The Data Hustle Transcript

Slow Data, Fast Business — Noel Gomez

Episode 8 · Jun 23, 2026

Duration: 50:22

Source subtitle file: 2026-06-23__F5wvR_Rnu0s__Slow_Data_Fast_business_-_Ep8._Noel_Gomez.en.srt

Generated from exported YouTube subtitles. Speaker labels are approximate.

[0:00] Tony and Juan
Today I'm joined by Noel Gomez, co-founder of Data Copes, where he helps data teams put DBT and Apache Airflow into production without spending months building and stitching together their own platforms from scratch. Noel has spent more than 15 years working on data platforms. And before data co, he worked in enterprise data, including at Amjen. What makes him especially interesting for the show is that he sits right at the intersection of what we care about, modern data tooling, real production constraints, and a mess reality of helping teams actually deliver value of data. Noel, thank you for joining us here on the podcast. So, before we dive into the questions, can you tell us something about yourself and um yeah, your background and how you got here

[0:40] Tony and Juan
>> for inviting me? Um, happy to join the call here today, the the podcast and um how I got here. Well, it's it's one of those strange uh things I would say that um well, so I I went to school to to be an industrial engineer. Um, and and so a lot of the stuff where you learn about lean manufacturing, Toyota production systems, some of those things still carry.

[0:40] Tony and Juan
I've always been a person who looks for like how do you do things right the first time?

[0:40] Noel Gomez
How do you avoid uh inefficiency and waste and all of that kind of stuff? And so from from you know like that that manufacturing background eventually got and worked on some software uh things where I was

[1:25] Noel Gomez
using Ruby and Rails uh the web framework and uh that taught me a lot. This was you know I don't know over 10 years ago and I learned a lot about having conventions over configuration where you can essentially look at a project uh any Ruby and Rails project and it will be very similar and uh when you're first starting out you're just following the rules you don't necessarily know why you just say okay somebody told me to put this here and that's where I put this stuff um and and eventually I appreciated that I appreciated that, you know, you you're teaching uh good software development best uh software development best practices without actually knowing that that's what you're doing, right? It it was set

[2:13] Noel Gomez
up by somebody who had more experience than you who who did all of that stuff. Um and again, then you appreciate it, but not in the very beginning. And so then when uh I was I was at a large enterprise and we were modernizing um this is around the 2017 uh time frame, I was uh wanting to do things I with some of those same principles that that we had from software engineering. So like um convention over configuration and and you know unit testing and data testing and all of these kinds of principles that version control all of that kind of stuff. Um and we were using a guey tool a traditional enterprise guey tool and we were moving more to like airflow and

[3:00] Noel Gomez
we ended up creating our own custom framework. So, it's essentially like saying we were making our own DBT. Um, but it didn't live up to what my ideal way of doing this stuff was. And so, uh, we eventually developed this thing, deployed it, and it ended up being better than what we had previously, but not meeting my expectations of what could look like. So fast forward to 2019 uh left the company started uh this company and we started with a different product initially um and it's very similar to what's happening today because we were doing uh this is around 2019 we were doing a product uh like a voice assistant for sales reps uh in pharma and I went into

[3:50] Noel Gomez
it with the idea that like oh you know everybody has good data and we just have to put this at the end and everything will work out right which is similar what to what we're seeing today with with Genai where you don't have the governance, you don't have the ownership, you don't have a lot of those uh processes in place, but I took that for granted back then and uh what we were seeing is that that you needed that first part, you needed the governance, you needed the the the ways of working and all of that and um you were doing consulting and helping people set up CI/CD and all this stuff and and that's when I essentially like ran into or

[4:26] Noel Gomez
learned about DBT and uh and started using Snowflake. I knew about Snowflake but hadn't used it. And then it all clicked for me cuz I'm like, "Oh, all the things that I had wanted to do, the the software development best practices and all of that. Other people had thought about it and I can see how I can make this all work." Um what was missing was the glue to connect all of these pieces because it isn't just a warehouse. I need to think about security for the warehouse. I need to think about security as code. um I needed to think about the the the CI/CD process. I needed to think about um how do I do data ingestion and on and on,

[5:04] Noel Gomez
right? And and also look at things from the enterprise perspective because you don't just have one way of ingesting data. You may have multiple tools to ingest data. Uh you may have uh steps that you have to do after the transformation. And so that's where Airflow comes in because I'm like, okay, you need an orchestrator, an orchestration layer that can can, you know, communicate with the fixing tools or new tools and connect the end to end process and and surface issues and all that. And um and then the other problem that we saw was that when you're dealing at scale um even though all of these tools are open source managing setting them up you know setting up the platform

