Viktor
00:00:00.307
The goal of everybody should be to make on every single level, life of agents easier. that's the current goal, Whether that's, MCP, whether that's CLI, whether that's skills, whether that's, not blocking your website from agents, It can be many different things, but that's the current job.
Darin
00:00:21.244
This is DevOps Paradox, episode number 359, Demos in the Age of AI Agents When was the last time you watched a product demo, maybe it was a 30-minute video, and you walk away from that thinking that you learned anything?
Darin
00:01:31.952
typically, if you could get your hands on it for five minutes, you would feel like you'd learned more, correct?
Viktor
00:01:36.518
Yeah, That's why I was kind of confused trying to remember because I really rarely watch a demo. I run demos. wanna run it myself when I watch something it's more like person explaining me something rather than demo itself
Darin
00:02:00.616
We have friends that can watch the video in 30 minutes, that can play in the sandbox for five minutes, except in their timelines it's a lot faster in human time, and they can come back and say, "Yeah, I've already tried it. It doesn't work." How does a vendor answer to that? Or maybe another one, you're watching a demo up on stage, let's say it's an open source project or at least code's available. Somebody's sitting out in the audience, they've got Claude or something running, and by the time they get to the point in the demo where there's actually been some problems, somebody's already put a PR in for what just happened in the demo
Viktor
00:02:38.512
Yeah. With the note that I hope that those demos are available in non-video format as well, just to make it easier for our friends
Viktor
00:02:51.247
Yeah. I have opinion that if I'm going to watch somebody, I'm going to do that. I'm not going to let Claude watch it for me. if I want Claude to give me the information, it should probably go anywhere but YouTube
Darin
00:03:05.604
Why are we talking about this? This is a conversation I had with a couple of people back at KCD in Austin this year. most people were vendors I was talking to, and they were thinking, "Okay, how will we show the demos for these things?" Because especially if in a sales, you know, you're, you're trying to introduce something and trying to sell something, whether you're actually trying to sell it for money or you're trying to get somebody to adopt an open source project within their company, you g- usually go through some sort of demo, uh, whatever that demo is. But if a company, your target prospect, is already going down the AI route, one of their first questions is, "Okay, w- where's your CLI? Do you have an MCP server for it?" And your answer is going to be, "Well, it's on the roadmap for 2029. Yeah, that's our roadmap."
Viktor
00:03:53.859
The goal of everybody should be to make on every single level, life of agents easier. that's the current goal, Whether that's, MCP, whether that's CLI, whether that's skills, whether that's, not blocking your website from agents, It can be many different things, but that's the current job. Just before this recording, actually, I was very-- I'm not going to name the service, one of the clouds, and I was, "Okay, get me all the… This is me with Claude. Get all the pricing. I wa- I want to see the price. I want to compare it with others." And it came back and said, "Blocked by Cloudflare that's not making it easy
Darin
00:04:34.861
That's just making it harder. I want to get back to what you're saying. It's our job to make things easier for the AI, or we should be targeting to do that. Couldn't we have people argue the point that we should have been doing that for humans all along? Why are AI so special?
