Description:

This week, Dan Neumann is joined by Tarik Smajic from Machine Learning Team and by Justin Thatil, an Agile colleague. Justin and Tarik are both Scrum Masters but Tarik’s work is in Artificial Intelligence or Machine learning. In this episode, they explore together with Dan, the differences and similarities between Scrum and AI as well as how they complement each other by sharing valuable case examples.

Key Takeaways

• What makes AI Teams different from the Scrum framework?
  • Scrum helps to reduce complexity, and certainly, machine learning is a very complex subject.
  • Scrum is a way to start establishing norms in AI teams.
  • In the traditional software development life cycle, there are established phases in order to build software and this includes an exploratory aspect.
• It is more than data.
  • We give the client for free only the data that we are willing to give them, but there is even more data that you can think about that in the past was considered waste data.
  • There are patterns that can be found in data, that is why it is called predictive data.
  • We used to want all the data available but we started to figure out that not all that data is needed, and in case it is necessary to synthesize data that has any predictive implication.
• The beautiful dance Scrum proposes:
  • Scrum works by just enabling the particular accountabilities to do their thing, to be empowered to shine in their field of action.
  • Once you stop trying to solve problems using predictive and prescriptive analytics and start understanding where the value lies and where models need to be built.
• Case: A Team faces a product challenge.
  • Let the Team have the time to research (but it can’t be forever).
  • The Team needs to go through one cycle to establish a baseline.
  • It is better if you adopt Scrum, starting from scratch.
• Sprint reviews in AI:
  • The race to the minimum viable product can look like looking at your data asset and learning from it.
  • Tarik shares several examples.
• It is important to establish what the development phases look like while the ideation and intake Team handles the values assessments and figures out what use cases there are; prioritizing them is the product management Team’s work. Then the research aspects follow; you want the engineers to build the pipelines and then do the testing.
• Scrum of Scrums:
  • Tarik shares how they use one Scrum of Scrums on a weekly basis that only lasts 15 minutes.
  • A necessary question to ask during a Scrum of Scrums meeting is: Am I putting anything in anybody elses’ duties?
  • How realistic are the expectations? The meeting produces a forecast of what can happen.
• Application of Scrum in the AI and ML worlds:
  • Tarik shares his experience.
  • Everything in Scrum is iterative.
  • There are three phases of learning something. It takes a while to master things; patience is required.
  • It is OK to bend the rules, you don’t have to do it all by the book.

Mentioned in this Episode:

Link  to a previous episode

Things Done: The Art of Stress–Free Productivity, by David Allen

Visit the website and catch up with all the episodes on AgileThought.com!

Email your thoughts or suggestions to Podcast@AgileThought.com or Tweet @AgileThought using #AgileThoughtPodcast!

Trainings & Courses

AgileThought’s team of experienced practitioners and certified trainers teach classes on a variety of topics related to agility, scaling, and technology.  We offer both introductory and advanced classes onsite or remote, many of which help participants obtain a certification. These classes introduce participants to new mindsets, tools, and practices that can be used immediately.

Transcript [This transcript is auto-generated and may not be completely accurate in its depiction of the English language or rules of grammar.]

Narrator (00:03):

Welcome to Agile Coach’s Corner by Agile thought. The podcast for practitioners and leaders seeking advice to refine the way they work, van pave the path to better outcomes. Now here’s your host, coach band, agile expert, Dan Newman.

Dan Neumann (00:17):

Welcome to this episode of the Agile Coach’s Corner podcast. I’m your host, Dan Newman, and today I have the pleasure of having not only one of my agile thought colleagues, but I have an external guest as well. And so I want to welcome Tarik Smaljic a scrum master for a machine learning team. Tarik, thanks for jumping into the podcast waters with us.

