Viktor
00:00:00.000
It's gonna be very unpopular, Darin, just to be clear, people are gonna hate us for this, but yeah, that's my claim, the least value. you're Giving to the company is writing code and you should be very worried if you are one of the people in a company that literally just receives instructions and translates them as-is to code, you're in real danger.
Darin
00:01:29.442
So here we are, January 7th. This is our annual predictions for the year. If you listen to last week's episode that came out on December 31st, you saw what we pretty much failed. Uh, I'm gonna go ahead and roll the dice now. I think we'll probably fail again this year, but let's give it a shot as we
Darin
00:01:49.577
Yeah, I will go ahead and pull forward. Uh, the title for this episode is 2026, the Year of Discovery. And the reason why it's the year of discovery is 'cause that's what Viktor said on the last episode, so I just stole it and brought it forward. I do want to go ahead and call out a, a small state of the podcast item. the podcast has been going since May of 2019. It's a long time, in, I'll just go ahead and say it in 2025. I lost my job. That was laid off. And, uh, at this point I can keep doing the podcast, but we need some revenue. So starting with this episode or somewhere around there, we have had ads added to the podcast. I apologize for that. But we need to pay for stuff so the ads will stay in. However, if you would like a. Subscription, quote unquote, that has no ads in it. There is a link on our website, and I'll make sure that it's in the Slack workspace as well, that you can sign up for a $5 a month, no ads feed, period, just five bucks a month. That's all we ask. That'll help knock the ads off of some stuff. So that's the quick state of the podcast. No, it's not going anywhere, but there are ads. So if you get angry, I apologize, but Bills need to be paid. 'cause that was the nice thing about having a job. It helped pay for everything.
Darin
00:03:12.972
no job. I need some way to pay for it. So until I figure out what I'm gonna do next. Alright, let's get into the actual predictions for the share. Now, I will state again if you listen before you know that we'll record these episodes, the year interview, and the year forward at the end of November as we learned last year. Things can happen in a really fast amount of time. So in between the time that we're recording right now, the last week of November of 2025, and the time you're listening to this, January 7th, 2020 sixth, there could have been people put on the moon, people put on Mars. you know, the, a uprising has already occurred. The singularity occurred on the day after Christmas. I, I don't know, right? All these things could have happened. So we're working with what we have here. Just what we know as of today. Now, that being said, we have a handful of points. we know that AI is going to play a part in this, and we're just going to use AI here just as the starting point, but we're not gonna be on AI on everything. However, unfortunately or fortunately, depending on how you look at it, AI is gonna be weaved throughout everything that happens in our jobs this year, whether we like it or not.
Viktor
00:04:35.376
I think it's still not going to play a big role, role in operations in 2026. It has, it played big roles 2025. And I think that it'll not play a big role in 2026. Some kind of observability. Yeah. But kind of controlling what to do in the cluster, either, either in supervised or unsupervised mode, is not going to be big yet.
Darin
00:05:03.390
I would agree with that, just like I am not gonna let AI book a flight for me or buy anything with my money online. I'm just not gonna do it, not autonomously anyway. Uh, and that's what you were saying there, right? You're not, you, you want to, I heard you talking about this a while back. I believe it was on the Gemini CLI, unlike Claude Code, where it ask, Hey, do you wanna keep using this? Can I keep running this command without asking you Every time like a, a get commit or a get push, right? It's like those, you're gonna say, no, thank you. But get status is like, yeah, go ahead. I don't care. Just go for that one.
Darin
00:05:49.185
yeah, that's, so that's where I'm saying it's like, okay, what things could AI help with? Okay. Obviously reading state of the cluster, doing the, the observability part of it, like the true observability, not interactivity.
Viktor
00:06:04.256
I feel even more pessimistic view than that. Even in a more kind of supervised mode where I, where I approve absolutely every single thing. It's still not going to take off this year.
Darin
00:06:16.834
Why do you think, do you think, here's my, here's one prediction on my, something will happen that causes it all blow up and everybody's gonna pull back really fast?