Viktor
00:04:48.900
the more time I spend with AI, especially with agentic AI, the more I'm certain that basically the rules for now, that will change, are the same as for people. So yes, everything I say for agents applies to people. Everything I say to pe- for people applies to agents. I think it, th- there is equation mark, sign between those two. Same rules for now
Darin
00:05:11.534
For now. But my argument is, why weren't we trying to make things better for the humans to begin with? I mean, sure, we've had UI, UX, you have focus groups, you have all this other stuff, but at the end of the day, people are gonna ship what they gotta ship because of dates here's a really bad case in point. Siri launched really way too early
Viktor
00:05:30.162
Oh, yeah. It, I mean, it, it was great when it launched. It show us future that, many of us could not imagine at that time
Darin
00:05:39.640
this isn't just a now thing. This is… Companies have been trying to solve this human interaction thing, but it seems like it's a lot easier for us to actually make it more friendly to AI than it is to make it friendly for humans
Viktor
00:05:52.245
I feel that the major difference is in speed with which things are happening. Like, when we were measuring things in days or weeks or months, sometimes even years, the quality of documentation seemed less important, which it shouldn't have been, because, ah, okay, so you're gonna going to spend a full day reading this thing, and you're going to realize that you're missing still a lot of information, and you're gonna spend, a couple of more times bothering people. Yeah, well, you still have five months and, twenty-five days left to do it. It- it's okay, Now it's kind of, "What do you mean, uh, it's not done?" What do you mean Cloudflare is blocking you? This is unacceptable. I expected this and 57 other things to be done while I was sleeping Another important thing I feel is that there is a big difference between internal and external facing, right? For some reasons that I cannot explain, we think it's okay to treat internal users with a very different criteria than external ones, so, oh, yeah, you know, we cannot have, bad UX. We cannot have, things not explained well for external users because they bring us money. But hey, internal users, who cares? I mean, we care. It's not that we don't, but we-- it's not even close to how much we care, right? As if they cost nothing
Viktor
00:07:28.140
Yeah. And ultimately shipping those features that, uh, you care about. it's bewildering how people don't see the relation between, productivity of your people and the amount of money that you're earning
Darin
00:07:43.052
We're gonna be taking a look at three different scenarios. We're gonna talk about the vendor sales demo. We're gonna be talking about open source and what you just brought up, internal. Because I could argue that's probably the most important of the three. Because if you think about it, at the end of the day, you go into an organization, how many apps are truly off the shelf with zero customization?
Viktor
00:08:14.582
There is no such thing as no customization. Show me that such a thing and kind of, hey, there is a dark and light theme. I'm trying to find the most ridiculous customization. That's a customization, right?
Darin
00:08:26.593
that's true. let's go back to the way the demos used to be done, you know, four months ago. Uh, you'd sit there, you'd get on a webinar, you'd be on a Zoom, you'd do whatever. Somebody sh- showing a screen and then saying, "Hey, uh, so would you like to sign up today?" The problem with that now is people are expecting to be able to just do the demos themselves. They don't wanna get on a sales call. Heck, we never wanted to get on a sales call, just to be honest. But we did 'cause we had no other choice
Viktor
00:08:58.165
I mean, now we do, I constantly discuss a l- a lot, and I mean really a lot of companies call me all the time and, and show me their stuff and things like that. And if I see on their website, hey, this requires sales call or custom s- or special setup or something like that, it's off-putting immediately. You know, I, I don't want to speak with you. Kind of fix this. If I cannot try it out, I, I don't wanna bother with it
Darin
00:09:22.696
Here's another one. Agents can read documentation faster than a sales engineer can book a calendar slot with somebody, they can read the docs and probably have a prototype going by the time you get that figured out
Viktor
00:09:34.683
Yeah, because think about it, who is reading docs? We-- I, I assume that we are now talking about technical docs, right? For our area, not docs like how to operate a fridge, there are two main reasons why we read the docs. One is for me to understand whether I should even bother with this. That's a very s-- rel-relatively superficial reading, at least initially, right? but the other one is how does this work? And those might be different personas. Now, the latter persona is not me anymore if you show me some super-duper framework for Rust, cool. I just wanna know whether I should use it or no. If I choose to use it, well, then, talk to my friends, the friends you mentioned at the very beginning. They need to know how it works Different personas
Darin
00:10:24.403
very different personas. DHH, David Heinemeier Hansson, the guy with Basecamp, I have a note here from something you posted back in February 26, showed an agent signing up for HEY, which is their email service, Fizzy, which is something else, and Basecamp with zero integrations. The agent is the new tire kicker. I mean, think about this, Viktor. You're probably already doing this in some variation. You're sitting there in Claude and it's like, "Hey, let's go integrate with this service. Go."