[5:50] Noel Gomez
and and and all all of the surrounding processes is not very simple and especially if you don't have a lot of experience and so saying I can run airflow uh which we we would run into people running airflow in an EC2 machine but that doesn't scale very well. Uh so saying like I can run airflow to saying I can run airflow on Kubernetes or there's a big gap there. And so that's what we saw that it's like okay you have all these tools they're great you need to connect them all and you need a platform and a way of of doing this stuff and I guess the final thing is um you we were working with people who were dealing with sensitive data and maybe

[6:29] Noel Gomez
they didn't want to use a SAS and so their alternative was I'm going to have to build everything myself you know because I want it in my cloud account and so we said well what if we made it so we can deploy it in your Kubernetes in your Azure ure or your AWS. And so it's like solving for all of those different things. And so a new developer comes in and in 10 minutes they're ready to go because they didn't have to install anything. And when you upgrade, everybody's on the same version of everything. And it isn't just the same version of DVT, but the same version of SQL Fluff or the same version of the AWS CLI or whatever it is. with multiple

[7:06] Tony and Juan
components and you have to make sure that those are in sync between your development environment, your CI/CD environment, your Airflow environment, right? So, it's keeping all of that stuff uh in sync. >> So, it's really about governance at scale maybe specifically about the technological part of governance. Correct. >> Both it's it's the the technology and the process that goes with it, right?

[7:06] Tony and Juan
It's how do you do all of those things together?

[7:06] Noel Gomez
So, I mentioned uh snowflake security, right? So people starting with snowflake may not know like what's a good way of setting up security model. Um and then the next piece would be like how do I do this in a repeatable way. So, uh you'll see people do different

[7:48] Noel Gomez
kinds of of uh ways of doing security and a lot of times people just just go to a uh what is that the a worksheet in uh snowite and they just type you know create ro and and grant ro to user and all of that stuff. How is that repeatable right? And so um you know we we were thinking about that stuff right from the very beginning cuz again I'm thinking scale I'm thinking not when you have three people but when you have 100 people you know. >> Yeah. And it's also like all every tool has a very clear boundary of like this is what we do and this is what we don't do. But that often means that when two

[8:26] Tony and Juan
tools are collaborating a lot of things fall into nobody's territory when it comes to governance. So, it's really nice that you already went ahead with that in mind. >> Exactly. Because, you know, like I said, DDT to me was great um as a framework, as a way to

[8:26] Tony and Juan
to do all those things that I wanted to, but what do you put inside the models folder?

[8:26] Noel Gomez
There's no convention. There's nothing that says like do it this way, right? Or do it this way so that when you have, you know, 6,000 DBT models, it it's not crazy, right? So, so you know I'm thinking in that direction because again um a lot of people are are coming at it from the perspective of like my project

[9:09] Tony and Juan
is small, I'm just starting or this the company's small. So they're not thinking 6,000, you know, I'm thinking 6,000 from the very beginning or or multiple thousand. People are thinking 100 or or maybe 500 DBT models, right?

[9:09] Tony and Juan
So the scale difference is there and and like you said in you you see this within organizations and across tooling vendors. Uh everybody is just takes care of their piece, right?

[9:09] Noel Gomez
Um but what I tell people is if I'm an organization, it's like saying I want a chocolate cake. Okay. Um but what they have is ingredients over here, an oven is over there, a mixer is over there, right? It's it's like all the pieces are there, but I just want a cake, you know? like

[9:53] Noel Gomez
help me get to a cake as quickly as possible. >> I guess with I guess with AI penetrating a lot of these enterprises, I'm seeing a couple interesting trends that relate to this sort of business of a data platform and governance. I guess on on the one hand I recall uh seeing this on call as they had this graph of the amount of users that had a certain amount of DBT models in their projects and you just see it grow exponentially over time and this was preai right and all I'm imagining now that you can generate this even faster it's it's going like asic right like there's it's not going to be so rare for companies to have thousands maybe tens of thousands maybe hundreds

[10:40] Tony and Juan
of thousands DBT models U at the same time I I feel like there's definitely some uh hubris let's let's say executives in these organizations saying well we're able to push this volume of of models why can't we build this capability inhouse now that we have AI like where do you where do you feel like >> data platforms like your own really fit in this picture and sort of steer that conversation into a more um wholesome uh direction. >> I think the way I look at it is the the basics haven't changed. So So I mean like the the same things that I've been talking about for years are still the case today. Um so if we if we think about AI as the same thing that happens

[11:29] Tony and Juan
when you say you have self-service analytics what you did is you said I went from I have a team of people who are very knowledgeable who who can uh you know set some standards and all of that to anybody and everybody can do anything right and so what ends up happening you end up like a billion PowerBI dashboards and Tableau and uh metrics are you know like duplicated and there's no ownership and right and and and so we we saw that happen and we said like no you you need to shift left you need to put that logic in in a centralized and govern place etc right well now AI is essentially doing the same thing it's it's it speeding up uh

