Transcript
Speaker: Over a weekend, the system built five models that were impossible for us to do in the past. So we were super excited, incredible gains offline. And we're like, okay, now put it to production. It took us three weeks because the code basis was a mess. But it's not because AI is not doesn't know how to the code. We haven't learned how to put AI in the direction of not only you can you explore, but the way to then put that in production is through these templates, these formats.
Speaker: managers are going to be in a really good position if they can get their hands on agents, as I've been doing for my own personal projects, that you're like, actually, I can build a team with this because this is what I've been doing. It's just that I have to codify it and code it my knowledge and and so on. So I think that's something that I want to bring out because people managers out there might feel, yeah, but all these things that are happening are for the individual contributors. And I am like and i understand the the the thinking, but I think there's a missing point. I think that if if we rethink the way we've operated as managers, Jai I think puts us in ah in a really good position.
Speaker: Today's episode is brought to you by Omni. Most companies I speak to want AI analytics by failing to put projects into production. That's where Omni is different. It's the semantic brain that grounds AI in the heart of your business logic, giving you governed answers, whilst also the depth to identify root causes. It's intelligence everywhere. from your spreadsheets to their chat feature, even within your product. Omni moves you beyond the dashboard.
Speaker: Don't just take my word for it Trust teams like Booklexity and Synthesia that are already using Omni to deliver intelligence that people trust. Check out them in the show notes or visit omni.co.
Speaker: That's O-M-N-I.co. Now, back to the show. Hello everyone, welcome to another episode of the Stacked Data podcast. Today I'm joined by Jose Panero Garcia who leads data science, marketing and tech teams at Skyscanner.
Speaker: Jose's actually had a really interesting journey into data. He started out in robotics, moving through consultancy and insurance and is now leading um a large team of of data scientists at Skyscanner, the world's largest travel platform.
Speaker: um and In today's episode, we're going to unpack what it actually takes to scale data science teams that really drive impacts from how you prioritize the right work to mistakes teams make as they grow and and how AI is really starting to reshape and the teams that that we're building. so and yeah Hopefully, it's be really useful for a data leader or anyone that's thinking about scaling out um data science teams and how to prioritize the high-performing impact work. so Jose, it's great to have you on the the show. Thanks for for joining me. How are you doing today?
Speaker: I'm very good, thank you very much and thanks for for having me. It is clear that I'm in short sleeves, which is amazing. You're in long sleeves, but we'll still have a ah warm conversation yeah despite the weather temperature. It's not quite the same weather in the UK as it is in beautiful Spain.
Speaker: But yeah, i I obviously gave a bit of an intro there um to you, Jose, but it'd be great for yeah for you to to talk a bit more about your your path into your role now and and how you got there so the audience could and get to know yourself a bit better.
Speaker: Yeah. um I think you captured it well in under 30 seconds. So that's very, very good. And I did the study engineering and did a master's in robotics and i was to sort the sort of thing i wanted to do but when you leave university and hit the real world and they didn't let me touch any robots so i was like okay this is done not for me um so i did pivot and and go into consultancy spent a couple of good years there didn't see myself
Speaker: spending more time in consultancy, it wasn't my take. Not that it was and anything that I would say it's not great, it's just it didn't fit what I what i wanted. Into insurance, a good time working there with pricing models and and doing like learning the craft of of becoming a data scientist, lots of statistical modeling, lots of data quality, building pipelines, so like really really, really, really useful experience there.
Speaker: and And then as of six years or something like that, and I joined Skyscanner. I did join as an individual contributor, so again, doing this sort of like the craft of building machine learning models, but then I did pivot into into the people side of things, leading teams, becoming a people manager. And that's where my journey started more on the leadership side of things.
Speaker: And I've been doing that for the past four years. um I've got now, I kind of oversee three different data science teams, around 20-ish people in different area So it's ah it's a different challenge that I have now compared to when I was first an IC, second time manager our delivery manager and now like having to deliver through other managers and principals and so on. Which is a different challenge, a very cool one. and yeah That's bit of me and fun fact and that you know about Harry. I joined Skyscanner in the middle of COVID.
Speaker: So that was very scary, but it did turn out really, really, really well. and I'm very happy to still be here working at Skyless Gunner. It's a pretty good place to be. But I was a very scary time, I can tell you that.
Speaker: Yeah, I mean, definitely not the time to be joining ah the travel space in in in COVID. But yeah, I suppose it'd be great to get a um an understanding of what yeah what was, when you joined Skyscanner, what did the data team look like? What were the problems that you were trying to to solve then? And then I'm sure we can go on to how that's evolved um over time. And um yeah, what what what was the landscape like when you joined?
Speaker: Quite a few differences, of course, because there's been six years from starting to to today. and One of course, that the number of colleagues, the number of people in in the data science team, where we have now grown over 50 and compared to, don't know, we were under 30 or something like that. I can't even remember the the numbers, but we've grown quite a bit.