Tarik Smajic (00:39):

Thanks for having me, guys. I’m, I’m honored to be

Dan Neumann (00:42):

Here. Absolutely. We’re, we’ll be excited to hear about Scrum Mastery in a different context than we often talk about it with. I dunno, for lack of a better term, more traditional software development. My agile thought colleague here is Justin Thatil, a frequent participant here. So I appreciate Justin, you joining as well.

Justin Thatil (01:04):

Yeah, thanks Dan. Thanks for having us and having the opportunity to explore the topic that I’m, you know, have got a fascination about. So Terry, I got the pleasure to work with him and we struck it off just, you know, on the side, just having a conversation and you know, we’re both Scrum masters in different realms, right? So one of the things that’s important as Scrum Masters is to evaluate the context that you run, right? Your teams, how they operate, you know, what a data actually deliver. So this particular topic, you know Terry is part of a machine learning artificial intelligence realm team. And I’ve, for, for one it’s a fascinating, I’m a fascination topic for, for me, you know, just it’s, you know, the cutting edge of technology, right? We’ve got a ton of Silicon Valley organizations that shine at, at doing this. And just a sheer notion that there are Scrum masters out there that are able to bring on value for a, you know, very different mindset in, in the sense for, for their team members to, to still deliver value is fascinating. So, looking forward to our conversation. And hope everyone listens. Enjoy, enjoys the listening to us as

Dan Neumann (02:32):

Well. I trust, I trust that they will. So, Tarik, maybe you could help connect the dots. I, you know, I’ve been scrum master for teams filled the scrum master accountabilities to use that new term, a lot of times it’s, Hey, we’ve got a backlog of, of features we need to create, and we, we kind of pound through that list of backlogs, you know, inspecting and adapting as necessary for that backlog. What makes machine learning or AI teams different, and how does that maybe connect to the Scrum framework?

Tarik Smajic (03:05):

Well, so I, I think it, it, it’s funny, right? Like one of the selling points of Scrum is it helps reduce complexity a little bit, right? Like, you want to use it for things that are complex. Well, I mean, I don’t know of many things more complex than machine learning and, you know just advanced analytics in general. So it, it really provides like a really good cadence to just kind of a start establishing your norms, right? Being of sh it gets you, it gets you through the shoe really first really, really quick, right? Like the first step, just you apply it by the book and you, you, you can pretty much tell at the beginning like where it does and doesn’t fit. So I’ve had the opportunity to seed the removal of some of the more prescri prescriptive elements of Scrum get changed in 2020 while in the midst of adopting Scrum with the team that I’m met with, and then having it actually make more sense, right?

Tarik Smajic (04:10):

To answer your question of how it differs from your traditional software development life cycle, right? It, it used to say, right, like an increment is something that’s releasable and testable. The, the problem is in, in machine learning there, you don’t always output working software at the end of a sprint. Sometimes it’s insight. And so having had that piece removed from the scrum guide actually made it even more applicable. But I think on the general high level, I mean, it’s, it’s one of the most perfect things. And of course, nothing is perfect, but it’s, it’s the closest thing to perfect for, for this particular complex environment.

Dan Neumann (04:49):

Mm-Hmm. So, so what you’re describing, I think is with machine learning, you’ve, you, you’re effectively trying to connect the dots, I think, with how do we take a bunch of data that may or may not have obvious connections to itself and get better knowledge, better insights from it? And so your increment using the scrum term isn’t a bit of working software necessarily. It’s an, it’s an insight or a learning about that data in all those connections to it

Tarik Smajic (05:20):

Of of course, yeah, it’s right. And that, that, that’s when the differences start showing up in, in your traditional software development lifecycle, you’ve already have very established phases, right? You, you understand what your ingestion or your intake intake phase looks like, your design phase, it’s, you already kind of have a well-established way to build software. We’ve been building software for a very long time. This has a lot of that exploratory research aspect to it. So it, it tends to stretch and compress some of the phases based on the complexity of the space that you’re in. And I think that’s where, where being agile and having the ability to, to apply Scrum in a bit more looser way really helps.