[12:16] Noel Gomez
what people are creating and yes you go from like you had a uh I don't know let's 500 DBT models. So now you could have 5,000 DBT models. But that has a negative like did you really need the 5,000 DBT models? Um because that's going to increase your your costs and your and your run times, right? And so what I tell people is you still need those guard rails. You still need to have a foundation for AI to work better. Um I have an example where I remember when we were doing this migration you know back in in 2017 I used to say like we have to name things a certain way like you know tables the the the naming

[12:58] Noel Gomez
convention for a table has to be a certain way uh for a schema for a database like everything like made sense and people would like say to me like why do we need to do this you know like you're just being you know like painful to work with right like you're caring about things that don't really matter. because like so what if it's called A or B? Well, fast forward later on in that project, we were doing uh we were using notebooks and we created a Python library where you could reference. So without knowing where the table was, what database or schema, you could just say like I want sales data for this product, right? So pass into parameters and and you would just get a data frame

[13:38] Noel Gomez
back. But then when when we got to that, people were like, "Oh, it's so good that we did that because it was easy because we had the naming conventions, right? The the tool could now decipher or figure out what what the name of of the table should be." And so like this is what I talk to people about right now that you know when you're talking about let's say a security role um a a folder and and and a a and and a DVT model and like a schema and snowflake all of these things if if they all make sense if you have good naming conventions then you can connect the dots and what I tell people is if a human can connect the

[14:20] Noel Gomez
dots then the AI will also be able to connect the dots, right? And and if it's hard for a human, it's also hard for the AI because you you're I I've seen things where the same type of data is in a table with very different names like like the pattern of of the table name is different. And I'm like, I can't look at this and tell you what is the difference between these tables because I I just can't tell like this is a Salesforce name and this is a product name and what is the difference? And so like if I'm having a hard time figuring it out, the AI will have a hard time figuring it out. And then people will say, "Well,

[14:57] Noel Gomez
the AI could figure it out." You know, qu okay, but now you're using tokens and you're spending time. And so like all of the stuff we were talking about with um self-service analytics are just exploding, right? Um and so it's it's it's always a balance in my mind. It's like set up a good foundation, set up the guard rails, the governance, all of that kind of stuff. then you can move faster and not do it in a in a way that creates a lot of technical debt. And you know what what is one thing that people are not talking too much about this uh working at scale with AI before AI the loophole the loop was always you implement a solution you

[15:40] Noel Gomez
start experiencing the pain of the solution that you build kind of like you said the having all of the grants in the same fing snowflake you look for a solution you learn and then you reimplement and going through that loop enough times allows you to build a product that actually works in scale I think that with AI that loop is still kind of happening but we remove the part where we actually learn. >> I think we we we just accept I I think it's very hard at the scale and at the speed of AI to to even validate that AI is doing something right. I have a I have a blog post that I did where um I have one of those images where it shows

[16:22] Tony and Juan
you like the same image left and right and it tells you find the three differences between these two images and uh so I asked Chad GPT and it came back and it said um you know it's A B and C. It only had one right. One of the three was right and then it said would you like me to highlight the the things that are different? So I said yes. So what it did is it actually changed the picture. So it it hallucinated. It changed the picture so that it was right. And and I'm like this this is an example of something I can validate because I can see the picture. But imagine if I'm talking about, you know, Python code or

[17:00] Noel Gomez
or something else and I can't validate, you know, when you're talking about thousands of lines of something or or some SQL and I don't know SQL. Like I have no way of of validating this. So in my mind, we have to think now of of ways to keep AI in in the right direction and having some better ways of validating what AI does so that we can be confident in what it's doing cuz you get an answer. It's just confidently wrong sometimes. >> Example of that is having an AI that changes or removes unit tests so that your code passes. That's what I've seen. It's like, oh no, please don't do it. There's many many examples that we're seeing and and I

[17:43] Tony and Juan
don't know that anybody has an appreciation like like on the leadership level they don't necessarily have an appreciation for this because everybody sees the good and not necessarily the bad of of AI. >> I feel like we're all in this sort of um Groundhog Day movie where we're sort of relearning the same things as an industry. Yeah, you know, you'll see like people that have never touched code discover git and version control and tests and I feel I feel like your coral area of self-s serve analytics is pretty on point there. There's a lot of that same behavior and sort of incentive there and the same problems we're seeing here as well. Uh you mentioned the naming of the modeling. Uh I recall

[18:28] Noel Gomez
reading a um article from uh fast MCP, right? like the the the the biggest guys out there when it comes to MCP and Python and they were talking about if you want to write a good MCP it has to have at most like 10 to 50 things it does and the name should say what the outcome is right and this sort of reminds me of what you're describing here right like what it it almost feels like this is the way a human would be able to interpret it but it helps the AI as well and it it just runs a lot better um >> you you're you're trying to eliminate the guesswork, right? You're you're trying to direct it in the right way.