Speaker: But not only it's about numbers, it's about also operating models. and In the past, we were more like embedded within engineering teams. So we didn't have like um a cohesive group of like data science. and so But now that has evolved into this like sort of data science drive where we have specific data science quads that work within specific tribes and verticals. And that really helps have like that ah sort of like t-shaped squad where you can have a that functional really deep team in an area and and sometimes people do move around but that's a big change and compared to when when i started and of course from a technical perspective what can say like we were in the world of um at the beginning
Speaker: having to deploy models the hardcore way with AWS and then Laterbricks came, there was all this AutoML thing and now LLM. So like technical world has completely shifted and from from when it started to today and the problems that we're solving have also been evolving quite a bit. and One good example, know, Flights Run game Flights ASB.
Speaker: Skyscanner score. And so the evolution of the actual machine learning algorithm six years ago to today has gone through all the journey of like heuristic, more classical machine learning approaches like three based models, deep learning.
Speaker: Should the next evolution be a hybrid between deep learning and LLMs? We don't know, but that's it's been a full evolution of of that sort of thing. Amazing. amazing so Obviously, you said that one of the big things, obviously, the the size of the team has grown significantly. how how howd you guys How do you approach scaling a team and what have been the typical challenges that you've come across and how you and how have you overcome them in that in that growth journey as ah as a leader?
Speaker: yeah so and Give an example on a smaller scale. So I think it was in 2022 2023. There was no data science team in marketing. As as I just said, like we've been growing and not only a number, but also on the operating model. And there was not really data science being embedded in marketing in a meaningful way.
Speaker: So there wasn't a question, should there be a marketing team given that there's a lot of spend from i paid marketing and should we optimize that spend? So when the answer was yes, we should.
Speaker: And the approach at that time of the company, because we didn't have like, okay, let's hire 20 people in a year, that did not happen. It was start mode. So that was a one strategy. So you start with one, two people, just focus on one single problem, um get the relationships working, figure out what the next problems would be. And with those, when the people working in in marketing, the data science team working in marketing, when you never get below P0 priorities because you're stacked with yeah the two people, 50 problems, you're never going to get below P0. That's a good indication that there might be a need for more people.
Speaker: So it makes it easy for stakeholders to request with more people and that's how the marketing team grew from two people to eight to ten data scientists. Of course, in that journey, you have to demonstrate value. There's no way you're going to grow a team if there's no ROI on that front.
Speaker: So that's one approach, which is the start slowly, build it up, and foundationally, the team kind of accelerates a bit. It worked really well, but it does, you do go through the not death by a thousand cuts because it's not something and that is negative, but the person running, like building the team, in this case it was myself, it's a lot of like having to always be on top of, do we need more people? We need to then redo again all the plan of business casing, we need more people. So it's a lot of work there, but it's a good strategy. It works.
Speaker: It has its cost. The other strategy is, And taking it to another and example is when two or three years ago, we were around 30 people and 30 data scientists.
Speaker: And instead of growing slowly each part of the team, the strategy was, okay, now that we've demonstrated through the past two or three years that we've added a lot of value as a discipline,
Speaker: we definitely need now still using the strategy of D. We don't get the low B zeros across many of the teams that we have. Then this is more like a higher level executive C level, CEO, CTO.
Speaker: Hey, do we could we add 15 more data scientists? That's a big ask. and So if you get that approved, then fantastic. That's the strategy of like go fast, go all in.
Speaker: But it does come then with the challenge of, oh, you've increased your team by 50%. So is the operating model like across a discipline working? So what but you hear will not get you there. So do you need to review all d the ways that ceremonies are done? a Have you got enough work for everyone? Have the teams, are they ready to have an influx of two people that can onboard into projects? Are they ready for that? So two strategies that I share there, both work, both have their their their cons that if you're just aware of them, then great, you can mitigate the effects of that.
Speaker: yeah It sounds like ah umm obviously planning is is key there. Have you got the right systems and processes to especially if you're you're moving at at that scale and speed? um yeah It's not just from the the hiring side and how you knowre scheduling, have we got the capacity to interview that many people to make that then the onboarding? um what's the utilization like? so I imagine there's a lot of work of really having good relationships with the wider teams, what is the capacity, what are the projects, and having a clear roadmap to then be able to plug people into.
Speaker: and Interesting. Jose, I'd like to take step back, you obviously have found your path into leadership. and One of the things I think you all we often see um with data professionals, they definitely get to the and get to a point where they have to maybe pick the IC or the the leadership route. I think the recently with the AI and the ability of what a ah senior IC, like staff principal level IC can do, we've seen a sharp increase in demand for that type of IC track.
Speaker: compared to previously. um but yeah how did you make that decision between going down to an IC or stepping into leadership and what's your lens on it?
