Temporal UX

Transcript

00:00:00Welcome to Think Different, Think AI, the podcast by Mark and Jens.

00:00:07Two technology-loving minds who not only talk about artificial intelligence but live it.

00:00:14Here there are clear classifications, real practical insights, and a fresh perspective on what is possible.

00:00:20Understandable, critical, and always with a wink.

00:00:24AI to think about, to smile at, and above all to discuss.

00:00:33We hear it already, that's totally great, because today it goes without the film counter

00:00:37shutdown.

00:00:38A bit too many mishaps.

00:00:39Welcome to Think different, Think AI.

00:00:42If you're wondering why I just said that, then it has a big

00:00:45significance, because Jens and I, as you've just heard, are actually sitting

00:00:50physically together today, meaning without video, and on the side we are totally

00:00:55excited to have a non-alcoholic drink and talk about a topic that

00:01:02It's a world of concepts that I can no longer get over my tongue. Jens, it's nice that we meet here today.

00:01:06Jens, what are we talking about today? Today we're talking about time design and the modern concept

00:01:11of Temporal UX. So, I have indeed heard of time design before, yes. There's

00:01:18really nothing better than when it’s said, it's nice to do absolutely nothing. Digital

00:01:24detoxing while sitting in nature and just looking at the mountains, I could have still experienced that.

00:01:30But the Temporal UX, I have never heard of that. Can you share a few warm words about it?

00:01:38Because I hope I’m not the only one sitting here with question marks in front of my beer.

00:01:46Sure. So, regarding time design, whether this term has existed for a very long time, there is some uncertainty.

00:01:52Recently, I also wrote an article about it myself.

00:01:54There aren’t many who are dealing with it in the field of AI, where the topic

00:01:59comes in UX or CX or also in service design.

00:02:03These are all design disciplines that deal with how to create an experience

00:02:07for a person as well as possible, how we can sell our products well

00:02:10and that everything comes together so we can make money, but also

00:02:14that our customers are happy.

00:02:16And customers are sometimes not happy at the airport, just to give an example

00:02:19.

00:02:20Situation when I arrive at an airport, having gotten off the plane, I go

00:02:27down the stairs or up this jet bridge into the airport, then I have to look for

00:02:33where my luggage has actually arrived, walking through this airport,

00:02:38the baggage carousel arrives and then I wait for my luggage.

00:02:41I know that.

00:02:42That's all time that I spend there, exactly, that I spend there in this

00:02:46airport, which can certainly be designed differently and can also be perceived differently in my perception.

00:02:51Experiments were conducted to see how

00:02:55I can shape this time that a person spends at the airport so that they perceive it as particularly

00:03:00pleasant. A simple trick that was done was to make the path to the conveyor belt a bit

00:03:06longer through the airport or to pass by the Duty Free again or other topics,

00:03:10all the nice things available. So, in principle, when I arrive at the conveyor belt, even though it always

00:03:16takes about 20 minutes for the luggage to be transported from the airplane to the baggage claim,

00:03:20it now took me 15 minutes on this long path

00:03:25before I arrived. So if I wait, I wait only five minutes instead of 10 minutes for the luggage

00:03:30and I perceive the overall time I spent at the airport waiting for the luggage

00:03:34as clearly shorter and more pleasant. So, for example, that's why such a

00:03:38Time management is a very important aspect, how do I design waiting times, how do I design

00:03:44routes, we know this from the physical world, and when we talk about Temp-UX, it's about

00:03:51how can we transfer this into the digital world, and especially in the interaction

00:03:55with AI models?

00:03:56I was just wondering if the answer is to bring the notebook from the study

00:04:00into the kitchen, so that the path is as long as possible while the agent is working,

00:04:04so that I can essentially cover some distance.

00:04:06I briefly thought, thank you very much, I always complain

00:04:11that I have to walk through half an airport to get to my luggage.

00:04:15Now I also have to be grateful that my time is so nicely designed and

00:04:19filled with content...

00:04:20You end up taking additional steps, right?

00:04:21I end up taking steps and I have a view, yes, I see the Tutifry Shop or I don't know

00:04:26what, that's really great.