Justin Thatil (06:06):

One of the parts that fascinates me in, in this, in this realm you know, it’s, it’s, we, you touched on it, the data, right? And there’s a coined term that I ran across here, and there’s an entire industry, Silicon Valley, again that’s shining on this. And it’s surveillance capitalism. And essentially what these companies are able to do with you know, what we’re calling the data, it’s, it’s more than data. There’s, it’s, there’s, you know, the data that we can think about are you know, what we give them for free in a sense are the, is the data that we’re willing to give them potentially your name, your address, phone number. But beyond that, what they’ve been able to do is to actually expose even more data that you don’t can even think about, right? What in the past was considered waste data.

Justin Thatil (07:03):

Digital exhaust data exhaust is another firm that’s been used in the past, but what companies have been able to realize is there’s patterns that can be extracted from, from this data that essentially harbors rich predictive data, right? So in a and then this can be used in different ways, right? So, so in a security sense, right? So if you’ve got ton of data and patterns of how people use your, your app, right? So let’s say it’s a financial app and, you know, from a financial organization’s perspective, you know, you wanna be secure, right? So the models that essentially highlight, hey, these are the particular patterns, behavioral patterns that we’re able to see from fraudsters, for example, you know, from a security perspective, that’s great. Hey, we’ve got the ability to pinpoint particular behaviors and patterns that more, more than likely is indicated that there’s a fraudster trying to access your system. So, you know, so building on top of that, and then I guess being part of a team that essentially doesn’t know what that pattern is from the get GOs essentially, right? It’s, it’s that discovery phase I think you were talking about. That’s, that’s really interesting. And how do you, how do you model that? How do you create that environment for your team to, to enable the, that kind of discovery? That’s very fascinating. . Well,

Tarik Smajic (08:45):

The Scrum enables that. It’s, it’s, it’s funny, right? Like, people often hear me talk about Scrum and I’m, I, I, trust me, you don’t have to put a gun to my head to get me start talking about Scrum. It, it look by just enabling the particular accountabilities to just do their thing, right? To be empowered to, to shine in their particular role, accountability, whatever you, whatever you wanna call it. This beautiful, like symbiosis happens. Usymbiosis might not even be the right word. It could be just a beautiful dance, right? Uempowering your product owner to be able to drive the priority of a product, right? Like you, once you start trying to solve problems using, you know, predictive and prescriptive analytics, building these models, it’s understanding where, of course, you know, nothing is free. So you have to understand where a value lies. Like where do we wanna build models?

Tarik Smajic (09:37):

Like what, what, you know, where can we actually create impact? Doesn’t always have to be monetary, right? It could be time saving, it could be, you know, just more efficiency. So as long as you empower your product order, be able to drive that priority, it allows the team to see that vision, right? The team that’s, that’s actually doing the work, doing the research to see the vision of where they want to take this thing. And so that allows them to, I guess, pursue those right thought tunnels and go down the right rabbit holes to find the data. And then on, you know, to kind of expound a little bit on what you’re talking about I’m a huge proponent of Gartner. This is my, I guess this is my plug for them. They have a lot of really, really good research on it. And at their most recent key conference, they were talking about the difference between big data and metadata synthesized metadata even.

Tarik Smajic (10:32):

So we used to want all the data, right? We used to want every bit of data that was available. But we’re starting to figure out, well, yeah, there’s trillions and trillions of, you know, gigabytes even more netter you know, terabytes of data out there, and you don’t really need it all. You just need to find ways to synthesize really good metadata that has have the predictive implications, right? You’re, you’re a lot of research has been done into feature factories where you essentially establish these data sets that are just a combination of features that allow you a high level of predictability on those patterns. And then you let the model figure out the pattern. You don’t wanna figure out the pattern. We ain’t got the time or the brain power to pursue that. We let, we let the computer do it. Mm-Hmm. <affirmative>.