[19:09] Noel Gomez
And so like um it's it's really interesting because I showed somebody like on my project where I took uh I I I did a demo I think it was using Cortex code, but either one co would do the same, right? And it was like load Singapore data uh for the last 30 days, weather data for the last 30 days. And so it it did it perfectly because I have uh an agents MB. I had, you know, like folder names that made sense, you know, I had examples of other other load scripts. So it followed like I was using UV so that it it would like essentially create that virtual environment on the fly with with just the dependencies and I was using DT. So like it did it

[19:55] Noel Gomez
perfectly because it had something else to to base it on. It didn't just invent everything. And then I said okay now I want you to make um a data mart that does ABC whatever. And not only did it know to do that but it knew that I have a staging model first. So my first layer was staging. So, it knew it had to do that before it got to the mark. And, you know, like it knew that I had uh YAML files with uh descriptions and so like it was following the patterns that were established in the repo and in the read in in the in the agents MD and and skills and all that kind of stuff that like because I had all of that then it

[20:37] Tony and Juan
was flawless, you know. But if I don't have any of that, then it's just going to make things up and it may not necessarily do things right. Um, I've seen it where like another one is where it it will do something but using an old way of doing something like like it was trained on old data, right? And you're like that's not how you would do this in in DBT 110. Maybe in DBT15 I would have done it that way, you know? So, so it's like little things like that that that that also uh I I see become a problem. >> Yeah. Which highlights what you mentioned at the beginning, which is having already really good examples in your repository working at with a tool

[21:16] Noel Gomez
that has been built with scale from the beginning and just telling AI, hey, whatever I'm doing, just replicate exact exact process. >> And you don't even have to tell it like because it has the guard rails, right? like when you do this profile the data then do this and do that you know like if what what we're doing is uh teaching AI what we know right like so so we're imparting that knowledge as part of the context so that it it is more likely to get it right I still don't believe that you can I I still think you need some people to validate some of this stuff but it's it's um I I look at it as like uh it's it's

[21:59] Tony and Juan
It's an efficiency play. So it's not about eliminating somebody. It's about making somebody more efficient. So if we think about uh another example is GPS, right? So before there was GPS, people read maps and there were books and then you say like I want to go to this location and you find the city and the streets whatever and then you kind of like map this whole thing out and then we have GPS on the phone. The the process didn't change. like I'm still going from place to place, but now I don't need to worry about like where I'm located because I have GPS to tell me where I am. I've put the location where I'm going. There are maps that that that

[22:34] Noel Gomez
do all of that. But if you remember when like Apple Maps came out and all this stuff people would there would be like stories about like somebody drove into a lake because the GPS told them to to go that way, you know? And it's like but that's that's what's happening with AI, right? like I can't just like just trust it. I have to have some common sense and I have to validate what it's telling me. But if I do that then it is making me more efficient because it saved me 70 or 80% of the work. Um and I'm just reviewing I'm I'm I'm guiding it first like this is what I need you to do and then I'm validating that it followed my

[23:13] Noel Gomez
instructions and it would do things in the way that I would do them, right? or or or some somebody with more experience >> thinking a lot about the extensions that arise out of something that you're describing here. There's a couple of things that you see happen um in the industry as a whole, right? I guess first of all if you are more efficient uh a lot of companies might make the decision to do it with leaner teams and especially on the junior side where you would typically kind of push off sort of this more menial work that is valuable but um not necessarily something you want to invest your time in if you can have AI do it for you. And on the other hand,

[24:03] Noel Gomez
um, if you do that, then maybe in four or five years, you don't have those experts to give those examples and and build that knowledge into your AI system as you're describing. And I I it it always strikes me that people don't really look at how these models have been developed. Like if you look at how are they even able to create all of these amazing things? Obviously they're trained on all the human knowledge and all the data out there but there's a whole process of post training that happens where there's you know like thousands of people going over Q&As's and saying like okay I like this answer best based on these you know four check marks right uh and

[24:49] Tony and Juan
what do you think the future is you know if companies decide to go this direction because there's obviously financial pressure right like there's a lot of stuff going on in the world and there's just outside of AI just a ton of pressure on bottom line of companies. Um, yeah. Where do you what do you think we're sort of waiting for in in the data industry in a couple years?