00:04:28And I wondered, is this also the phenomenon when you drive an hour

00:04:33to your destination and you're of course driving at the maximum speed limit

00:04:40the whole time, you were essentially here and then you stand in traffic for 10 minutes and those 10 minutes

00:04:44of traffic are already annoying, but if you can roll a bit, then it feels better.

00:04:49I somehow feel like that’s also a little bit about that,

00:04:51only that no one else designed it for me, because traffic is known to be only

00:04:56annoying from behind. From the front, it's quite alright. From the front, it's quite alright, but of course it's also a

00:05:00good reason to implement speed limits and also dynamic speed limits,

00:05:04to improve traffic flow, because of course it is better for us anyway

00:05:09to drive at 50 kilometers per hour on the highway than to be completely stopped, while everyone is somehow going in

00:05:14at 120 or more into a traffic jam, which actually makes the jam longer, because everyone hits the brakes, which...

00:05:20I'm trying to remember this clearly, when...

00:05:22This is actually something that can be consciously managed, basically, if you intervene early,

00:05:27it makes things better for everyone later, because we feel that we first arrive at our destination faster anyway

00:05:32because we're not stuck in complete traffic, but the flow is better, and we also perceive it quite

00:05:37pleasantly at that moment.

00:05:39Now we are sitting together today.

00:05:41I mean, I don’t know if the listeners are aware.

00:05:43Normally we are separated by a certain number of kilometers.

00:05:46I am more around the Karlsruhe area,

00:05:49while you are more in the Cologne area.

00:05:51Yes, aside from the initial letters

00:05:53we are mainly connected through digital channels.

00:05:56And we had the chance today,

00:05:58I'll say at our Cologne office situation

00:06:02to meet for work, because we are somewhat together

00:06:05doing vibe-coding with it. That’s essentially why we’re here,

00:06:09on the idea of meeting in person for a podcast episode in the evening

00:06:13also on our topic that we came here for, because we were in contact with vibe-coding

00:06:18exactly about that. And how did we come into contact with that? Would you like to

00:06:22briefly explain it, so that someone thinks that time management, vibe-coding is fun.

00:06:26Yes, to explain this setting a bit more,

00:06:29because it makes it even more dramatically exciting. So there was another

00:06:32colleague involved and we were thinking about how to actually solve a problem together with Mark,

00:06:37and the colleagues and I described the problem.

00:06:41Mark then coded it extensively with his AI agent.

00:06:45My very own.

00:06:47Just my own.

00:06:48Exactly.

00:06:49This led to the fact that we of course initially had a setting in this room,

00:06:53where we all sat with our laptops, so the colleagues and you had no insight

00:06:58into what was actually happening in the background, except that we occasionally saw Mark

00:07:01can see typing and then see waiting, which means something has already happened in this

00:07:07spatial dynamics, which is strange because Mark was at least the only one with a visual input about this,

00:07:13that the AI agents in the background are doing something. My colleague and I were a bit cut off from that.

00:07:18Additionally, with Mark, this also happened. He kept looking at

00:07:23the screen and gave us status updates that he derived from, let's say, sporadic feedback,

00:07:29that the AI he was given said, okay, now it can't take that long anymore,

00:07:34it is now at part 4 of 6, and it still took 20 minutes until part 6 was finished

00:07:41and not quite finished, unfortunately, but there were still errors in it, but that

00:07:44was to be expected, but still this timing, this focus on the outcome of a prompt,

00:07:54that I provided, which hasn't really been consciously designed until now, to be honest.

00:07:59And the situation reminded us a bit of old times, yes, you stand there and the AI doesn't say yes

00:08:05now just ten minutes and you could set the clock by it, but the thing has gone down, we have indeed started

00:08:10Craft Agents with an Opus, with an Opus model behind it, and we gave it a task and it planned and did and acted

00:08:20and you can always see nicely which feather goes in and what works and what it has planned and then there was just a

00:08:266-point list and for me that was of course implicitly. What does course mean, it just looked like

00:08:32yeah well, I have to do those six things to get your nice result to discuss with people

00:08:38maybe it has changed or I don't know what. So a bit like the

00:08:41old installation bar with Windows, where you might have used your 32 diskettes for Windows

00:08:46and laid them out. The thing then runs, 5% installed, 10% installed and that's how it was

00:08:52for us too, right? He checks off a point, checks off the next point, checks off the third point,

00:08:56checks off the fourth point and it takes a while, and it takes a while, checks off the sixth point. Done.