Darin
00:10:53.443
Right? You just… That's what you're trying to do. And if it gets it, you're a manager
Viktor
00:10:59.305
I mean, managers don't go into every single detail That's not their job. They couldn't keep up
Darin
00:11:06.847
Yeah, so now we're de- demoing to agents, maybe. We need to be demoing to agents. We already talked about that. We need to be focus- focusing, that may be too hard of a word. considering,
Viktor
00:11:18.022
Let me rephrase that slightly, and this applies to both people and agents. We are trying to convince them that that's the right tool, service for project to use, Whether that's a demo or through some other means, right? We are trying to convince them that, hey, when you need to operate a browser, it's Playwright, or it's something else, that's what the job of those product owners or companies is, to convince me or my agent that, it should be used. And then we get into discussion whether that should be a demo or not, right?
Darin
00:11:56.483
Well, the old job of vendor demos, or actually probably still the job for a lot of them, is to prove capabilities, here's what it can do. It can do X, Y, and Z, and here's how it's done
Viktor
00:12:09.471
Yes, but think of it this way. Why do we do demos? Wh- why don't people just read all the docs, every single page in the docs instead of watching a demo?
Viktor
00:12:22.410
let's imagine that they're good. Let's imagine for a second that they're good. I'm still not going to in the c-- I, I need to be convinced phase, I'm not going to read absolutely everything in the docs. I want a demo just to see does this even, is this worth my effort, right? because I'm not going to spend a full week reading it to say, "Ah, no, not for me. Next." But now, what does it take to read average amount of documentation for agents? Like a minute, 10 minutes? I mean, ma-make it 15 minutes. Let's go crazy. Let's say that there is some latency problem it takes them less time to digest all the documentation than to go through the video
Darin
00:13:04.393
But that's also one thing that can be a little weird too if you're trying to do a- an AI demo live and all of a sudden something's waiting for 45 seconds or 90 seconds to be done. It's like, "Oh no, things are horrible." It's like, no, they're not horrible. They're just what they are. Because if you're trying to get a human to do the exact same things, 45 seconds to 90 seconds? Come on. Think 45 days to 90 days. Okay. Maybe hours, not days What about open source? I I still think a video demo is still okay. I have strong opinions on how long a video demo should be, five minutes or less. For the-- a technical person trying to figure out what's going on and trying to implement an open source project, five minutes or less is what you want to… it's okay to go to 10, but it's got to be tight. It's like I'm trying to solve this one thing. Again, I have strong biases. Now, the videos that you have on your DevOps Toolkit channel are more… They're, they're much longer. but they're more thought pieces and you're attacking it from every different side. That's not the same thing as, I'll use this example for you. you're trying to sh- show this one Crossplane composition process. You could probably shoot that in three to five minutes, right? Just a, a rifle shot of here's how you would do it
Viktor
00:14:25.277
if I would extract my vid- uh, only to keep only the demos from my videos, that would be in low minutes
Darin
00:14:31.431
But now for open source, couldn't we just have a good README, maybe a good quick start document and
Viktor
00:14:42.413
all it takes. That's all it takes. That's my argument. I'm not sure how much I will be watching demos. do want to hear opinions of people I follow, but demo, demo. I mean, if the demo is followed by, "Okay, I'm, I'm just showing you this because I want to make a point, and the point is this," then yeah. But the typical demo, right, kind of, oh, you execute something, something appears on the screen and stuff like, I'm not, I'm not watching that anymore. Um, ask Claude
Darin
00:15:11.875
That's one thing is you could have a skill, 'cause skills are a little more broad-based, too. You have a good agent's MD, you have a good skills directory, and people want to… You want to say, uh, a real demo, you'd have a quick start skill or a demo skill and just let it run.