Speaker: um I guess there's not one single reason. and Basically, because if I put myself back into my own shoes when I started to get the people row, I don't think it any IC, despite the amount of books that the person can read on leadership, there's no chance anyone understands what this is about.
Speaker: So it's a, there's an opportunity to build a team. and Okay, fine I'm up for the challenge. Let's do it.
Speaker: But I was not, I could not answer at that point that helping people drives me and it really, it's something that I look look forward to because I was not doing it.
Speaker: And so it's impossible to have that view at the time. m Plus, when I made that sort of shift, as I said earlier, with the first strategy, because I was going slowly, I was making the adjustment of, hey, I'm still 50 percent IC, 60 percent IC, but also a manager of one or two people.
Speaker: And then that gradually expanded scope. So my my percentage ratio of doing IC work and doing management work was kind of slowly adjusting. So I had that that and opportunity to experience it and slowly in a way that I could enjoy it. And now I'm super grateful for that because whilst I really like the IC work, a bit of a geek, you know, personal projects, GitHub and stuff, m I still feel that I can add way more value by helping the team um than doing
Speaker: the best system in the world. There are smarter people than I am in the team who can do a better job at IC roles. So even if I look self-executely my own career, I know there's some other people out there in the team that are going to over um pass me over. So that's that's ah that's a a byproduct or from a positive front in this sense.
Speaker: But i really, if anyone is looking to make a transition from a IC role to a manager role, And everyone, like we have to, they have to be sure that they can, they will be happy without having to be the ones controlling, be the ones doing, be the ones solving it.
Speaker: And because that's what not the manager does. The manager creates the processes, the environment so that others solve the problems. That's what you have to, to try. You're solving problems in a completely different way.
Speaker: and And it's very rewarding if you if you like this sort of stuff, because you're thinking way broader than the flowchart of an architectural diagram, which is super cool, but you're like, oh, okay, but should these two people work together? Why this first? Should we phase this out in a different way so that we can go fast? Or should we slow down because we need to invest in a system so that it accelerates in six or seven months time?
Speaker: get the input from the from the principal from the staff data scientists what should we be doing and map that out a bit better but but yeah and that's I didn't make the decision and I go it was why path that was kind of given to me and so maybe a different story compared to two other like tech companies that basically you make the shift directly from IEC and boom, there you go, six, seven, ten people. I didn't experience that. and I know no that could be a bit daunting, but and regardless, I think I would still make the the same decision if I were to just step back. and so
Speaker: Nice. and I think there some some people get it sort of naturally happens. um Others, it's you know you've got to sort of choose that path. I think it's really interesting to get your your lens on it. and As you've grown as a manager, what are some of the mistakes, a manager and a leader, what are some of the mistakes that you've made and or that you've learned from and that you know now and that yeah I suppose are common things that maybe you see other managers now that you're ah you you manage managers that you see them making um and and how do you help them mitigate them now?
Speaker: Over control is one. and I'm not saying micromanaging, it's more the... and because you You've come from the ranks, you know the technical details and and you want to be in all decisions.
Speaker: You just want to be there. Not because you want to control everything, it's just like, you know, in all meetings, in all decision points, in all planning sessions, in every technical discussion. and That works only if you've got a couple of projects and maybe one or two therapy parts, the moment that you scale to a bigger team.
Speaker: and you will become the bottleneck and that's the main issue the person who who should be helping the team is actually hindering the team because the process set up is and that the manager has to be in all these decisions and that's not right that's not true and so that's one thing that i evolved over time and and to for me managing to get to that point also learning how to provide clarity on expectations. So for the reports, now here's my here's my expectation on roles and responsibilities, you're what you're accountable for, were you what I would like, would it be okay if this happened in this quarter or this month? I always, it has to also come from the individual contributor to say this is feasible, this is not feasible. But once the plan is in motion, then you have to like, let go. And the good thing is,
Speaker: If the person is not delivering, everyone is clear that that is an issue. The final accountability is on the manager. So if something breaks, like you are the face of the team and it doesn't really matter that you had a pact with someone and they didn't post it, that's not ering anyone's problem more than yours.
Speaker: But still, it does provide that clarity to direct reports on we're a team, but you have to own this sort of stuff to the level that they're accountable for. You can't have the same bra responsibility to a junior data scientist, a senior, to a staff.
Speaker: But I think that providing that clarity and removes a lot of, I don't need to be in this. I don't need to be in this decision because this is yours. And if there's an issue, the responsibility is bring it to me.
Speaker: So I only intervene when the team thinks I should intervene. And then to avoid the over-reliance on, well, we don't We don't want to make tough decisions. Let Jose take them.
Speaker: Over time, that also is an evolution for the team because you are also giving it the, well, what would you do? Why do I have to make the decision? Let the team try to make. So it it's a fine balance. There's no, as I said, there's no playbook.
