm. I'll challenge the customer. Like, if you have access to your data, right, or whatever the use case is, let's see what we could do in thirty minutes. Mhmm. Not that this is a product that's gonna be in production. Right? Thirty minutes is not enough. You you you you know this. Right? Like, five hours. What you guys built in five hours during the AI agent bake off was impressive. Right? But five hours is not enough for a full production Yeah. You know, AI system. But I think thirty minutes alone of just focused time, especially if you're in this analysis paralysis, could sort of be a very good exploratory tool to just see where is the limit of your use case. Mhmm. Right? Because, again, it it sometimes it's just you have to go into it, and even this thirty minute exercise a lot of times unblocks a lot of the anxiety of getting started. Right? Yeah. It's like it reminds me of of people that run marathons and stuff like that. Right? It it's really more of what you put in daily, not doing, like, one, like, doing one marathon, you know, like, running just the marathon out of nowhere. Right? Like, you have to really prepare. You run much more miles prior marathon, right, than you do at the actual marathon. I think for that is is spending thirty minutes and playing with it if you haven't, I think opens up the the art of the possible very quickly. Right? Mhmm. And then you could quickly see, oh, you know, some of these models allow text input and image input. What is our use case if we just provide images? Right? Because a lot of times you you'll be surprised even startup founders, right, who, you know, sort of are known to be sort of the tech leading will ask, hey. Can for my type of documents, is Gemini good enough at processing it? And I'm like, I I don't know. I I don't know how your documents look, but let's try it Right? Yeah. And see just quickly with a very simple prompt, nothing crazy, how far do we get. Right? And, usually, that's a good place to know just, like, one shot prompt. Where are you? Right? Like, that should be something everyone in the org should feel comfortable in doing because, again, we're talking about a ten minute, thirty minute exercise at most that at least gives you a really good idea of where your application could start. Right? Because that's that's the worst your product could ever be. And if that excites you, great. Then build something robust around it. If you're like, it's not what I was hoping for. Right? Then that's when you have to probably do a little bit more homework and see where are the limitations. Is it what you're asking, or is it the specific model or the framework that's being used? Mhmm.

So what are, like, the best practices since you're using these tools every day that you have seen? And, yeah, what's the usual process you follow to to build something? You know, I I I know that, you know, I work at Google, people probably think, oh, he has a very sophisticated way

of using AI. And it's not just, you know, Chatty Beat. I mean, not just Gemini. You could use it with pretty much anything. But my flow, whatever idea I have, is always I'll start and try to write as long as a prompt as I can. Many times what I'll do, is I'll record my voice as a sort of brain dump of, hey, Gemini. I'm trying to build this tool for my sales team. Here are some of the problems I'm having, and here's what I want. And then what I'll follow-up with is ask as many questions as you can until you feel confident in what I want to implement. Right? So, really, is is that first initial prompt, you know, Gemini will reply, and then it'll usually ask me five to 10 questions. I'll answer each of them again as thoroughly. I like to use my voice because I speak a lot quicker than I type, but you could just type. And then just doing this little exercise, you know, less than three minutes long, it helps me sort of, like, put down what in my brain I want to do, but in a much more verbose thing. Right? When when working with any LLM, the more you give it in context, just the better it behaves, especially, you know, in in the way that I'm using it. And then by that time, I tell it, okay. Cool. I wanna build this tool. Please devise a game plan for me that I could follow step by step. The reason why I do that is because if I just tell it to do everything for me, sometimes it hallucinates, sometimes things break. But at least if I follow it step by step, right, I know what are the things that it could go wrong with. Right? And sometimes I'll ask it as well in each step to list out what are the biggest problems that I could face when doing, you know, the the the issue. So one example is I built a sales tool when NanoBanana came out, where you would give it a domain, and the output would be a, the company's domain made into a GIF with Nano Banana and VO. Right? So it would essentially be like a marketing tool, like, hey. Do you know that Nano Banana's out? Here's a little GIF of your website with little robots crawling all around. All that was vibe coded. Right? And it was just me following this process of, hey. Here's my idea. Here's what I want. Here's the inputs. And telling it that, like, what are the problems that I could face in each step helped me realize because authentication is always a problem. Storing databases is always a problem, right, as we know. But it just told me, hey. Heads up if you have these problems. And I think that's a very easy, simple way for me at least to build these tools for my team internally that doesn't cause as many hallucinations. And, again, it's it's not a very intense intense way way of of prompting. Prompting. Right? It's not you know, it's probably five chats with Gemini, and then I start building the theme with Gemini CLI specifically.