Dan Neumann (11:25):

No, that makes sense. So there were a couple of things you were, you were touching on there. So Justin, you were mentioning the, you know, the outcome might be inhibiting fraudsters it, if not preventing them, at least make it more difficult or less likely that they would be successful. And, and turik, you mentioned the symbiotic relationship, and I think that’s probably a fair term where you have the product owner to do their accountabilities, needs the developers to do their accountabilities, counting on the scrum master to take care of theirs so that that entire ecosystem can function properly. So I think that’s the right term. Maybe you could help connect to, if you do have a product challenge like inhibiting fraudsters, what does it look like from a scrum standpoint to take that through the lens of an ML team or teams that are dependent on each other?

Tarik Smajic (12:20):

You gotta let the team have the time to research. We like to say that our product is knowledge and our increment is usually knowledge gained. Retros are extremely powerful for research teams, I’ve noticed, because it gives you the opportunity to start discussing these things, right? Like, what are the things we’ve learned in the last spring? And it really helps drive that. So you can’t give them forever though, because we are, in essence, in a time boxed space, right? Like, no one wants a project to go on for 17 years. So how do you do that? So you start establishing what, what baselines look like, but in order to get there, you have to at least go through one cycle of doing this to establish a baseline, and then you just try to improve from there. I will say that based on my experience, it is a lot easier to adopt scrum starting from scratch than it is in the machine learning space, please, than it is to take a, an existing product and try to get whatever group is or team is about to take over it. Start practicing scrum at a high level. You go through the same growing pains. It’s just a lot less technical that I guess when you start fresh. I hope that answers that question, Jeff.

Dan Neumann (13:53):

Yeah. Yeah. It, it does. You touched on a couple things that I think are are important there. The, the value of the time box. I know I’ve heard, oh, I don’t know, people can Google why doesn’t scrum work for AI or whatever. And you’ll see responses like, well, it’s research. We won’t know until we know. It’s like, well, yeah, but at the same time, you need to know if you’re heading down the right path. You know, if you’re in an academic situation, you probably have a grant that covers your learning. They will want some output, and that grant will run out. If you’re in a company, it can’t be just, oh, just, you know, whatever in a year or two, I don’t know, right? I don’t know. Is is an interesting place to start, but you can’t stay there very long. You might put it within a box. How do your what do you do with sprint reviews kind of going off maybe a, as you were talking about that, the increment is knowledge, and you have stakeholders who are looking for periodic value delivery of increment knowledge gained. Do do your stakeholders kind of follow you guys with the learning? Are you speaking over their head? Like, what does that look like?

Tarik Smajic (15:04):

The the goal is to obviously take ’em on the journey. The other thing too, i, I wanna measure, I, I mean, if I had to sell you the pitch in a one liner, it’s, it’s the race to minimum viable product, or I, you know, minimum valuable product, right? What does that look like? Sometimes in, in this particular space, if, especially if for a company it’s new, someone is just starting out, a lot of the values generated by starting to look at your data asset, you start learning a lot of things about your data asset that you might have not known before, right? Whether you’re backing stuff up well enough, whether your lineage, your recency is good, right? Like, what’s the quality of your data? So a lot of the values gain there. So one of the examples is just by working with a hyper consumer of data as a team, right?

Tarik Smajic (16:04):

Which is usually an advanced lenux function, the business unit working with them starts gaining these additional fringe benefits of much cleaner data, much more valuable data, right? Because as we are using them to generate things that create predictive output, we can use just by going through that journey through that research and stuff, you, you’re starting to generate the data. Now, of course, you wanna get to the first functional model. And so one of our secrets is just, you know, we will limit our, you know, feasibility or like our research phase into this. We’ll, we’ll take what we know about it and then we’ll just start engineering features and seeing which algorithms perform better. I mean, if you’re just doing some, you know, I mean, and of course it depends on what you’re doing. Are you doing something like regression, like some kind of auto aggression? Are you, are you focusing on creating reinforcement learning agents, right? Like, is it deep learning? Is it all, there’s a lot of additional questions that either increase or decrease the complexity of the thing that you do, and then you have to also take that into account, right? Like, that’s the awesome part about scrum. It takes reality into account now to your sprint reviews. Mm. The output sometimes, Hey, we now understand the behavior of your customers better.