Speaker: You can read and read and read books. I can tell, i can mentor people saying, this is my experience, this how it should be done. But until you actually go through it and experience the delegation right in clarity, deciding if this is a problem that you should solve or not, yeah it's only coming to experience. So...
Speaker: As I said earlier, if anyone wants to become a manager, the time is yesterday and not tomorrow, because this can only be learned through it. yeah i I really resonate that with the, I suppose, the yeah it's not micromanagement but but but control and wanting to be be involved. and Especially, I think, when you're in that technical standpoint, you're making all of them decisions, it can be hard to to step away from that. but Sometimes, I think you've got to let people you let your team fail um and yeah make the make make so make some mistakes. You can't be over everything all the time. you're
Speaker: you don't have the capacity to. so That's really interesting too to hear, Jose. I suppose one of the things that we're really seeing massively over the last 12 months particularly, but is from our side, um is is a drive for everyone to be able to articulate their that their impact. and how data is really moving the the the needle for for the business. So um it's something that I think teams and individuals, even leaders do sometimes struggle with in in in the data space. you...
Speaker: um how how do you typically so um make sure your teams are working on the right projects and the driving the right impact because there's always a million and one requests. I'm sure Skyscanner is no different. What's your framework to to making sure you're you're working on the right bits?
Speaker: i'll um I'll share my strategy. I'm sure and'm sure no, I know. Because I've spoken with other leaders, other companies work in a different way where the data science team a of solve multiple problems, smaller problems. um My point of view is that, one, if I want my team to solve a problem,
Speaker: and hopefully it's a one a long-term problem something that is foundational strategic and
Speaker: and but to get to that long term one request that i always have to stakeholders would would be after any poc initial thoughts could we could we get the data science team self-sufficient Like that's, for me always, it's the final goal.
Speaker: In any system that we're going to be in, how can we give the team the chance to iterate fast? Because one of the main issues with machine learning and AI, and this was like a long-term report, it's a marathon.
Speaker: And one thing is doing deep analysis. and Another thing is optimizing something that exists and you're going for the fine grain, you're tuning, those percentages that it the more you shave, they look small, but you shave them continuously and it's a lot of value.
Speaker: like But it's still a marathon. So if it's a marathon and the team are not going to be set up to be able to iterate fast, then you're going to add to that marathon that bottleneck of, no, you have to depend on this team to release this.
Speaker: You have to depend on this team to be able to change the way the system operates. So that's one big thing for me. If the answer is no from stakeholders or from, you know, whatever the system, because it's really complicated and so on, that gives me, in an initial data point, like, okay, the problem has to be really big For us to like step in, invest a lot of time and like kind of be in this handover mode and then maybe come back in the handover mode, I really prefer that sort of, you know here's a box, we sit in here, do whatever you want in the box, but we can move our faster. we have So for me, speed at some point is is a key.
Speaker: um Second, not only it has to be, could it be long-term because it's an optimization goal? What about the problems that cannot be solved today? So if the problem is categorized as unless we have data scientists running something, this cannot be done.
Speaker: And that's another data point for me in the sense of, For example, hey, there's a system that, an auction system. So I run auction system through my team. And there's a system that the deployment mechanism is really complicated for a human to to click buttons around and and set what bits do we want in different item levels.
Speaker: There's thousands and millions of item levels. It's basically impossible for a human to run this sort stuff. If we come in, that unlocks this amount of potential value. And even if it's not the biggest one, it's like something that no one else can do. That's net value. So that's another big point for for me.
Speaker: And so if we try to combine those, we we go into problems that already are working the system. not For me, nothing is broken. but when i When I hear sometimes people say, yeah, the system is broken. Well, it's not really broken. It works. Like it's in production. It works. It dies.
Speaker: What you're referring to is that it could be better. Happy to be there. That's where that one data point. And the other is like, this system doesn't exist. Can we do it? Yes. happy to be there So that's the two things are very important.
Speaker: Just to finalize, to like to finish my answer, those the data points that I normally try to use for, is this important or not? Also have to be surrounded with, are the data scientists surrounded with the with their right tools and elements? then so normally this comes through, do we will we have engineering support?
Speaker: And a POC could probably work without some engineers. Sometimes you could do some analysis, some position making, some cutting corners. But if we're going to be, but I don't see it in today's world, if anyone works on data science, on ML data, like production grade data science, not product data science, which is super useful, but you can live with deep a analysis with no No, no, like ML production heavy, low latencies, millions of users impacted, you need engineering. I cannot see those two teams not working together.
Speaker: So that for me is also, i repeat this because it's normal that stakeholders might see the two disciplines as different disciplines that simply work together.
Speaker: So i there's always a journey until stakeholders understand that they should be together. to repeat and repeat and repeat and repeat and then the results speak for themselves. But it's not a given that everyone understands that for me these two teams literally should work together.