[24:49] Noel Gomez
>> This is this is an interesting one because it gets into the society the the impact on society of these technologies, right? So if we think about um when you went from agricultural to the industrial revolution, there were good and bad things that came out of that, right? And then there was a

[25:31] Noel Gomez
there was sort of like a push back from society to say like, okay, we need we need worker uh protections, right? We we need things that make sense like people shouldn't work seven days a week um or so many hours in the day. Um and so that's when you have unions and and and you know that that that push back that that that rebellion came out of um you know the the the the capitalists taking so much having so much power over the people and we're we're there again. I think we're the beginning of of that again. Um, and so I think it's it's going to be interesting because like I'm a technologist. I appreciate it, right? But I also can uh appreciate the

[26:21] Noel Gomez
negative the good and the bad, right? That comes with these things. So you'll see like, okay, now you have all these data centers and and they're polluting water and and so it's like, okay, how how are we going to uh to deal with that? because it's not it's not necessarily about us, it's it's our grandchildren, right? Or or people in in the future. And and so I think there's there's bigger implications that we haven't really dealt with yet, I think. Um and it's going to be interesting to see because yeah, it's it's it's like you you sort of like um I always talk to people about systems optimization. If you if you look at an organization when you have a project you

[27:08] Noel Gomez
usually just optimize that part of the project regardless of what's happening around you and so you do local optimization versus system optimization right the system means I would I would optimize multiple things and maybe it's not optimal for that one process but across all of the processes is optimal right because there's many things to consider and so I think in here we're getting to that same point where like we're optimizing so much for one area at the expense of, you know, environment or uh society, right? And then it comes back to hurt you because then it's like, oh, you know, now you have higher expenses in in, you know, society expenses like like homelessness or uh health care or all of these things,

[27:56] Tony and Juan
people that can't support themselves. And so like what we did is we we we optimized for one area at the expense of another area. And so it's going to be interesting to see how the world takes care of this kind of thing because you see different countries doing different things. I mean it's the same thing that happens also with um personal data, right? If you look at like how personal data is treated in Europe versus in a lot of the US, it's it's a very big difference, right? And so that's what I think will need to change, but we'll have to see how how it goes. And like you said, like how do you get if if you're only relying on AI, how do you

[28:39] Noel Gomez
get people who don't have the experience to get good enough to evaluate the AI? So it's just AI evaluating AI that, you know, now you get into the matrix type of thing. Sorry, >> that's another interesting angle to break this down. Maybe we're just a few months or a few years away from AI becoming very expensive and companies realizing it's cheaper to hire juniors than to token vaccine. So there is been some research where the graph of the of the AI agents overtaking even the senior engineer in in some in some regions. Yeah. this uh internet theory that I saw that kind of broke my heart that there is like a cursed generation like I forgot what were the years but there is a range

[29:25] Noel Gomez
of like three years that if you were born in those years your parents probably struggle around the 08 crisis you either had prom or the first years of university when we were in lockdown and now you enter a job market where like nobody's hiring juniors but maybe it's just that generation I'm sorry if somebody recently is from those years. Maybe it can course correct a little bit if indeed token maxing and the pricing just kind of makes it again unfeasible to fully automate a lot of tasks at least from a monetary perspective. I >> I don't know. I I think I don't think it's stoppable. Like I don't think you can It's like saying you're going to take people who move to the cities to

[30:11] Tony and Juan
work in factories to then go back and work in a farm, right? It's it doesn't happen, right? You you only go forward and and so like it's it's something else that needs to happen that we don't know know yet. But I don't think it's saying that junior engineers like they're going to hire more of those because even if it's more expensive, I can still do more. I I can do more with uh it's it's more efficient, right? So, what what

[30:11] Tony and Juan
I talk to people about is um you can't just tell me what the AI did. I didn't hire you to just tell me what the AI told you, right?

[30:11] Noel Gomez
Because then I could have done it myself, right? And so what

[30:55] Tony and Juan
we need to get people to do is to think, right? To to to to say like, okay, AI is telling me this. Does that make sense? What other things, you know, how should I direct it? Right? So, so you become more of a conductor and and you're you're getting people to like learn better ways of working or or even challenge the AI, right? To like, did you think about this? Did you think about that? Right? Um and and so like that's where you want the human to to get to. But I don't know that you get to say like I'm going to go from AI to uh junior engineers because even if it's if it's more expensive today I mean all these companies are

[31:37] Tony and Juan
fighting you know you know there's so much money and everything in there like I don't know I don't know you're going to go back to to the way it used to be. >> No it's just and I also don't know what what will happen. It's just a funny angle to think that it's also possible that it's not a straight line. But you actually brought up a really important point that I wanted to mention and that is I do agree with you no that some people feel like intermediaries just like vomiting literally what AI says and like why do I even need you in in the middle of the chain versus some people that really know how to leverage that. Can can I actually ask you to to phrase

[32:13] Tony and Juan
that a little bit more into career advice? Imagine you are just working to uh you're just learning how to work with AI.

[32:13] Tony and Juan
What is your advice for those people that truly want to like be in control of AI instead of just being media that is not bringing too much value?