Darin
00:15:31.256
and you've defined it. Yeah, it will show you because you as the author have defined what that skill is supposed to do
Viktor
00:15:39.672
This is the part where I feel that we are getting to demos tailored for me. Because what I'm doing increasingly is I would enter into a conversation with Claude, "Okay, what is this? Why do I want it? How is this related to the projects I'm using?" And so on and so forth, right? And then it-- If I get verbally convinced, I would ask Claude to actually, "Okay, can you show me this in action?" and then, "Yeah, but I actually want to see this. Oh, use Postgres instead of MySQL," right? and I end up with Claude assembling a demo for me that is based on the conversation I had with Claude, which is much more powerful because it's actually a demo that answers my questions, my doubts, covers my needs, instead of it being generic And that's absolutely amazing, and it really needs the documentation for that, not necessarily a, a demo itself
Darin
00:16:37.929
Let's flip it around. At least in the open source world, we're talking about trying to get people to use your project. Think about having new contributors come on. We've talked about do we need contributors anymore, but let's say that we do. you've got a new contributor come on, they clone the repo, point an agent at it, and it's like, "Does it build? Here's everything you need to know. Oh, by the way, I see on your machine you're low on disk space and this…" I mean, there's all these different things that become possible now that we couldn't have done before easily
Viktor
00:17:04.938
Assuming that that repo has the things, instructions that the agents need, right? Now we are going back to agent dot md and, uh, maybe some skills and things like that.
Darin
00:17:14.229
But again, that should become baseline. And how many people are doing that? I know on my projects I'm not doing a great job yet
Viktor
00:17:22.508
that's one of the One of the complain-- you know, people complain how AI contributions from other people are going down the drain, and my argument is always the same kind of, did you make that repo agent ready or did you let it fantasize? And or you can put it the other way around, right? Kind of like, are you really yourself effectively working this repo with agents, right? If you are, then other people will be doing, good job, right? most likely you're not. Most likely you are-- When you work on that project as a maintainer, Let's say that you're doing a good job at giving the correct instructions to agent, because you know the project, which is great, but you're not really codifying those instructions for others
Darin
00:18:07.919
In an open source, you think you would want to codify as, or codify, depending on how you want to say it, those instructions as much as possible because you may not work on that project the rest of your life. You may decide, "Okay, I'm done with this." And then if somebody has to pick it up afterwards or, you pass away, you do something, you know, just things happen, but that project still has to go on. having it codified already is going to help out a lot
Viktor
00:18:34.240
You know, many project maintainers are acting in conflicting ways and this is nothing to do with AI, just to be clear, complain how n- they're getting enough support from others, not enough contributions, and then at the same time they complain how bad the contributions are make up your mind. it cannot be great because you have infinite amount of experience with this project and somebody just landed on it so make up your mind. Do you want to be the sole contributor or you don't? We want to make that investment. And now that investment that we were before making for people, now we're making them for agents. Agents operated by non-maintainers, which is important
Darin
00:19:20.543
Let's move on to internal demos. In the past, internal demos were talking about, "Okay, w- here's the changes, here's what we're gonna be doing." Now it's about changing the workflow. Like, "Here's how things now flow through the system."