Speaker: So that's always like the pitch that I have to say. we have engineers? Who is going to work with us? so on.
Speaker: I suppose obviously that's um ah particularly a product of Skyscan as well in the environment that you have. you can the the The changes you can make to ah an ML product, even small, can have a huge impact across such a large customer base. But you need the the robustness of the engineering to make sure that then the changes are are production production ready.
Speaker: um That's really interesting, Jose. I suppose one of the the big things that's obviously come and affected this this space and and do all the whole industry and um is obviously the the birth of of Gen AI and sort of ai tooling that has come around, which has dramatically increased the capacity that someone can can output. um How has that changed?
Speaker: you how you manage and how you think about scaling a ah team?
Speaker: I think you're we're going to have this question for for some time because think it's super exciting to be living through this dramatic change, like everyone is kind of starting at the baseline, the same baseline.
Speaker: And so there's no playbook on this. Like how, how, who has lived five years ago a change where skills are basically apparently being obliterated by this system.
Speaker: There's, there's nothing out there. So for me, um
Speaker: we, I personally want to focus right now on giving the teams space to experiment in because I don't know what is the art of the puzzle. I'll give you an example.
Speaker: One of our systems, and because we work on the options side of things, we work trying to put bits on with Google. To do that, normally you need an API.
Speaker: An API in the old world would have been built by engineers, so we have to go through that, through them, and then requirements, and then work together and i get that done.
Speaker: Now, this a quarter, we needed big changes to to the API so that the challenge was, hey, could we build the API?
Speaker: We've got Claude, we've got the documentation, let's just build the API and see what happens. And it worked. It worked. and We completely bypassed an engineering team. We built something that is really useful.
Speaker: So that's the big highlight. We enabled ourselves we to scale our system massively. But here's the low light. We don't understand the API code. So what do we do with that?
Speaker: Are we talking now, are we doing an am ownership problem? So whatever we gain in speed today, it's going to slow down because we have to go back to the engineering team and say, now you own this.
Speaker: And probably the code is not correct and you're going to change it. So that already adds slowing down. The contrary also happens. The engineers would have probably in the past come to us for analysis.
Speaker: And now they're doing the analysis. But what if the analysis says that there's an impact of 10 million? Would an engineer be, I know the answer because they don't, would the engineer be happy with the, yeah, the analysis that I kind of don't understand too much says that I should be doing something that might have a 10 million impact.
Speaker: They probably would be like, how can you double check this? So that's the paradigm that at least in a bigger company, I guess startups can just go in, fail, go in, fail and and learn a lot quicker.
Speaker: But we're learning in a slowish, but steady-ish way because these problems you only learn by doing.
Speaker: And this is just one example of of what AI is doing. It's enhancing us to do a lot of things, But when you actually like look under the hood, it's difficult.
Speaker: and What I'm not saying is that...
Speaker: So this is definitely going to change the way we operate as a data science team within and the remit that we are experts on. I gave you an example where we are not an expert on this. So it's enable to do enabling us to do things that...
Speaker: who just wouldn't have done at all zero.
Speaker: So that's really good. What I do see us and I'm pushing the team to do is look, we're experts on building models, evaluating outcomes, evaluating predictions.
Speaker: And that was building the model was never our net hand. Can we automate that piece of our net value add because LLMs can do way more things than what we can do. They can be working overnight, building 30 models where in the past it's you and your screen for eight hours and you're kind of done.
Speaker: So there's a big shift there, and which we should be able to control a bit more. and And that's a lot of exploration that was hoping we we're we're trying to do. um but it's a It's an unknown, untrusted area.
Speaker: How do you do it at scale, robustly, with care, quality? Do you go and break it? Do you not? it's i you only have You can only do it by purposely allocating time for the team to try things and give them a challenge.
Speaker: The challenge can go really, really, really bad. and I also have examples of that where we built some models that
Speaker: over a weekend the system built five models that impossible for us to do in the past like absolutely impossible so we were super excited incredible gains offline and we're like okay now put it to production it took us three weeks because the code base is was a mess but it was not because ai is not doesn't know how to decode it's we haven't learned how to put ai in the direction of Not only do you can you explore, but the way to then put that in production is through these templates, these formats.
Speaker: We're learning a lot. to dax piece I think that's the the thing that seems to be am the from from articles and people I've speaking to. The LLM's are great to at execution, but if you haven't got the the context and they don't have the context, that's where they they can really really slip up.
Speaker: I suppose on that, Jose, la what is... What do you see as the new sort of skills that really are going to help someone excel in their career? Because yeah as we've sort of mentioned, like the execution point is is and the actual sort of writing of code is is arguably sort of decreasing, but there are definitely other skills which people need to bolster to be able to still still deliver. so yeah how How do you see like the the change in in the skills profile of of individuals and and teams as as we develop? And is it something you're thinking about right now?