Viktor
00:06:26.576
no, I think it's. You know, different reason how to put it at territory. Every technological wave in software industry, there are normally groups of people or types of experts that accepted fast and other groups that actually rejected and keep defending their no as strong as they can. now in many areas, I think that the office was leading the Yes, right? Kind of like containers. Oh yeah. Virtual machines. Oh yeah. Maybe my data center, but Okay. Eventually, yes. Cloud, those all sorts of things. Right? and like let's say the application developers were mostly kind of, uh, I dunno, I don't care. Um, why, why are you talking to me? this is now the other way around. I see massive adoption of AI and I think it'll continue increasing up among application developers. Not everybody, I'm not saying like 98% and nothing like that, but the very, very. Big raise in adoption over there. And when I look at the ops side of the equation, very much rejection, right? Oh, it's not deterministic. it's this, it's that, whatever the reasons are, and I think it's that history repeating itself where, okay, we have now different group that basically tries to defend status quo. whether that's because of job security or because, you know, I know better or whatever that is, and I think it'll continue in 2026. AI will not be widely used in operations. That's my prediction.
Darin
00:08:23.341
Well, that would model, again, going back to your examples of Yeah, yeah, yeah. Whoa, wait a minute. And then until it finally it took on. So right now, I think until AI is fully. Subsumed all of development, which might happen. I don't think it'll happen in 2026,
Darin
00:08:48.011
or, yeah. Well, you still have to have somebody to drive the AI agents, I think, for at least another five or 10 years, because how do you get the business requirements into the agent to hand over to the agents, right? Somebody has to come up with the requirements,
Viktor
00:09:06.978
you know, let's not predict anything that is more than a year away, kind of. I don't wanna participate in that.
Darin
00:09:12.581
Well, we didn't do so well for just one year ahead, so, okay. Uh, we, I, I would have to agree with you that we are going to not see it take over in observe or in operations, but from the development side, I think It's going to, it's, it's there. It's just a matter of how much is it. It actually helpful versus hurting versus what? And that's what I don't know. 'cause it's gonna be how are people adopting it?
Viktor
00:09:41.784
all the, all indications are there that it's very helpful. It's going to continue being very helpful for application developers. I think it can be very helpful for operations. So my hope is that I'm wrong on this one. Just to be clear. I am advocating for me being wrong on this one. I just, I see. I just don't see it happening. Mostly it's a human issue in this case, not even technological.
Darin
00:10:05.893
So if we're not gonna see AI really take over operations, let's step into the operation side of things. is DevOps really dead now and platform engineering is one.
Viktor
00:10:17.247
Uh, yes, and I think that. Platform engineering is going to continue driving for next couple of years because eventually it'll be AI driven. That's the, that will be the end state. I'm just not sure how much that is happening going to happen in 2026, but just think about it. Forget about the current limitations or what's not, right. You have, you have a platform. That you can say, Hey, give, gimme a lollipop, or gimme a database, and you get that database. And that database is done right based on your com company, best practices, rules, policies, general information from the internet, you know, when all the stars align. That sounds great because ultimately what I want is gimme a database. And ask me questions that I, that I need to answer, and I will answer them and don't ask anything more. And that's it. And I don't want to know, I, I don't care whether it's Terraform behind it or it's Crossplane or it's Ansible or what kind of like, I, I, it does things as long as I get the database, I'm happy just as if I would make a parallel. If you go to AWS, you have exactly the same attitude. You don't, you don't know what's behind the AWS, which tools are running and what is the exact process that happens when you say, gimme C two instance, eh, you know, you don't care. And I think that that's where pattern engineering will be going, and that's ai, it is going to play a massive role. I'm just not sure that 2026 is the year. There will be atoms for sure, but I'm talking about on the level of adoption now, just to be clear.
Darin
00:12:02.724
So correct me if I'm wrong. And you will. So platform engineering all along has been about either giving a block or giving a a book of how to do something.
Darin
00:12:17.949
What, somewhere in the middle there, right? Either in or in the middle Here. Here's the reason why I'm asking. It is really in 2026, the year that we actually finally codify all of our standard operating procedures. All of our runbooks, all of our everything. So that, so that AI operations in the future beyond 2026 can actually be something real.