Tarik Smajic (17:32):

Let’s say you are a lollipop company and you want to sell more lollipops. I don’t know, I’m just making think of that. You have to first figure out how you wanna do that. Do you wanna do that by mailing people like little flyers? Do you want to do that by email marketing, right? Like, is, is that a marketing thing? Do you wanna focus on existing customers, non-existing customers? And then just, you know, going through that and going, Hey, you have the data to do this or you don’t. Here’s how we can acquire it. Here’s how we can build it. And then presenting what you’ve found. Hey, we’ve managed to load your data asset, you know, do some ETL and make it really good for this particular thing. We were able to generate some census data that allows us to now do more things.

Tarik Smajic (18:15):

Here’s where additional benefits come in. So not all of the value in advanced analytics is created by just building the piece of software. Some of it is what we like to refer to as French benefits. So, and once the business is along for the right, as they’re learning these, they can start seeing it and understanding it better to the point where, you know, we’re like two years into one of our engagements, and man, our stakeholder like speaks our language. We’ve established that lexicon, right? And it’s a learning journey. Everything is iterate. Mm-Hmm. <Affirmative>, including your relationship with the stakeholder, right?

Narrator (18:49):

Have a topic you want us to tackle. Send an email to podcast agile thought.com or tweet it with a hashtag agile thought podcast.

Justin Thatil (19:00):

Yeah. The lollipop lollipop example. I’m thinking, you know, another iteration could be, you know, when, when is the person likely to buy a lollipop? You know, it could be like, oh, maybe you know, they’re coming out with their kids from a movie theater and yet to have lollipops present right there, they’re more, more than likely to buy it, right? And, and there’s , you know, so that, it is just fascinating how, you know, this kind of data can be leveraged to, you know, to drive again, that surveillance capitalism. One, one thing Terry, I remember us talking about last time is, you know, I think you touched on it a little bit, the ideation phase, the multiple phases and how, how you had to leverage, you know, some of the aspects of and scrum together that made sense for this team. Wondering if you can touch on that a bit.

Tarik Smajic (19:52):

Yeah, so we, we just kind of start at, at the, we say team level, the department level, and view that as the main main team. And that’s broken up into subsets, right? So we’ve have, we’ve got your research teams, we’ve got your ideation team, we’ve got your, let’s call it your delivery team, which is your like final mile team. There are people that build your APIs. There are people that build your infrastructure that houses your data, you know, your data engineers and so on and so forth. Your pipelines for delivery due to the nature of what people do, right? Like the research team can be pure scrum. Now, do the two week sprint cadence, refine, do, do soft solve refinements, right? Like have, we like to do like actual dedicated blocks of backlog grooming where we refine, we get together and we start building new stories and refining stories.

Tarik Smajic (20:47):

You, everyone’s dream is m plus two, right? So we work, we work our way towards that. And then the researching the data science team, they do the thing, they do their sprint outcome reviews every other week, retro learn some repeat. And they’re doing great. I mean, we got with two teams, we have like a hundred, we 160 sprinted. The variance also, once you start doing it, because you’re not doing everything relatively, their cadence and their normalization actually occurs quite quickly comparatively speaking. And then through time they just, you can actually watch amongst other things. I know some people feel, well, velocity and estimation, so I don’t want to divide the house, but let’s just say you establish your own trend metrics and allows you to, to understand what the health of that team looks like on Scrum. To your point, yes.