Speaker: um We are thinking about it now. i I don't have super clear answers because literally we're thinking about our hiring processes, right? Where in the past, imagine that company was doing like a coding interview.
Speaker: Is that relevant anymore? Maybe it is, but with an AI tool. But then how do you know that the person is using the AI tool to not create sloppy code?
Speaker: So all these questions come to mind when we're reviewing this sort of of of things, because if we want to hire people, in the past, we had very clear definitions of, you know, raise the bar and what do we really need from someone who's technical from someone who's going to lead people um but now that's changing a bit so we have to we have to figure that that out and in terms of who who do i see maximizing these sort of tooling um
Speaker: the those who with a critical mind and uh and those who want to who were always like pushing the the the need for i want to create something i want to learn something new ai is going to 10x that and i've experienced that myself because you know in the past if i wanted to to learn um I'm just going to give the example I wanted to learn Plotly, which is just a visualization tool using Python code, which is very cool.
Speaker: Well, it took me a long time reading all documentation, maybe reading books, With AI, I can expect that. But always, i have always if work if I'm critical, I'm like, okay, no, but always challenge, always challenge.
Speaker: So that's super, super important because if not, we are already seeing the that those who just accept PRs, there are a hundred files to be changed. It's becoming ungovernable. so and we're going to have to balance the speed with quality. You can only do that if you're critical, if you're like disciplined, and and if you see this as an opportunity and not as something that's going to hinder you. So I think AI is is pretty amazing, honestly. I'm very, very positive of what it can bring. The amount of
Speaker: The amount of software engineers today in the market, I think is actually the demand is going up because who's going to govern this code? and Only the experts who have done it in the past and can actually understand what is happening. right Same with and machine learning. Who's going to understand that a system is trustworthy, robust, non-biased? I can only imagine an insurance where it's really regulated or banking that iss really regulated.
Speaker: You can't let AI get loose because regulators are going to ask you, why did you put this price to this 90 year old lady? Well, you have to be able to explain that. So only experts can govern the guardrails of the system. But I guess the system is just going to expedite everything.
Speaker: But yeah, those who are going to see this as a simple like a chatbot, not too sure. I think we really need to lean into and how did you download your knowledge into these systems so that they work for you on the areas that you were good at, but the real net value was always pull this problem, which is undefined, understand what exists in the company, how the company works, and then output something that you know that's the real metric. How do you define a metric? How do you define success?
Speaker: What happens in between was the cool part. But the cool part probably is something that we could probably ultimate in the next few years. So we'll have to see how people are up to that, and myself included, by the way. Yeah, I think what you said there is definitely what we're seeing. I think if you're in a position where you're you're senior enough, you have a excellent fundamentals. You know what good looks like and you know what you want to build from the get-go from when you' you when you're speaking to a stakeholder and you you've got your plan. i think
Speaker: If you're in a position like that and you know what you want to go and execute on, then that's when leveraging Clawcode and the rest of it could be really powerful. I think it's at that entry level, mid-level, um engineers, scientists, data professionals, which are ah probably struggling because they they don't know what to review. They would tend to find them find find themselves along that execution path, whereas yeah they that they don't know what good looks like. so I think that's probably where the I'm seeing the the biggest the biggest gap. Yeah.
Speaker: Still, like if you're that junior, And where in the, think about it also this way, in the past, you would have worked together with a senior and you would have worked at a pace of slow trying things, the senior reviews you get to learn.
Speaker: You could still probably learn a lot if your companion is AI because you're literally asking it 10x the amount of questions you would ask your senior. Why does this work this way? How does this compare to this? why Why can't I use the other thing and give me pros and cons? And you're learning really enhances because you've got in front of you a lot of examples and scenarios that you're probably either you cannot cover all of them with the senior or maybe you can't have as much interaction with the senior so i still see those juniors who who really revoke ai to to push themselves i think they're going to be uh brody
Speaker: better seniors that in the in the future than was. So we'll see have how that goes. Another interesting point that I wanted to mention is I've been thinking on how tech leads, managers operate, because what is happening seems that the industry is going systems that agents are like, sorry,
Speaker: systems that are um controlled by agents. And each agent has a role and each role has a definition of you know success or not. If you think about it, that is literally what managers do.
Speaker: They have to distribute work, define what the definition of done is within a timeline, define success criteria, not on their own, but with the team. But that literally they're responsible for not one system, single system that does one single thing. It's like a full system with four different projects.
Speaker: um And I have a feeling that managers are going to be in a really good position if they can get their hands on agents, as I've been doing for my own personal projects, that you're like, actually,
Speaker: I can build a team with this because this is what I've been doing. It's just that I have to codify it and code it my knowledge and and so on. So I think that's something that I want to bring out because maybe people managers out there might feel, yeah, but all these things that are happening are for the ah for the individual contributors.