[32:13] Noel Gomez
If if we're focusing on data, there are things that you should learn about regardless of who's doing the work, right? So So what's changing is who's doing the work. It was me and now it's AI. But in either case, I should have some clue of what I'm doing, right? So I should know what Kimble modeling is or data vault or what is uh change data capture or or you like

[32:57] Noel Gomez
there what is git and why would I use git and what are good patterns of branching strategies? Um why do I need pre-commit hooks or you know like so like I I focus more on the on the why and the how versus on the who is doing the work, right? because the who is what's changing but but the why is where you need the the the uh the person to essentially direct the AI. So I'll give you an example. Anyone will always show you like okay I'm going to have AI like make an application to do X and and they'll do it at different levels right and and maybe you get it go to fewerations. Okay, but none of them talk

[33:38] Noel Gomez
about like did the AI use the right design system so that whenever I add a new button, it doesn't pick a different color, right? Like I have a way of like defining the colors in one place, right? So that that my code is dry that that it's easy, you know, that only comes from experience and knowing that that's a better pattern, right? to so so you have the people who are just sort of like the naive genai users which can get something done but how well it scales or how well it it survives right is questionable because you're really just doing that that that self-service analytics thing right so self-service analytic I just did something and I gave you the report and

[34:23] Noel Gomez
all of that but in a year I have a thousand of these things and I don't know what anything means versus saying like, hm, okay, I know I'm going to need this metric multiple times. Let me put it in the in the DBT model. So then I could just leverage it in multiple dashboards, right? So So that that that's where I tell people learn that learn learn the like the how things are done and the because those those rules don't change, right? These are like the fundamental rules that I call them the like the data truths, right? doesn't matter what you change. There's always more data. There's always data ingestion. There's always data quality issues. There's always ownership, right? There's a lot of these

[35:03] Noel Gomez
things that don't change. And so, focus on that because AI doesn't fix a bad ownership, right? A decision. Who has to make a decision a lot of the time even like when the developer is doing the work, the developer doesn't know how to make the decision? They just don't have enough context or or something like that. like if I have data that I'm ingesting some data and it has some PII data I don't know I don't know which fields you know and and what are the rules of the company and the different levels of of data security so so those are things that are external to the AI that somebody needs to learn and and understand so that they can

[35:39] Noel Gomez
direct or at least set up that those guardrails for the AI because that's the other thing right you still need that >> interesting so it's kind of like AI accidentally promoted you from implementator into manager, designer or architect. And most of the advice that you just gave is like how to be a good architect or a good manager or a good designer. And like a good product manager. Yeah. Yeah. Right. Yeah. Exactly. Like Yeah. It's it's like a general contractor if you're making a house, right? You don't have to be the electrician and the plumber, the painter, right? But you need to know like how to sequence them and and and how to evaluate multiple proposals to see like should we go with this person

[36:26] Noel Gomez
or that person, you know? >> Cool. Take what I'm getting from this. I I just have a flashback from from training junior engineers at the start of their career and I was always trying to predict like who's going to like have a easier harder time picking things up, right? Like let's say we have four or five junior engineers, we're training them. Let's see what happens in two years. So with the same amount of time, same effort and often times what I would find is the ones that struggled the most were the ones that didn't ask questions in in general like they were not curious enough about how things work, why they work, could have even done better. I think you illustrated this in your story

[37:10] Noel Gomez
like you were having a data platform, you're like it's not quite to my standard yet, but you had that standard. You had that drive. you had sort of that intuition of this is what it could look like. And I think that doesn't change with AI. Like I I think sort of the the way I look at it, it's like wisdom, it's taste, whatever you want to call it. Um you have to sort of fight that urge of of instant gratification of look, I pushed something. Here's another feature. Here's more volume. here's more stuff to saying, okay, that's all nice, but can we do better? And I I guess the what I would tell to anyone that's a junior now coming on the market is if

[37:52] Noel Gomez
you focus on doing that really well for long enough, you will find a place uh where that's appreciated. Where that is, who knows, depending on where the market goes, but there is always going to be a demand for that. I mean, even if you look at if you look at the code or the data models that AI generates, it's sort of like it's all right, but there's always things you can pick on, right? You're like, well, that's good, but what about the governance? Or, well, that's good, but what about performance? And then it's sort of like this average representation of what a data model would be, but not necessarily a good one. It requires a lot of your input to

[38:34] Noel Gomez
go from okay to great. Uh, and I I don't know if that's the issue that people don't appreciate yet because we're not f far along in it is that what you're doing when you're when when you're not caring about that stuff, right? When you only focus on delivering fast, you're creating a lot of technical debt and and then it becomes like, oh my my warehouse bill, my data warehouse bill, it keeps going up, right, exponentially. And now it's like, oh, now I need to go focus on that. But it's like, but you shouldn't have had that problem, right? If if you started the other way. So I I talked to people about like there there's two types of of uh