Viktor
00:19:35.020
I'd like to start this by stating that internally, I don't want to see plans from anybody anymore. I don't want somebody telling me, "Hey, this is the plan. This is what we are trying to do. What do you think?" Instead, just show me that something done, not production ready, not code written with semantically exactly as you want it and not, cut the corners, that's fine. But when you're pitching me idea for a new feature, I wanna see the feature. That's the low b- lowest bar now. If you don't show it to me, if you're gonna tell me about it, I'm not interested
Darin
00:20:15.634
this takes us back to what we were doing, but an extension of with open source. That sentence made no sense, but that's okay. You want to be able to have a conversation about the demo and you want to see it done, not production ready, but you want to see it done in the way that you want to see it done. I may have a different way of wanting to see it done because of the size of the system. So being able to have the conversation with… Let's think about it this way. we have a team that's doing an SAP implementation. If you don't know what SAP is, be very glad. But you're doing an SAP implementation and you're concerned about AP, I'm more concerned about the HR side. Our demos are completely different. But if I could just have that conversation with SAP that's like, "Hey, what's the status on this? Show me what's going on," and they can show me, that's the kind of demo that I want as a stakeholder
Viktor
00:21:22.913
Yeah, but now we are. That's the point. Again, point me to, the docs, with all the information. I will provide instructions of what I want. We are going to have a conversation, and I'm going to get the demo. Maybe not from the first attempt. Make it five. I may spend a full hour on it, But I'm going to get the demo that I'm looking for Unless it's SAP, then you're not getting… Then, then you're not going to s- set it up yourself, just to be clear
Darin
00:21:53.489
Not only are you going to be getting the demo that you want, you're getting it done on your time, and you're probably generating more questions to come back, feedback loops, to be able to figure out, "Okay, here's what we really need." Because most of the time you're sitting in a demo and you're sitting there on your laptop or on your phone texting while the demo is going on, and you miss the things that are important to you until three weeks later when it's in, and then he's like, "Oh no, this isn't supposed to be the way it was." "Well, we told you about it." "Well, I was busy." Okay, number one, that was your fault. Number two, I should have been pressing it on, on it more as the developer. But now if you're actually interacting with it, the chances of you doing something else at the same time you're interacting goes very low. Because now you're context switching versus multitasking. Meaning, and multitasking basically meaning ignoring everybody else and just doing what you want to do What do we do about this? Well, one of the things we can think about for both vendored and OSS projects, sandboxes first, demos are second. If you don't have a way to get a sandbox up, whatever sandbox may mean for your application then you haven't built it right. Because number one, this takes us back to the beginning. I want to be able to go in and point, click, run a CLI against, do something. I want to interact with something. I don't necessarily want to see a video first. Now, if it's a two-minute video, great. that gets me tied in, but that two-minute video gets me to the sandbox, which then leads me to longer trainings
Viktor
00:23:31.395
I feel that that's where SaaS comes in even if I plan to use something by self-managing that something, managing it myself, I'm going to install it and stuff like that, I really, really appreciate where there is… Again, project yesterday I was working on, or a tool, service, whatever, I have zero plans to use SaaS. I'm going to run it myself. I went for SaaS instead. Just because, okay, I w- I wanna see this in action up and running, kind of like what's the minimum setup? How can I connect it to my cluster without installing it like this? Perfect. Great. Is this what I want? Yes or no? Yes. Cool. Now let's talk about me spending time and setting it up and all those things. Not before I answer the question, which is whatever the question is, but the answer is yes or no So I really, really appreciate, when there are projects, and almost all open source projects have some, entity behind them, some company or multiple companies, when there is that something as a service, oh man, I'm really happy even if I'm never going to use
Darin
00:24:41.523
Another thing to think about is making your docs agent-friendly. We've talked to vendors about this. Some vendors are still writing human first and agent second, and others have flipped to where it's agent first and human second. Clean markdown, LLMs.txt, uh, there's lots of different things. OpenAPI spec if you've got an API. All these things, uh, especially if you've got a, a SaaS and you're wanting to be able to connect with it using a CLI, you probably have endpoints somewhere, so why not publish the spec? Because not everybody may be able to run your CLI, but they can sure as heck make a call against your API
Viktor
00:25:22.169
I feel that those things might be less important, important but less important, le-let's say that you have docs specific for agents. I think that that's amazing, that's great, they're going to spend 15 seconds instead of half a minute on it. That's great, but not critical, If it had to sift through HTML, not good, more tokens spent, but ultimately I hardly notice a difference.
Darin
00:25:49.602
We should be demoing the failure modes. Most of the time in demos, you're always taking the happy path to where nothing fails. Now it needs to be the opposite.