Speaker: and i am black And I understand the the the thinking. But I think there's a missing point there. I think that if if we rethink the way we've operated as managers, actually, I think puts us in ah in a really good position to really build things, not for production grade for the company, but like enhance everything to a level that you couldn't do in the past because you were depending always on the, e all the time on the engineers, all the time in the data scientists.
Speaker: can build agents to review the quality of the work on explaining how something works and I can make better decisions because I can go to stakeholders and I have information I didn't have in the past. And so, yeah, it's just an extra point. I don't know if many people are thinking about how running teams is going to be pretty similar to running agent teams. So we'll see.
Speaker: That's a really interesting point. I haven't heard of managers speaking about that. I've heard of principals saying yeah now they've you know they've got agents that are acting like mid-level devs, but not how you've described um agents. how is Is this something that you've started to sort of trial out at Skyscanner? Yeah. like what Where have you seen the biggest victory in that space?
Speaker: Well, i just I've tried some things. and The way that I've always learned is learn by doing, so i haven't yet tried it at scale from like people management and so on. But the first thing I've wanted to do is let me see if I can run a team of agents that does 80% of my newslet a newsletter work because I really enjoy writing.
Speaker: and But an article take took me quite a few hours to write. and I've got over 60, 70 articles published and I'm like, well, that's a lot of knowledge on how I write, how I structure my thinking, this is how my style of writing, which is very pedagogical and so on.
Speaker: and Can I get a team of agents that get me 80% there? Not fully release it, but... Can I do that? And I did it. So I still think I retain the same amount of quality that I had in the past because it's me still tweaking the 20%. I don't like this flow. I want to change this other thing. I want to create diagrams because I'm not doing the diagrams that I have in my head.
Speaker: But there's a lot of things that I was doing in the past that this is doing for me. Build that through agents that do researching, do loops of reviewing, formatting. I have to, I've had to define rubric criteria scores so that ah the agents understand that this is good enough or not.
Speaker: So I've seen it. I've seen that it can be done. And through that, now I want to bring it back to, to, to my work, to the team. Sometimes it's not me, it's just even doing the work. It's telling the team the art of the possible. It can be done.
Speaker: I know now that if it can be done, I can um push the team to do it. If I hadn't done it, I don't know if this is possible or not. So I guess that's right now, given it's, I'm going to say early days in the Gen.AI world, some people might say, like oh, it's been over two years. Like, yeah, sure. But in the corporate world, all this change is very early days in many aspects.
Speaker: So...
Speaker: If anyone is giving it a try, they should they should fail on trying things and and to learn on their own. I don't think managers today can live in the world where where where they didn't touch code in the past or these tools in the past. I think you have to allocate X percent of your time to do something which is not going to impact the team, but that's going to get you to learn.
Speaker: and So yeah, I can see these agents, if I build them in a way that can help me with calibrations on on people management. When you've got a very big team, sometimes subjectiveness flows into the calibration. Can we have documents that are the baseline truth and that can tell me, well, you know what you said here is actually not exactly what the guidelines say. So are you sure this is right?
Speaker: Are you sure are you benchmarking these two people correctly? Because this person seems if they're doing this way more, but you put the same rating. So hopefully they enhance the way that i can critically think about things.
Speaker: and yeah I have weighed lots of ideas on lots of ideas on on the age front, but yeah ah like do things The big takeaway is yeah go try experiment, um unleash your curiosity. We've said it many times on the pod before. I think yeah one of the the best attributes in in in data is to be curious. um so like yeah What you're talking about is just going and experimenting, trying, failing, um and pushing what the art of possible.
Speaker: is and Then you know you you use that and and what you've done personally to to empower your team and to get them to go out and and do the same is is the the message I'm i'm hearing there, Jose.
Speaker: The role of a manager is is not being threatened by by ai but um can be can be enhanced. Yes, totally. Awesome. We're coming to the end of the show, Jose. i suppose We like to end by looking back and some some reflection. so I suppose, what what are some of the what's the biggest lesson that you wish you'd know that you know now that you wish you'd known sooner in in helping you you in your career?
Speaker: That's a good question.
Speaker: That's a really good question because I always go back to the to the shift from IC to people management, but that that is not necessarily probably the thing that might have impacted me the most.
Speaker: I am going to answer not from a, I made this mistake, which normally is one of the things that helps someone grow. I actually am going to make it from, I've been doing this forever and it has worked. So I don't, so hopefully that. The change is not broken. Keep keep keep it the same. and Although, you know, what got you here will not get you there, which at some point I might hit that ball. We'll see.
Speaker: Which is, tied to the curiosity that you were saying, but and not really on being curious. It's more about if there's an opportunity, just say yes to it.
Speaker: Even if you're not ready, just say yes to it and see what happens. What's the what's the risk, right? And and night that's the mindset I've had. So for example, and three examples I can share. one I always lived in Spain and when I moved up to the UK, I lived for years in London and a couple of years down in the south.