Viktor
00:12:46.072
so. We are not going to cod it. Actually, I think that we are less likely to codify it now than we were, uh, to cod it before. Right. For the same reasons as before, kind of. Because yeah, we always plan about codding everything and we never do it right. We do something and most of the things don't. And now it's even less likely because there is even less need for it. Right. Let give you an example.
Darin
00:13:15.210
Well, hang on a second. Let me, I want to correct myself because I, I sort of see where you're heading. I used the wrong word.
Darin
00:13:23.070
I should not have used the word codify. I should have used the word document all of our things as marked down finally, so that machines can actually read it and use it. Instead of in Word documents. Okay. Now continue if you want to continue on.
Viktor
00:13:38.992
Oh no. Now we are in agreement. Right. I was thinking that when you said codify, that if I would translate it to policy that you meant VER or OPA
Viktor
00:13:48.802
I was thinking Word document that explains why we should never use latest image.
Darin
00:13:54.615
Correct tho that's what I think will come in 26 because that's gonna lay the foundation for. AI agents eventually taking over that work.
Viktor
00:14:06.653
What I predict will happen is that we will see solutions emerging in 2026, but we will not see wide adoption yet. Right? Because those are tough decisions. People simply don't wanna, don't want to relinquish part of their power first. Second, don't wanna, don't want to introduce randomness in production.
Darin
00:14:32.414
How about a third? I've been a developer my whole life. I don't wanna sit and write documentation the rest of my life,
Darin
00:14:41.864
at least it would be a markdown and not in word, but, you know, at least it would still be plain text. I, but it feels like that's where, if we're gonna be successful in years to come, we've gotta get that process in place now, I think. All the wikis become markdown documents that can be consumed by whatever models or processes you're using.
Viktor
00:15:06.182
Here's the biggest challenge. Let's say that you, you as a company, you already have all that information somewhere. You don't know where it is. You don't know which format it is, but it's somewhere, I dunno whether it's worth, whether it's markdown, whether it's slack, but somebody said something, wrote it down, or maybe recorded voice or wrote up. What will happen is realization that that's mostly useless because we've been struggling with this for decades now, and that's simply inadequate, silly processes in companies that prevent any meaningful progress. We've seen them five that five years ago and 10 years ago and 15 years ago, right. At one point, we are going to start feeding AI with our company formation, and it's going to be disaster, a complete disaster simply because you cannot, it's the same problem you remember working in the day when companies faked going agile and then. Basically kept the same processes from their waterfall once a year release, but organized everybody into two week sprints teams. And that was a complete disaster, right? Because they said, okay, this is cool. We are gonna do sprints every two weeks. But the rest of the company stayed exactly the same now. We are, we were in a good pos better position back then because I, as Viktor could choose to ignore it until I get caught. But once we start feeding all that, those wrong processes into ai, it's not going to be pretty
Darin
00:16:53.996
Well, that's what I'm thinking is we can't just feed in what we have. We have got to, we have got to feed in what is reality now. And I'm gonna say to a point, Greenfield, I mean, you can copy and paste to get your greenfield going, but I don't think we could just take, Hey, here's 40 years worth of data. Go. No.
Viktor
00:17:15.288
so that's the thing that there are actually three separate things here, right? What is being documented in whichever format you want? What is really happening in reality, which is often very different. We are all faking it. We are all faking that we are doing what is documented and trying to circumvent it as much as possible. And then something third will have to emerge because simply AI will fail more with what is documented during decades, but it'll fail just slightly less if it tries to applies the current ridiculous processes as well. Simply because everything is faster, everything is different now, okay, so here's the thing. Let's say that you were developing something for a week now, uh, in the past, and then you wait for a couple of days because Joe needs to, uh, approve your pull request. And Jock has now no idea what your pull request is, but he needs to be the bank who approve, approves it, right? And he's just overwhelmed by, by clicking buttons. Now when, when I change that and say, I'm, I'm doing it in a day, am I willing to still wait two or three days for Joe? And Joe is just one of the many pieces in that puzzle
Darin
00:18:29.405
I don't see how you could, well, you could, and that's when you go play golf or do whatever you do for a couple days till Joe gets his things done. Right now, instead of working 40, 60, 90 hours a week, you're now working 10 to 20 hours a week.