[39:18] Noel Gomez
leaders, right? There's the ones who are just focused on like quick wins. Let's just do something, you know, whatever. we'll just get something out there. Um, and so they move fast at the expense of long-term, you know, moving slow. So, they move fast now, but then they move slow where I say to people, you have to flip that around. You go a little bit slower now to go a lot faster forever, right? Because the conventions, those things that we were talking about earlier, like you you said that once and now it's like, you know, I I do this with people like in 6 weeks we like we figure out the CI/CD, are we going to do blue green deployments? Are we going to

[39:56] Tony and Juan
do uh uh how we're going to do the the the security model in Snowflake? All all sorts of things, right? It doesn't have to take a year. It didn't take that that

[39:56] Tony and Juan
long in the grand scheme of things, but it's for people to at least have the conversations. Like I've had the the the you know met with teams and they're like oh we hadn't thought about that or you know like what are you gonna do when this happens?

[39:56] Noel Gomez
DBT is a perfect example. So people starting with DBT they don't realize that like DBT only creates things. It doesn't ever get rid of anything. And so then they end up with orphaned models in production right? Like that's a very common thing where

[40:30] Noel Gomez
it's like I renamed something or I removed something. It's still in the data warehouse because nobody deleted it. So I tell them okay this is what's going to happen. this is what you need to do, you know, and and then here are the implications like you can't have something like so what we do is we have a macro that goes and just drops anything that's not in the repo, right? So like the repo is your source of truth, but that has another caveat which is only dbt can be writing those models, right? If you have some other tool out there that dbt doesn't have the code for, then it's going to get dropped also. again you have to understand these kinds

[41:03] Noel Gomez
of things and and I don't know that people have that appreciation and uh you know and and that's when we were talking about like where people move to like having that curiosity to me you know building that common sense like the why why are you doing this stuff like not just do it but like why am I doing it and um another good example is when people talk about like I'm going to just have one big table so I'm going to take whatever sources are and I'm just going to make one big table and it's going to have 500 100 columns. Okay. But then how does that impact your genai, right? Your tokens and like you know like you have

[41:38] Noel Gomez
to like be thinking about multiple perspectives at once. And if you're only focusing on delivery, you're not thinking ahead. You're not thinking the what is the north star? Where am I heading to long term? Um another example that I use is is uh light. Okay. So we have we take for granted that any of us could go to the wall and hit a switch and light turns on. That's it's amazing, right? Like we said like we we can't even think about the fact that like it had to be generated somewhere. It had to be transmitted. It had to be transformed that you know like all of these things had to be perfectly working right and and and deal with like spikes in in in

[42:20] Noel Gomez
usage, right? And so many things have to go right so that we can have this call right now. Like we we wouldn't be able to do this. We wouldn't be even if we if we don't have a stable power grid. We're not talking about GNAI or any of this other stuff. But we took we're taking for granted that that that didn't just happen by magic, right? It it it took a lot of people to think about this stuff, to think about resiliency, to think about, you know, power spikes and on and on and like all of these things have had to be developed over years so that now we can hit a switch and light turns on. Oh, >> you know, both of you made me realize

[42:54] Noel Gomez
something and that is we're maybe heading to this this dystopian future where the most important part of an interview whenever you were applying as a programmer could be a personality test because it seems that we somehow agree that there are certain psychological test that indicate that you're going to be a better AI user and we talk about things like intellect, curiosity, but maybe the need for having things clean. I don't know the list, but if you're a psychologist listening to this, do some research because if you're the first one to develop that interviewing technique, you there is some big money in there. >> Yeah. And I bet existing personality tests might be able to tell you something, right? That there's that uh

[43:36] Noel Gomez
DISC assessment. I don't know if you've heard about it, but anyways, DSIC and each you take you answer all these questions and at the end you end up with a personality trait. And then you have the people who are just like you just want to do something, right? They're disruptors. But then you have the people who are more like they're very thoughtful and and are thinking long term and all of that. And and I think it's always a mix. Like it's not like all one direction or all the other direction. Um but yeah, having curiosity I think is is definitely a a huge thing. Like um when we had co and they were saying like we can't find cobalt programmers. I don't know if you heard

[44:14] Tony and Juan
like during the co right there's I'm like I've never seen cobalt you know let me go learn a little bit about it. I'm like this is really interesting and they reminded me of SQL and all of that stuff and I'm like oh okay and I could kind of see like okay it was kind of for business you know and so like it's interesting like I don't have to like say like I'm a cobalt programmer but I at least know enough to have an appreciation to like how that was developed. I would say personality is one way to describe it. But I mean it's funny like somewhere in your co conversation you mentioned you develop some common sense. I would I would rephrase that as wisdom