Speaker: But when I made the move, I was working in Spain. It was the middle of the crisis of of jobs and so on here back in Spain. And I moved to Spain following my girlfriend, who is my wife, mother of my kids.
Speaker: But that jump was, i went there without jobs. I went there, like, see what happens. What's the worst thing that I can do? That I have to come back and come work back again in Spain? And it was the best decision I've ever made.
Speaker: Another example, I never had worked with data really. i I touched some code for through robotics. But in the middle of the big data world, which was the big data era back in 10 years ago or a bit more or 2012, 2015, I can't remember.
Speaker: and was like, well, let's just see if I can find a job in data. I've got no idea what I'm going to do, so I'll just learn it all the way. It worked really well. And then, of course, they jumped to people. And so, I don't know, I guess it's easier said than that done.
Speaker: But that's the mindset that I that i want to keep. If there's a challenge, pick it up. i'm and see how it goes. That's for me the the the biggest the biggest stake I think that's great great advice. um yeah you You need to start jump into opportunities and I think you solving jump opportunities, solving problems, whatever it might be, whatever whenever you get given something like that, if you if you can make an impact and you can do something with it, then yeah you get recognized. i think that The people that excel are the people that solve the most amount of problems. um
Speaker: Excellent. final thing, Jose, is um what's your advice for other data leaders as they're scaling teams in this sort of ai era and and how how what how are youth yeah what what advice would you give to to other other leaders in in in how to drive impact and and in this new world?
Speaker: The most important one, as always, is data quality. I guess all data leaders know this, but it's just reinforced. they it doesn't really matter if you if you drive ah a ferrari if you're if if your wheels are or your tires are like a bit uh edgy and sketchy right so um data quality but that's aside from that which is like the always the baseline mantra and if you're starting or you want to get your team to do things i wouldn't as i said earlier like
Speaker: Go fast on trying things, make the space to try things. Because you're only going to learn what works and what doesn't work through through that. Don't try to to figure out the best system.
Speaker: It won't work. and These things are changing every month. So it's better to start now, get yourself a challenge or problem, get the team to do something about it. Let them fail, they will fail. They will find some good stuff and they will break five other things.
Speaker: But hopefully those will reinforce the, okay, at once you've got those five, six, 10 learnings, that should inform enough a plan on, okay, this is what we're going to do with AI as a team. We're going to increase speed or we're going to increase quality. We're going to focus on this problem. Maybe it's only data analysis as a partner.
Speaker: Maybe it's a system that actually runs things for you. You can only learn that within the tooling that you've got ah around you, the people that you've got around you, how your company operates, but please do not stay put figuring out if you have to build the perfect system or if someone else is going to build a plugin or a package or a service that you can buy, you'll be late to the party.
Speaker: And just lean in, do something. As long as you understand that that thing, something at some point will just not work, great. just go for it. Right now I don't have any better advice than just to give it a try.
Speaker: Trial, experiment, fail, off be curious. yeah In other areas, it's easier to understand. There's a playbook, there's robustness, machine learning, and MLOps.
Speaker: There's a lot of things out there that you can follow. Blueprint. There's nothing. There's no blueprint today. There's some. but And therefore,
Speaker: Jump onto the opportunity, give it a try, see what happens. what What could what could go wrong then? You haven't done anything, but you've learned a lot? Great, let's take it forward. Awesome.
Speaker: Jose, it's been a pleasure to have you on the show. Thanks for sharing such great insights and yeah your journey into data and obviously leading large teams at Skyscanner. Thanks for for joining.
Speaker: Thank you very much for having me, Harry. Excellent. That's it for this week, folks. We'll see you in a couple of weeks. Thank you and goodbye. hi everyone just a quick one from me if you've enjoyed today's episode i' be so grateful if you could hit that follow button or leave us a rating even better past to show on to a friend who might also get some insight from it it really helps us grow the community and continue to share amazing conversations I also wanted to take a minute to talk to you about Cognify.
Speaker: Those of you that don't know, Cognify is the leading recruitment partner for modern data teams. We help some of the world's best organizations scale data and drive real value from the hires that they make.
Speaker: you're thinking about building a team or making a hire and you're struggling with talent or just want some insights on the market, then I'd love to jump on a call with you and tell you a bit more. Equally, if you're looking for a job and want to find your next dream role, then reach out to myself or any other Covenant Fire team.
Speaker: We'd be happy to see if there's anything on our books that we can help you with and give you general advice on the industry. Finally, big thank you to Omni, this season's sponsor. If you'd like to learn more about the AI analytics that Omni can deliver you, then check out the link in the show notes or come speak to me. i can happily point you in the right direction.
Speaker: Again, thanks for listening and look forward to seeing you a few weeks' time.