Viktor
00:18:43.096
Yeah, but the, but then that's, that's when companies will realize that there is no return of investment on ai, right? Because if you just, if we improve, if you have a FA factory line, right, where you're as assembling something and you improve one piece of that assembly line, cars are still, uh, it just, it still takes the same amount of time to produce a car, right? Or whatever you're producing, So you need to change the whole assembly line. If you want to go faster, you cannot change one piece. So you cannot just give developers, here's ai. Everything else tells the same. And just to be clear, same story as we had with Kubernetes. Same story as we had with cloud, It's just that now it's hitting up and it's moving even faster than ever before.
Darin
00:19:34.314
Well, that actually leads us into our next prediction about developer experience metrics. If I'm as a developer, I'm able to get things done faster, but I'm still being blocked by approvals, qa, infrastructure, you name it. Again, I'm gonna be sitting on my hands, or let's say I'm not sitting on my hands and I just keep on cranking stuff out. Well, that backlog is gonna keep on building up and building up, just like a beaver, building a dam in a river
Viktor
00:20:10.549
Yeah. Imagine if you get to the point, and it's not impossible to imagine that. Okay, so typically we release every month, couple of features, and then a year later we have actually, we have hundreds of features, maybe 200. In the release pipeline sitting for months, nobody even knows what what they are for, right? Because everything else is equally slow.
Darin
00:20:39.702
I think part of this too. So let's, let's back up into what we consider a normal release pipeline. We have business analysts or we have business bring things to developers.
Darin
00:20:52.452
I'm oversimplifying it here. Developers hand stuff off to operations. That's, that's the basic pipe
Darin
00:21:01.512
to testers. Alright, fair enough to testers. Testers sign off and then it goes on to operations.
Darin
00:21:17.667
get to operations. You actually call out more important things there too. So that's, that's a fairly basic flow of idea to consumer using it.
Darin
00:21:28.541
if business, and I really think business has taken on ai. Or at least some of the people I've talked to that are working business like, like it's completely changed their lives of how they think about things. And so if they're already using it and the next thing down the pipe is developers, and now developers are becoming efficient at using it,
Darin
00:21:50.711
then everything else down the line, we were talking about operations, still not being there. Well, operations can't be there until everybody else in front of them are also moving.
Darin
00:22:01.824
So having to retool the whole organization, I'm going down different path here, but it's, it's gonna be imperative to retool the whole organization. If somebody in that pipe doesn't want to play, then you as the business owner have a choice. You fire them and put somebody in, or you just eliminate it all together.
Viktor
00:22:24.888
So there are actually two distinct things there, right? somebody who doesn't wanna play ball. With what's coming now or what's already here, but there is also somebody else who really wants to play ball but is still constrained by the company policies, practices, what the process is, what's not. Right. So I can, that pipeline I was saying, right, kind of whatever the pipeline is. I can also do it more efficiently with ai, right? But the real question is, do we need it? And if you do, how it should look like, even if you get to the point that everybody plays, ball processes are going to be what, what, what is going to be the, the biggest, bottleneck Because I cannot just kind of, Hey, I'm Viktor. I choose to ignore, uh, processes because they're silly. I cannot do that. I think that actually I can do it. I just cannot do it. Allowed. Right? So we will all be build. So here, here's what one of my prediction, we will all be building some new solutions with AI that will help us do the same things we were doing before, and that's going to fail, not necessarily this year. 2026, but ultimately fail, right? 'cause that's, that's the natural in, that's the natural instinct that everybody has, right? Kind of. So you are that you're in charge of X, whatever X is, and you're doing X for years or decades now. And now you get your hands on ai, you get excited, you want to use it. The first thing that you'll come up with is, how can I use AI to do X? That's a natural, that's instinct. Kinking in instead of saying, okay, why, why am I doing X? Sorry, I, I've been doing it, I, I just don't know why. This makes no sense. Right. And then apply AI or agents or whatever it is.