Tarik Smajic (21:40):

That includes establishing what your development phases looks like, it looks like, and what team works on it. Your ideation and intake team handles the value assessments and you know, figuring out what use cases they are and prioritizing them. That’s your product management team, right? Then you go into the research aspects once the research team is done, right? Once they’ve built a model, you don’t want them productionizing it, right? Like you want actual engineers like your Azure engineers to build the pipelines. You want QA testers too, right? So you have your testing and engineering team handled the last mile aspect of it which is also if you do a little bit of research shows, that’s where most teams fail, right? That’s where most advanced analytic initiatives fail, is actually getting the thing to the person, like to the person actually using it. There’s a plenty of companies that can build machine learning models.

Tarik Smajic (22:37):

There’s not a lot of companies that can actually get ’em to their users and get ’em used. I hope I didn’t get lost in my ramble, but yes. So we have three kind of lanes of teams. One is pure scrum, and then we have two con one con bond team, and then we have one scrum with bond team. Because while the engineering team does have scrum work that they can do, which is plantable and plus two, they also maintain an entire environment. And we all know maintenance is not very predict. I mean, you can do preventative maintenance, maintenance predictably, but you cannot account for things blowing up and things blow up. So that’s their combine aspect of it.

Dan Neumann (23:18):

Got it. So kind of a, a combination of right, input, input queue and, and that that refinement as, as ideas are generated, they need to be validated. There needs to be some kind of deciding which of all the good ideas we’re actually going to pursue now, because there’s always more ideas in capacity. I don’t know that I’ve been in an organization that hasn’t said our, we’re resource constrained their term. It’s like, like every company is, that’s why you have to decide what’s important to do now. Cuz you can never, if you run out of ideas or have more idea, more people than ideas like that, it just doesn’t happen. It’s not sustainable to have more people than ideas to deliver. And, and so your ideation folks are validating those, your research folks are doing your proof of concepts or, or the pre-production version is what you’re describing, and then your dev folks are building that last mile as well as reserving some capacity for the oh, something happened in production. Gotta go figure out what that is. Very cool.

Tarik Smajic (24:25):

And then at the center of it all is scrum of scrums, right? Who wants scrum to bind them all?

Dan Neumann (24:32):

Okay. I was wondering how you coordinated those. So even the conbon ideation folks as well as the research team and the, the dev team they’re, you’re using a scrum of scrums to keep those all in sync?

Tarik Smajic (24:47):

Yeah, so we, we, again, we’re, we’re well past the half phase we’re entering re right? So we do one massive scrum of scrums between all the teams, and we do your usual, you know, hey, this is we, we do it weekly. And the reason we do it weekly is most of them are on a sprint cadence. So the last thing we wanna do is just add an additional meeting and restricted daily. So we do, and believe it or not, it’s a 15 minute scrum of scrums once we, which is amazing. That’s nice. That’s what I’m most proud of. But yeah, we do the usual, Hey, this is what we’ve worked on. This is what the team’s worked on since last time. This is what we’re gonna work on till next time. I, you know, is anything slow me down? And then my favorite question, I think, I believe every scrum, scrum should have this.

Tarik Smajic (25:33):

Am I putting anything in anyone else’s way? It’s like a preemptive, Hey, by the way, there’s stuff coming down the pipe. I’m probably gonna toss something on you guys’ board. And then we, we like to have what I like to refer as a three-legged tool sessions where we, the, the, the product lane, the, you know, framework lane, the scrum lane, whatever you wanna call it. And then the development lane, like the three of us meet and we talk. It’s, it’s, it’s your true planning session, right? We cheat a little bit. We try to water, we, we try to sneak waterfall in there a little bit because you, you need to guess at some, you need to guess deliveries, right? People, people don’t, don’t, after a certain level, people don’t like hearing well at some point. So you have to give some kind of forecastable stuff. And the way we do it is we update the real, how realistic that data is. Once once to twice a week. Now we have these sessions and we actually go through it and we, we try to keep roadmaps milestones that we’re looking for and Go ahead.