Viktor
00:25:58.817
How can you sell it if it fails? It cannot-- It never fails, otherwise you, you cannot sell it. But, but Darin, didn't you work in companies? Don't you know how business works? It never fails. None of-- I ca- I can sign that none of the projects I've worked on, none of the companies I've worked in have products that ever failed
Darin
00:26:23.127
And you're crossing your fingers behind your back. Yes, I can see that. Why do we need to see failure points? we need to be able to see failures because we know in real life failures happen. Disks fill up When a hard drive fills up, there's no amount of, technical magic or AI magic that will take care of cleaning that disk out for you So we know there's going to be failure modes. How do we just deal with the failure modes? And that's because I w- a- as somebody, if I was buying, quote-unquote buying, whether cash or brain cells, I want to be able to see it fail and see how it recovers, and how long does it take to recover? I think that's something th- that's always been a disserv- I mean, you might do one, like, off on the side, but you never show that as a lead. I think showing it as a lead, because resiliency now is more important than ever
Viktor
00:27:14.443
I think that you just hit on the probably the biggest or one of the biggest selling points on top of open source, you can use open source, it works. You reach certain scales, scale, it will fail. ka-ching, ka-ching. That's the sales motion, right? which is fair, completely fair. somehow those companies, those maintainers need to earn money. they precisely focus on two things, that's failure, what happens, or scale, call it however you want, and security You want it to fail less? Well, ka-ching, ka-ching
Darin
00:27:49.720
For internal demos, again, I talked about changing the workflow. Here's what a Tuesday looks like, Or here's what a Friday looks like. We lock everybody out, especially during the summers, 'cause everybody's off on Fridays. Yep. Internal demos we have to treat… Now, internally it's going to be a little different, but, or at least I think in my head it's different, but maybe not so much anymore. Because you let an agentic AI run off into your network safely, it's going to find your system. Why not integrate with it? Heck, if people are integrating with their Google Workspace or their Microsoft 365 workspaces or anything else, aren't they going to expect to be able to connect up to the internal application just the exact same way?
Darin
00:28:39.184
They would like to, yeah. But then it's your problem to implement that. But it wasn't in the spec, so now we have to go back and spend another five years coming up with a spec so we can implement it over the next five years, and by that time we're all out of jobs
Viktor
00:28:54.098
There is a word for that. No, no, we're not out of jo- that's job security. I still have three years left in me to, to actually, uh, implement this thing. Cannot fire me You can fire me up when I'm, once I'm finished, which will be never
Darin
00:29:10.058
Never finish. Yeah. This, this is a, um, this is a never-ending problem. It's the arrow shot at the wall going half the distance. It will never reach the wall doing half the distance. Here's something that we should have been doing, but now it makes it real, or at least AI, agentic AI at least makes it possible. Sit down with a number one stakeholder, an early adopter, and they sit and help build it with you Long gone are the days of 300 pages of a PRD. Well, actually, th- it's there, but it's used differently now. Instead of 300 pages coming out from Word, now I've got 300 pages that are being fed into an agent that I'm working with the business person to know what's going on, I know technically what's going on, and give us three days to a prototype A good prototype?
Darin
00:30:05.007
Yeah That's what we want, right? We're trying to create business value so that we stay employed. Hmm, actually, create business value that makes the company money so that we can stay employed
Viktor
00:30:18.184
That made me think, you know, in the past, when I worked-- in companies I've worked with, we always try to stay away from customizations requested by customers, because that means that either we bloat the code or we fork it, and then we need to maintain two of them. Those things now become so much easier, You can give each customer actually what they want instead of, a subset that you decided was what everybody wants. Imagine that, fully cu-customized solutions based on a product or set of products. That would be awesome
Darin
00:31:00.355
Sounds like you're gonna be busy this weekend building that. We've already talked about this. Cut your demo lengths. If your demo takes an hour to do, it's too long. People are gonna be checking out after five, 10 minutes
Darin
00:31:17.532
mean, I can't sit through a Marvel movie for an hour without letting my mind wander
Viktor
00:31:22.715
I feel that conferences are a good test of your demos because by then you have a booth over there as a company, you have a demo, and basically you have very limited amount of time. You need to accomplish two things. First is that somebody passing by sees what they want to see and stops, which is very hard. And then when that person, if you accomplish the first step, kind of like congratulations, and then that person does not leave, before the end of it, which is extremely hard as well, right? Because how long does somebody wants to stay in a booth? Uh, and the answer is not enough
Darin
00:32:08.441
Not 10 minutes. Unless they have chairs. Now if they have chair, nice chairs, maybe. It's been a long day.