Darin
00:24:36.190
Going back to the developer experience metrics, though, Dora's been sort of the gold standard for a decade now, close to it, I think, uh, how is that gonna change? Is it just gonna be, is developer happiness gonna come into play or developer satisfaction even gonna matter in that?
Viktor
00:24:54.692
we see, but Dora has been transformed from D to ai, from DO to AI in last year to already with you.
Darin
00:25:04.210
Yeah. So it'd be interesting to see how that plays out. I've got a list here. I'm, I'm just picking out some, uh, we're not gonna get into all of these 'cause it'd be boring. Uh, here's an interesting one. Ops. Does GI ops even matter anymore or is it, is this, or is it more important than ever?
Viktor
00:25:23.514
I don't think it's more important than never in terms of the process. I feel, but I don't have the concrete, solution. I feel gits will be under direct threat. Not again, not necessarily 2026, but it's, it's coming, because it's actually very inefficient process and, and time consuming process, which is good because it gives us peace of mind, but I'm not sure how it'll play out in whether we will, that we need to store money, what we do somewhere in gi I I or somewhere else, maybe one day. I have no doubt about that. We need some kind of, but everything else, full waste mechanism, periodic synchronization from, from Git to, to, to, to the system. I'm not sure about that.
Darin
00:26:15.117
I don't know here. Okay, here's my reasoning why I think it matters, but not in the way that we've used it before. Imagine this using GI Op. Get ops based AI policies or policies that help drive ai.
Darin
00:26:32.967
So, you know, we, we write the new, just like what we're talking about before, we were writing markdown files all the time now, right? Markdown is the source of truth to feed the, to feed the machine. So now I'm writing it out, I commit it, but then how does that get consumed? Get ops is the answer or an answer for that, right?
Viktor
00:26:55.054
Okay, let give you an example. There is a issue production right now. We need to fix this. We need need to, we need to solve this thing, right? And I came an ai and it is, so this is future, just to be clear. It's running it, it's, it's inform the just ilion of data points. Figure it out. I'm going to give you two options. Now what happens after he figures it out? Fixes it, it's option number one. And then. puts the proof of that fixing git or whatever afterwards. Option number two creates manifest, pushes it to gi, then waits. Patients leave for anything between seconds to minutes. Or double digit minutes until a synchronously that something is fooled into a cluster. And during that time it is constantly monitoring whether, did I get it? Finally, this is it here? Is it here? Is it here? Is it here? Is it here? Oh, okay. It's here. Um, did it work? Let me check. It did. Cool. It didn't, uh, let me, let me create a manifest push it to get again, wait for a synchronous pooling to happen and so on and so forth. I feel that there is a problem with that. That whole notion right now that we are moving it much greater. We are going towards the world where we are much moving with greater autonomy and greater speed, and I feel that that will not be. Adequate. Now petition in the first example without Git the gits. I'm not saying, oh, you didn't put the proof of what you're doing or what you did in Git. I'm not saying that right. Uh, but I'm now referring to the Git GitHubs as a process
Viktor
00:28:58.223
Now we will, we will desperately need it for quite a while now, because we have zero for operations, we have close to zero confidence in AI for good or bad reasons, whatever it is. so it'll have to create a poll request. I will have to look at it. I will have to approve it. And then it'll be synchronized, then it'll pick it up, check whether it really worked, and so on and so forth. Right. We desperately need it. Right now, I'm just saying when we gain confidence that my adjunct system is working correctly in under certain specific cases, not all. When I get that confidence, that process becomes a bottleneck.
Darin
00:29:44.292
Not a prediction. Can you ever see a time where when you are onboarded to a company
Darin
00:29:53.592
as a human onboarded to a company, that at the same time they're building an agent version of you as well?
Darin
00:30:02.914
Because it, this takes me back to, oh, what was the book? The Gene Kim book? They had all the other, oh, what was it? The
Darin
00:30:16.114
Yeah. Phoenix Project. Phoenix Project. That was it. It's been a few, I, I need to reread this one again. Uh, because there is a character name in that novel called Brent. Brent is the guy that knows everything, but there's only one of Brent.
Darin
00:30:31.805
what if as part of your, would you take a job? Let's, let's put it this way as a question. Would you take a job that mandates that you train your agent from day one?