Yeah. And then in terms of the evaluation, coming back to how people usually evaluate, the feedback loop is just simply prompting. This is how you evaluate. Yeah. And then and the feedback loop is, do my sales

does my sales team use it? Right? Did it actually get a response? Right? I think, you know, for me, that that's when I realized, oh, yeah. Building internally is really useful. Right? And, yeah, it's not a my team's a 100 people, and let's say, you know, 20 people use it, I think, or 30. That's still a very significant, you know, number that that now these 30 sellers felt empowered that they could showcase this, right, and sort of be proud, right, of, like, oh, this is cool. Right? Like, this this is something that maybe you could use in your business.

Yeah. That's for sure. I mean, if you have 400 customers, yeah, and you have 20 people who are using it Mhmm. Probably also all have 400 customers. Yeah. Yeah. And, also, the benefit is quite a

Yeah. And another benefit was these 30 sellers that used it. Right? They didn't have to build anything because I built the tool for them. So it's it's a good way to scale out an idea and test it. You know? I've had many that I've built and no one wants to use them too. Right? So they're not all successes. But again, I'm not afraid to do that. Right? And that sort of happened to me as an moment last year of, oh, yeah. If if you just build that muscle the first time, building the next little idea, right, and now you know how to measure it, it's not as intimidating. Right? So now it's, oh, okay. If if I know 20% of my team likes it, then it's maybe something we should scale out to even more teams. Yeah. If no one likes it, then I'll just pretend I didn't make it.

I will keep a close eye on your on your repo to see what that should. Just to to to finish off, maybe just one last note. By taking a little step back Yeah. Before Google, you worked at Accenture, Deloitte. You worked probably with, like, more traditional companies. Yes. Knowing what you know with all these startups and how fast they go and about all the resources and tips that we've just talked about, what are what what would you suggest, you know, everyday companies to copy or or implement?

You know, I think one understand that a lot of these startup founders, they tend to see developing more of a experiment, right Mhmm. Of exploratory and and trying to understand. Because through the experiment alone, you'll learn a lot. Right? So it's not as intimidating as maybe knowing, like, we're gonna we have to make or hit this KPI. Right? I I do think you know, and I know startups are different because they're they're starting from zero, right, and going to one is a lot different from going to, like, you know, 99 to to a 100. But I think that way of seeing AI development is very beneficial for organizations. It's it's an experiment. Right? And just doing the experiment alone is, in my opinion, should be an okay or should be, you know, a thing that you guys implement because there's really no better way to learn what can be done, how AI could benefit your company than just doing it, and it's such a low barrier. And I think it's just dedicating making that space available. Right? Mhmm. You know, the whole, like, fell fast, fell often thing, like, it's not so much of that. It's more of culturally, you don't want your company to be scared of AI. Right? It's a tool. And many people don't know what they're doing, but they're willing to experiment with it, right, and see where it it it falls behind. Right? Like, for example, what I said is if you have your own documents and you just wanna see how AI can handle processing them, this is an you know, this is a five minute exercise. Maybe it's something you do every two months. Right? And you just keep that in mind that, oh, every two months, I wanna see how how far it's getting better until maybe it hits a point of, oh, great. This is perfect. Let's actually use this as a product or a feature. Right? And I I think it's just much more approachable than people think, and it's getting much more approachable.

Yeah. Awesome. Well, thanks thanks for for for your time. Yeah. Do you have maybe some place where people can find you? I know that you have your podcast. Now is the time to to to promote your podcast, the channel, everything that you're interested in. Like, yeah, where where where could we Yeah. Yeah. Yeah. LinkedIn

is fine. You could find me at Abraham Gomez. If not, if you wanna see what I'm doing with my silly little podcast, it's called Who Invited Abe, and I'm the host, Abe. Mhmm. You know, just whoinvitedabe.com works.

Awesome. Alright. Well, thanks thanks so much for your time, and we'll be in touch on that because I have your podcast there. No. I appreciate it. Take care, Vlad. Nice chat. Have a good breakfast.

Thank you. Bye.