Viktor
00:32:16.071
Nice chairs and if you're one of those who actually sponsors popcorn and beer, yes
Darin
00:32:25.785
Is this statement true? Do you think most demos are still being created for the 2018 buyer, the 2016 buyer? Are we still there
Viktor
00:32:36.119
I mean, most of the industry is still trying to figure out what's going on they're not saying that. They're all, "Oh, we are AI, we are agentic." They, they have no idea what's going on and how does their IP fit into that something, because that's the real question
Darin
00:32:52.877
i-imagine this, being able to take one of the demos you normally do. Let's say you're a sales engineer and you're typically doing this demo because you don't have self-service yet, so you're doing a demo. Could you take that demo, turn it into an agent plus a sandbox plus agent docs, and how far could that prospect actually get through your demo without needing you at all? Now, of course, you're not getting the credits and blah, blah, blah. Who cares about that? Because think about how many times you get on calls that don't pan out, I guess is the better way to say it. So you still wasted time, in theory, because you didn't take what you lost in that deal to refine your demo. Now I'm just saying have the demo be sandboxed and be done with it, SaaS or not
Viktor
00:33:38.690
So we're not talking about demos that are not watched by somebody, but more like a demo that you present to somebody, right? Um, and in that case, I think that you should have it set it up. You should have an idea of what the demo is, and then you should roll with reactions I think that if you're good at what you're doing, when you show a demo to customer A, that demo is going to be different than to customer B. First, because they are different. Second, because their faces are going to react differently. And third, because they're going to ask you questions that you should show in the demo, and those questions are going to be different. So what I'm really-- Where I'm really going is parts, if not the full demo, it differs every time you present it If you do it right. Kind of, oh, okay, so we have this thingy that does something with your code. cool. Uh, you have a demo that is, let's say, using Java. Cool. And then you go to somebody who is writing things in JavaScript. Well, the demo cannot be the same anymore if you do it right. Doesn't have to be super complicated, but kind of, oh, okay, so, uh, you want this build to be based on what you're actually doing, which is JavaScript. Cool. How difficult it is for me to change the build to do JavaScript if I know what I'm doing? Not much
Darin
00:35:13.963
And unfortunately or fortunately, that's where agentic AI can help out. It's just up to you. Before we had that tool, it was hard to make those kinds of changes, especially fast. But now it's possible
Viktor
00:35:28.478
One thing I, I was brainstorming with a friend for a talk, but now I think that that should be done in demos. You know, when you're in, um, in a room with a couple of people from the customer you're visiting or company, we should just kind of turn on, always listen for Claude and kind of like, okay, you build it while we're discussing it, Oh, Joe just asked this question. Cool." Kind of like adapt. Go for it. Just listen and react
Viktor
00:36:08.712
Yeah. Yeah. So we're going to sign the 100K deal a year, and I'm like, "I, I don't care about tokens." You cannot spend enough
Darin
00:36:19.362
Well, you could, that's the problem. But let's say that you were doing a two-hour workshop, 'cause this is typically where those kinds of things would come into play. Do a two-hour workshop with somebody. It's listening. Let's say you're using Fable. So it's listening, it's doing all the work, churning, looping under the hood, and by the time you get to the end of the workshop, you say, "Uh, hey Claude, what you got?" And there it is
Viktor
00:36:47.472
this is now not anymore that we're talking about running 50 agents in parallel. We're talking about single agent doing something. I'm pretty confident if I would say that whatever the cost is, it's less than any of the people being in that meeting. Not to mention all of them combined. Most likely