Viktor
00:30:43.363
I will go even further and say that I can easily picture. I just don't know when. So not a prediction for 2026 that when somebody hires me in a company, they hire me because I come with my agents that do the things that I believe is the good way to do things. Just as today, we are hiring people. We hire Darin or Viktor or John or Stuart because we feel that those people will contribute greatly to our company, right? Uh, at least that's how I hope that companies are hiring. Sometimes hiring, because we need 57 Java developers. Gimme 57 random people. I'm talking about when you hire an expert, you hire that person because that person is an expert in something. What did that person work on, before you hired that person? Create transforming That his views, his practices, his everything into his own agents. People will come with their own agents. I'm a hundred percent sure of that.
Darin
00:31:48.430
It's a very interesting thought process, because that's effectively like any big enterprise. They normally go and hire big company X uh, Accenture. I'm just throwing out the big names, right. Just whatever. They hire Accenture and they bring in a busload of people. Well, now I can go and hire small guy. Pay them a lot less, but it's still very good for them because they have brought their whole team with them and they manage their team.
Viktor
00:32:18.857
I mean, you are not gonna pay that person less, just to be clear, because that person is managing his own team. That person comes with a team. It's like, you know, when you hire director of a company, that director the next week brings people. That person worked in the previous company. Right? That's happening. Just kind of like the time that director and those people are my agents. Right? Uh, and that's happening. If you think about it, that's happening right now. You dar, you're doing it. Right now that same thing. Not necessarily as sophisticated agents and what not, but conceptually you, you must have written some prompts, right?
Viktor
00:32:58.712
And those prompts are how you instruct based on your experience and your way of working, how you instruct ai, what to do and when to do it, and how to do it and what's not, right? You are already doing that pro potentially for your personal use. But is it a farfetched to say that what you're doing right now for you when whatever you're doing, the debt, would become actually your own tool set that you'll be bringing wherever you go?
Darin
00:33:29.095
Absolutely. That would be the, because that's how I'm building out my business process or whatever It's like I just, It's like I need to hire somebody. Great. I'll go spin up a new agent that does the thing that I need it to do
Darin
00:33:44.474
in the way I think it should be done. Whether it's right, whether It's right or wrong. It's the way I think it should be done.
Darin
00:33:50.744
Because again, if we're all, all, if all we're concerned about is outcomes. Then we talk about this in the human context. It's like, as long as we get as good, if not better outcomes, preferably better outcomes, then I don't care what they do in the in-between time.
Viktor
00:34:08.571
No, still there is. So let's say that majority of us right now, and I know that there will people listening to this to say much bigger number, but majority of us can manage one agent right now, maybe two. I most of the time I work with one, sometimes two, But as time passes, I will be working with 5, 10, 20. So I will always have, I will always be, I will never go and watch Netflix, It's just that the, the number of agents or the amount of workforce I can manage will be increasing over time. So you, you'll still be, be employing me to be eight hours for you a day, right? And you'll be paying me good money for it. And over time I will be able to manage more and more. But you're never removing me, at least in not near future. You're not removing me from the picture, You're just increasing the number of people, or ais in this case below me. It's kind of like a, like being a manager. Hey, you're a junior manager. You cannot handle one person, and over time you become experienced manager. You can, you can manage a team of 20, right?
Darin
00:35:23.037
I think what we could see is the transition of senior developers, principal developers, whatever. You wanna call 'em, they're gonna be turning into managers, but not managers of people, but managers of their tech. And some people may not like that. Principals will because all they know how to run is PowerPoint. Sorry, principles. Uh, but. You know, that's just my opinion. Alright. Which is I, I wanna go ahead and lead into the next item because it ties into all this. We're just talking about developer burnout is probably gonna go sky high in 2026, not because they're burned out because of all the new stuff that they're having to learn at such a rapid speed. What do you think?