Dan Neumann (26:37):

Sorry. And you used an important term there. It wasn’t the commitment, you said it’s forecastable. Here’s our forecast based on what we know. And, you know, maybe it’s probabilistic within X percent probability. We think you’ll get this in this time range. But yeah I don’t know. Won’t get you very far. I don’t know. But we think it’s in this range with a certain probability that’s a much more valuable type of uncertainty.

Tarik Smajic (27:03):

Yeah. I I call it the fuzziness factor. Anything out more than two sprints starts getting fuzzy. The further out it goes, the fuzzier it gets. I like reflecting that by reducing the size of the font on the, on the timeline. So it’s further out it goes, I just make it small , I can’t read it. It’s, it’s a wild guess.

Dan Neumann (27:23):

I love you get the Photoshop, you can start pixelating things as as it goes out. I, I actually kinda like that idea. I think I’m gonna do that.

Tarik Smajic (27:30):

It gives it a more dramatic feel.

Justin Thatil (27:36):

I used to refer to it as the cone of uncertainty, but I like, I like this pixelation aspect.

Dan Neumann (27:43):

I love that. I love that too. So we’re getting towards the, the back part here of, of our time together. And I’m wondering if you could help put a bow on kind of the the application of Scrum in an AI or an ML world and, and maybe there’s some takeaways for folks that aren’t doing AI and ML as well. If they’re like, yeah, but we don’t do that. I bet there’s connections here we need to make for them.

Tarik Smajic (28:07):

So let, let, let’s do this my favorite way. Let’s let, let’s retro it. All right? So let, let’s retro my my experience trying to adopt Scrum four years on a new team lesson number one, everything, everything in Scrum is iterated, including it’s adoption. And I, I think that’s the most important thing. People should not forget. People don’t become the greatest soccer basketball team overnight, right? Just because you put it on the jersey doesn’t make you a pro. So you gotta practice it, you gotta learn from it, you gotta iterated it something better. And that, and that’s, and that really, that’s the biggest takeaway, right? And I mentioned sh Hari, and it, it, for those who understand, they’ll, they’ll understand. For those that are not familiar, I, I encourage you looking it up. It’s just the three phases of learning or mastering something it’s important to start somewhere.

Tarik Smajic (29:01):

So starting by the book is okay, it might not be perfect, but it’s okay. It might not even fit what you’re trying to do, but you gotta start somewhere. From then on, you can start researching, figuring out what does and doesn’t work. Do your retros, right? Like, that’s also important. And just make it fit based on what you’re learning, right? It it, it takes a while to master things in simple things. When you start introducing complexity at these scales, it just take, it takes a while. So patience is, is a way to do it. And, and then I, I know some people don’t like hearing this, having ways to convey health of teams. I like metrics, but I like metrics. Not for performance purposes, but for indicating the health of the environment that a team is putting in their effort, right? They’re working.

Tarik Smajic (30:00):

Once you establish a baseline as a scrum master, you, you know, you know when something is up, right? The velocity in the vacuum means nothing, but you pair that with a few other things. You start looking at any possible impacts that might have happened to the sprint, you start seeing why there’s dips, why there’s spikes, right? So that, that, that’s, that’s kind of my thing. I’ve read a lot of papers on why scrum and ML fails. I think the takeaway from that is you don’t have to do it by the book. It’s, it’s okay to bend the rules a little bit as long as the, the goal at the end of the sprint is to deliver something that generates value based on the understanding of the product owner, the team, and the stakeholder. I mean, that’s why we have definitions up done, right?

Dan Neumann (30:53):

They, they are handy. Yeah, for sure. And we will, yeah. So everything’s iterative that, that sh Hari learning stage. And we’ll put a link to a podcast we’ve done in the past on sh Hari. The folks can check that out retro for sure. And then the scrum framework is so lightweight. Look, make, look for ways to make it fit what, what’s happening there.