[44:55] Tony and Juan
or like a a crude battle scar so to speak. I'm I'm I'm personally now learning how to golf, right? And like I wouldn't go and have an AI tell me like oh like train me to become a good golfer. know, like why would I pay a coach or a teacher because they can point me right to the point that I need to focus on to like triple or quadruple my progress and that's what I'm like paying for. And I I think that's a similar thing if you know you're new coming onto the job market and you're trying to like put yourself out there if you're more experienced or even if you know you're a SAS company or a platform. I mean that's the the the the main

[45:35] Noel Gomez
reason why that will not go away. I I think I see all these articles like there's a SAS apocalypse and there's not going to be a place for all these companies. I'm like cool. But they thought about what is like the essential features that people need cuz they were in a world where implementation was very expensive. So everything they built was very very very carefully curated. Whereas now I can just like turn on the fire hose and do whatever and often end up in like a really weird place. But there's an article, I don't know if you know this company, uh, Base Camp. Um, it's the people who who put Ruby and Rails, they have an article start with no. Okay. So, I remember I I was doing

[46:18] Noel Gomez
this application. Um, it was for vendor on boarding and all of that stuff. And we did all this workshops to figure out the features and prioritize releases, all of that kind of stuff. And I tell people like there were a bunch of stuff that were just never done. Okay. And what would happen is uh the the the business person I was working with at the time, people would go to her like, "Oh, we we must have this feature to do this." And I sent her that article and I said, you know, like just whatever it is, whatever they're asking for, just say no. Because if it is very important, it'll keep coming up, right? It's not just from one person, it'll come from

[46:52] Noel Gomez
multiple people because that feature, you have to look at it as like it isn't just doing that feature for that person. it is that now that feature you have to maintain forever and ever, right? Like it's very hard to now say I'm going to remove this thing that only one person in the whole company is using it, right? Or or or you know and and and there's testing and regression, you know, like unit testing, all of that. Like so by saying yes, I committed myself to this long, you know, like life cycle of this thing that maybe nobody actually needed. And and that's where like to me the AI thing is like if I can let somebody do self-service and do AI and it's a

[47:33] Noel Gomez
throwaway whatever but if they start relying on this and more and more people start relying on that stuff that's when like that needs to shift over to something that is more governed and and better and so like how do people make that distinction I think that's still going to be very hard and people are going to learn over time like I said when they're when their data warehouse house bills goes up when things are failing all over the place and nobody knows how any of the code works. You know, like there there's going to be an inflection point where people will appreciate these things that I'm talking about, but it's we're not there yet because it just hasn't happened enough, right? Like I said, the

[48:12] Noel Gomez
industrial revolution, nobody cares that like one person is falling off a building over here or running over by a forklift over there. But if it happens many, many times, you're like, "Oh my god, we have to do something about this, right? This is not good. >> Yeah. So, another piece of advice you just gave us is for anybody, if you're in a centralized data team, check what type of AI products are gaining traction internally and well, maybe rebuilding them but properly from the beginning and make it scalable. That could be a really interesting use of your time. >> And that's why you need the people who are curious, who are asking the questions. We were talking earlier about like who are the good people that you

[48:52] Tony and Juan
can like the people who ask questions and I think it's it's the same thing. So like I'm not maybe I'm not asking questions of a human but I'm asking questions of t

[48:52] Tony and Juan
he AI like why did you make that decision or is there a better way?

[48:52] Noel Gomez
What are my options? You know uh what what is Kimble modeling? What is data vault? What are the pros and cons of each? You know like somebody has to care about that versus just saying like I don't know just make this thing into a dashboard. Done. I didn't care. >> For the people listening in that that are saying, "Yes, I want to work with Noel." Where can they find you? What are you working on? Where can people find

[49:28] Noel Gomez
you online? >> I think the the biggest thing is LinkedIn. If you go to uh LinkedIn and search for Noel Gomez, I think my LinkedIn is I don't know. It's Noel Gomez, I think. But anyway, um yeah, I post about a lot of things in there. uh next week. I don't know when this is going to be published, but I'll be at the Snowflake conference and then I'll be at the DBT conference in September. So, um yeah, find me in any of those places. I'm always happy to talk to people even like, you know, I'm not a I'm not a salesperson, a marketing person. So, sometimes, you know, people meet with me and they're like, "Oh, this was a good conversation. You weren't

[50:05] Noel Gomez
just showing me slides, you know, and I'm like because I care, you know, I'm more passionate about this stuff." So, I'm always looking to meet people in this space and have these kinds of conversations um because I enjoy it, not not just because I'm trying to sell something. >> Nice. I will see you definitely at the DVT one.