Darin
00:36:03.895
Or, or e, maybe not burnout 'cause I don't like the word burnout, but their cognitive load is gonna be so high
Viktor
00:36:15.737
Oh, I, I, I can feel that on myself. Just to be clear right now, I mean, not right now while I'm recording this, but these days, right, kind of my brain is working much harder than ever before, simply because I'm thinking, you know, before I was thinking on every week on a new feature and spending the rest of my time typing on keyboard. Now I need to think about, I don't know, five features every day. and this is my claim, right? And people might disagree. that's what, what is really tough, not typing on a keyboard writing code is easy. This is my bold claim that people are gonna hate, hate me, for the easiest part is to write code.
Viktor
00:36:59.938
And the less I write it, that means that more time I'm spending on trying to figure out everything else.
Viktor
00:37:12.583
Yes. It's gonna be very unpopular, Darin, just to be clear, people are gonna hate us for this, but yeah, that's my claim, the least value. you're Giving to the company is writing code and you should be very worried if you are one of the people in a company that literally just receives instructions and translates them as-is to code, you're in real danger.
Darin
00:37:36.412
Yeah, that's a good way to think about it. I think everybody's angry at me already 'cause we had to put ads in the podcast, but you know, that's a different conversation. Uh, what do you think about IDPs this year? Are IDPs backstage, all of that? Is that, is that gonna start going away?
Viktor
00:37:59.655
I think that interfaces will have to, like interfaces we use to interact with platforms will need to change, they're going to be more text based or maybe audio based. I expect those tools to be transitioning towards. getting my input in a more free form format, whatever the exact format is, and be focused more on explaining me what's going on. And that can be very visual, Uh, but not necessarily here's a form and then feeling those, that form and something will happen. No, I think that we will be moving away from that.
Darin
00:38:41.307
okay, you brought up audio based, I think about audio devices in my house, like I'm, the the, the a word from Amazon or s word from Apple. I don't wanna set off all the alarms right now. Are you talking, you think people actually want to do that in an office?
Viktor
00:39:00.779
I mean, offices would be separate. Let, let's ignore whether it's our office or, or home or wherever it is. yeah, I think that we, people will be talking to computers. Yes. I do think that,
Darin
00:39:12.521
Very interesting. Now, one of the big items, at least at this point, is whisper flow. Is that what it is? Whisper flow.
Viktor
00:39:26.358
I'm not using it for one reason and one reason only, and that's, that it doesn't. Press enter after my, uh, I pause.
Viktor
00:39:46.557
right? Because. What I would like imagining me is kind of like, you know, uh, not being tied to, I have big monitors here, they're really massive, huge kind of sitting here, or I'm stand up and walk a bit, but I'm talking to it, right? Uh, okay. Yeah. Accept yes, no. Don't do it like this and that kind of, because wouldn't you like to talk, I mean, isn't it enjoyable to have those discussions with a person? And if it is, do you prefer having a conversation with the person about the task you're working on, or, uh, slack messaging
Darin
00:40:35.186
Yeah. There we go. Okay. Uh, fair enough. Okay. Uh, I guess if, if I could have a good conversation via audio, then I'm not opposed to that,
Viktor
00:40:48.753
I'm not saying it'll replace it a hundred percent, that we'll not type anything or anything like that. There are especially some things that are more precise by typing. I don't want it to, to start guessing, you know, strange words and stuff like that. But it would be like, one example is that every once in a while I have a conversation with Whitney, right. You know, Whitney, we all know Whitney. And after couple of conversations, we had to establish the rule that whenever she talks to Claude, she needs to tell me because it reached the point that it's ridiculous. It's kind of like she's, we, we and her are talking and one moment kind of with, you're not making sense. Ah, yeah. Sorry, I'm not, I'm not, I'm not trying to do that with Claude. Kind of like. Yeah, this is threesome, but I don't see the truth one.
Darin
00:41:36.685
So context switching is important in understanding the when context switches happen. So it just goes to prove you should not be context switching. That's just a bad thing.
Darin
00:41:46.615
Okay, I want to go to the last one. I think there are a bunch more, but I think this is, this one seems reasonable. Uh, we've been hearing shift left for how many years now. 7, 8, 10 years.