Tarik Smajic (31:17):

And that’s the thing, they’re, they’re iterating Scrum too. The scrum got has been iterated, right? They’re seeing all the places as being applied and they’re taking the lessons learned to heart. I’m sorry, go ahead Justin.

Dan Neumann (31:29):

Absolutely. No, yeah.

Justin Thatil (31:33):

Now for me, this is, you know, again, a fascinating topic and it’s awesome to hear that, you know, fundamentals that we learn as scrum masters, applying Scrum working with teams can still applies very much regardless of, you know, the context and this context being the machine learning AI teams. Terry, cuz you was talking about metrics. One of the things that, that I love that you mentioned last time we spoke was around here what you were calling it, you were coining it just the wellness metrics, right? And the metrics tell me how well the environment around you supports you, right? So, so that kind of information, I love how you worded that last time. So just wanted to put that out there for our audience, you know, how important metrics are, but use them in the right way, right? Is it supporting you? Is it supporting the team? As opposed to if it’s performance oriented, then it’s just, it’s almost like an right. So good stuff. Thank you. Thanks for sharing. Yeah,

Dan Neumann (32:34):

No, yes. Thank you. And I this is where I hope I’m not putting you on the spot. Tarik Bec I don’t know if I mentioned, we usually ask people what’s on part of their continuous learning journey at the end of our calls. And given that you mentioned like reading Gartner articles for fun and the fact that you work in a job that literally has learning involved in the title with machine learning. I’m curious if you would share with us and the listeners what what’s on your continuous learning journey? What’s got you kind of curious these days?

Tarik Smajic (33:05):

Oh, lord, a lot. I, I I am a organizational psychology nerd. I that’s what drew me to being a scrum master. I I, I thoroughly, thoroughly enjoy working with people and helping create environments and helping people just achieve things. That, that’s kind of, that’s my advice. I, I really, really, I just enjoy collaborating and learning new things with people. So I have, I focus a lot on trying to better understand organizational psychology, liberating structures is another thing. I mean, as much as I’ve read about you, you never fully learn it. Every time you kind of glance through it, you learn other things. But currently the thing that I’m really deeply in is btd getting things done. It’s a really interesting framework. I I thoroughly enjoy it. It’s David Allen, he’s, I like calling it a me mental kung fu.

Tarik Smajic (34:00):

It actually goes hand in hand. But Scrum, I’m gonna butcher this and I’m sure people are gonna get upset with me for saying this, but this is my understanding of it. When you take things out of your brain and you write ’em down and put ’em somewhere and then order them in priority, you know, what you should be working on and what’s just taken up room in the brain, right? Same thing here. Once you establish a backlog and you order the priority, you know what you should be working on. So GTD is a, is the thing that I’ve kind of gotten into lately. I haven’t finished a book yet. I’m, I’m trying to, but that, that’s, if anyone wants to read something and have some time spare, I’ve encouraged that.

Dan Neumann (34:35):

That’s perfect. Thank you for that. I think I have the book, but I didn’t get it to done, so I I may have to,yeah, I may have to revisit that lesson and,see if I can get something smart out of there. Tarik, Justin, I wanna appreciate you guys hopping on and sharing with our listeners on the topic of Scrum Mastery within an AI and ML delivery function. So thank you very much.

Tarik Smajic (35:02):

Thanks for having me, guys. I appreciate it. Thanks again, Dan.

Narrator (35:07):

This has been the Agile Coaches Corner podcast, brought to you by Agile thought. The views, opinions, and information expressed in this podcast are solely those of the host and the guests, and do not necessarily represent those of Agile thought. Get the show notes and other helpful tips for this episode and other episodes@agilethought.com slash podcast.

The post Podcast Ep 15. The Beautiful Dance Between Artificial Intelligence and Scrum with Justin Thatil and Tarik SmajicPodcast Ep 15. appeared first on AgileThought.

This content was originally published here.