Darin
00:37:08.347
Yeah, so if, if you only had one or two people in there, okay, yeah, probably more expensive. But the way I'm thinking about it is, okay, if I can sit down and do a live demo, I'm saying the word demo very loosely now, for two hours, I think I would probably be hard-pressed even with Fable at the cost that it's at, to spend more than $5,000 in those two hours
Viktor
00:37:28.721
Fable is a special beast. Let, let's exclude it for a second. But if you go with Opus, a kind of expensive model, and let's calculate it on yearly basis, and then we can easily extrapolate to an hour. So what does any of that pe- those people in that meeting cost to the company? 200K a year? Or should we go lower? So let's say v- it's, it's cheaper labor and what so not, let's say 100K a year. I mean, you cannot spend 100K on a single agent running all the time. It would be close to impossible. You ca- you cannot spend even a fraction of that amount of money. And then if you extrapolate it to an hour, still holds true
Darin
00:38:12.724
But using it as this one-off tool to sit there and build while you're having the conversation
Darin
00:38:22.733
You just have to float the amount of token, the cost of the tokens for the meeting.
Viktor
00:38:26.911
Yeah. And especially, look, if it's a demo, And let's say that you prepared your stuff, you have environment where the demo can run, and you have your skills, CLIs, MCPs, whatever you're using, kind of it's all ready. Well, uh, let me tell you that Haiku is overpowered for that. Assuming that you set up everything right, not vanilla one. It's overpowered. And Haiku costs nothing I mean, nothing is nothing, but it's cheap
Darin
00:38:53.744
So if you could get your standard demo flipped into a, okay, what does an agent look like with this and sort of sandbox, what's left after that is the true demo that you need to do. those are the gaps. You don't need to d- give the demo of the whole thing. You need to give the gap demos
Darin
00:39:11.793
until you can turn the gaps into part of the story along with it. I know a lot of this is probably focused more at vendors and people working in open source, but I think the biggest win for this is how to think about how to do this for internal applications. If you are on an internal applications team, every company has at least one, if not 1,000 teams that do this. just think about how to do it faster. And if you can really work with a business person and you're allowed to use agentic AI, there's no reason to not work with them live. There's no reason for them to sit in isolation or write a PRD for you. There's 0%, 0% reason for doing that anymore
Viktor
00:39:52.139
yeah, they can write PRD still, but then everything that comes afterwards, It's uh, it's kind of… Okay, so we had a discussion, the three of us, you, Joe, me, Claude, that ends with PRD, and, uh, when I close my laptop and leave that room, it will start working on it
Viktor
00:40:15.626
Unfortunately, probably differs from case to case, but it makes me, it makes me nostalgic. I've, uh, now, now after this conversation, I wanna go back to professional services
Darin
00:40:26.694
Sounds inter- well, not professional services, solutions architect. That's where you want to go
Viktor
00:40:38.915
I don't have a real problem talking to humans. I have real problem giving instructions to humans. Kind of do this, and why, why did you not come to work today? You know, all the management part of, of the story, kind of. But talk to humans, I love that. I loved it when I worked with professional services. it's just that I talk to humans the same way, uh, independently of whether I'm in an office with humans or in a bar. That rubs some people the wrong way, but other than that, I have no problem talking. I have problem managing people
Darin
00:41:11.838
So what do you think? If you're stuck doing demos, maybe you're an SA in a vendor or you're trying to figure out how you're going to try to get more eyes on your open source project, how are you gonna solve it? Hopefully we gave you a couple of different ideas today. Head over to the Slack workspace, look for the podcast channel and leave your comments there