Viktor
00:42:06.435
I mean we have very successful shift left and I will give you an example of probably know, what gives you, uh, AWS is shift left. Very successful one they shifted left to or to company operations,
Darin
00:42:25.043
Yeah, that's, that's fine. Uh, but is it now going to be, because if we've got AI working everywhere, let's say we're, you know, efficient in doing that company wide, business wise, is it shift every, everywhere now instead of just shift left? Security is going to get shifted into everything that we do. Not just further upstream into development. I mean, think about it. If this would be a continuation of Shift left, should we be shift lefting up into when the business requirements are being written, if we're gonna be having agents write the code from those requirements?
Viktor
00:43:01.189
I tell you, this goes back to the previous conversation, right? Imagine there is a Joe, he's security expert. He comes with his own agents, he plugs them into the system and. I dunno, Kuber networking, he's, Michael is networking person, right? He comes with his own agents, also her own agents plugged into the system. And then I am an application developer and say, Hey, gimme a database. And then. Uh, some communication is happening behind the scenes and they all agree kind of like this is the solution, right? Kind of like it, it, it got the, the, the security agent info. What, what to do to make it secure. It got info from networking expert over there, made it right, so on and so forth. And it comes back to, to me, with a solution, and this is how it works today between humans,
Viktor
00:43:53.730
Just infinitely slower, right? So it takes weeks. So basically I think that with AI, we are replicating what we are doing. It's just that now I'm codifying in a way my experience internet, because before actually internal developer platforms and all the address plat internal platforms were all about me codifying my experience to provide it as a service. To other people in, in a company, right? and that would be me creating Terraform module or Crossplane compositions or, you know, answer something, something I just say, you just execute it, right? And, uh, it'll do. And if that's not doing what you need, well tough luck. And now we are doing the same. We will be doing the same except that I will be creating agents that will talk to some large language models, uh, and force certain rules, provide some best practices. It'll be a lot of things going on, but it'll be ultimately an agent. Right. And they will be talking to each other.
Viktor
00:45:05.092
But one important note, this is not my prediction for 2026, just to be a hundred percent clear.
Darin
00:45:10.235
Yeah. Well, I believe we're gonna start going in that direction. 'cause if we don't have those kinds of guardrails up from the beginning, we'll never get them in or we'll get them
Viktor
00:45:21.117
Because here's the fundamental problem I, I think that many people understand, is that you will hear often, oh, AA is so bad at this. And sometimes it is, but we reach a point that it's not. It's not bad at all at whatever you want. Clusters services, applications, code, it's very good at it. The real problem is that it has no bloody idea how you want to do it. Doesn't have the tribal knowledge that you keep in your head. Once we figure out how to provide it with instructions, how to make it personal, how to make it Victor's agents or their intelligence agents, then we are talking.
Darin
00:46:02.468
We've talked about Daniel Meisler, Meisler Daniel, can't remember how to say your last name about his personal AI infrastructure. PAI, this is just taking that concept into business and I, I'm not gonna make this prediction this year. But I don't believe it's far off that we're gonna have a one person billion dollar a RR company,
Darin
00:46:28.153
and I could see it to where it's a five person, a a r billion company this year. I could see that. A one person I'm not so sure about
Viktor
00:46:35.690
Yeah. Uh, to me that's already not important. We will have. Single digit teams reaching 1 billion. Yes.
Viktor
00:46:45.170
not sure. 2026. But That's coming. That's coming. Oh, I'm so certain about that.
Darin
00:46:50.473
Yeah, Viktor, we could get something going, just two of us, a billion dollars. That would probably set us up for at least the next couple of weeks.
Darin
00:47:04.382
and they, they can pay the $5 a month to, uh, listen to the episodes here without ads.
Darin
00:47:12.377
Oh, okay. Once, yeah, it's, it's sort of like talking to financial planners. It's like, oh, you, you have less than a million. Okay. I'll only charge you a thousand. Oh, you have more than a million. Okay. It's gonna be 2000. You have more than 5 million. Okay. Okay. We won't do that to you. Uh, so we went through just a handful of what we think are predictions and just ideas for 2026. What do you think? However, the slack workspace look for the podcast channel. Go over to episode number 3, 3 2, and leave your comments there. And yes, if you need to leave a sad comment because you're listening to ads. Sorry, we had to do it.