00:09:01So like back then when it says here Windows is being installed and at 98% you have the

00:09:06feeling that IT has dealt with the subject of mathematics for the first time because

00:09:10how can it be so difficult? You know what you are installing on an empty computer?

00:09:15Okay, you can't calculate the speed of how back then people used diskettes,

00:09:20If I know what the chain is, yeah, so please look on Google, that's what okay, is that storage symbol still in Word that looks like a diskette?

00:09:27Yeah sure, of course, there is still one.

00:09:29Okay, okay. So what most probably only know from the storage symbol and then you switched this chain, yeah, each chain was 1.44 megabytes, little fun fact, they did exist with 27 kilobytes and by the way I mean real kilobytes to my children's teachers, real kilobytes, yeah, because I was still taught to calculate with 1024

00:09:49and not with a thousand today if you are in school today and say how many mb or

00:09:54how many kilobytes in a b then they say a thousand and you are like no it has never been like that and it never will be

00:09:59yeah because but good is another topic please all teachers write yes and

00:10:05there you stand and back then you switched the diskettes and as I said yeah the

00:10:08thing knows okay 32 diskettes or something like that you install you’re at diskette 28

00:10:13and the thing says 98 percent it can't mathematically actually be that hard to calculate

00:10:18What percentage am I really, and how much longer should it take?

00:10:21And we felt reminded of that today.

00:10:23When the AI said, yes, 4, 4, 4 out of 6, 4 out of 6, 4 out of 6 done.

00:10:28What's going on there?

00:10:29Yes, definitely.

00:10:30When someone tells the story they're telling now, it's of course a nice analogy.

00:10:33Back then, it was, put bluntly, not quite fitting now.

00:10:36But each individual link was essentially like new prom that you might have entered before,

00:10:41where something had happened in between again.

00:10:43Sometimes it went faster, sometimes it took a bit longer.

00:10:46Sometimes you had to insert disc 4 again after disc 16, which you also didn't understand.

00:10:50Yes, there were also such things.

00:10:51And as I said, Mark always brought up this nice example of 98%.

00:10:55I believe that was the case for every installation of anything.

00:10:57Often, the bar quickly reached that 98%.

00:11:00And then the last 2% actually took twice as long as everything before it.

00:11:05That was a very, very, let's say, a very adventurous design of wait time,

00:11:12that sometimes our installation routines demanded from us back then.

00:11:16It was similar with games.

00:11:19There you would start an effect you also knew from normal installations.

00:11:23Sometimes you would sit there, and then the directories that were being loaded would fly past.

00:11:28Or with games, it would say, oh, now the textures for the landscape are being loaded

00:11:32or now the wireframes of the spaceships are being loaded.

00:11:35And then you started to play with it a bit more consciously.

00:11:39There was a little game where they made a joke by simply

00:11:44inserting crazy or totally nonsensical loading messages.

00:11:48The demons will have brought them along.

00:11:50Exactly, the maximum, absurd cold jokes loaded or something like that.

00:11:53So, there were all sorts of examples that brought a little smile

00:11:56to bridge that time.

00:11:58So that you could say, if I'm sitting there and not doing anything else, then I should

00:12:04at least be somewhat entertained during that time.

00:12:06So, that was the thought.

00:12:07What were joke generators at loading queens?

00:12:10Hm, small idea, maybe I should talk to the Pudderdorfs at our place, whether that's not

00:12:14a good idea.

00:12:15Telling flat jokes during the loading queens or other processes.

00:12:19Yes, I used to design a lot of websites, and back then there were

00:12:24still some websites, you built flash back then and you always needed

00:12:27it as well, yes, yes, it was nice that we did that

00:12:34back then.

00:12:35Yes, yes, yes.

00:12:37There were some nice champagne websites that I designed back then with the theme.

00:12:41But what we sometimes did to specifically address that waiting time,

00:12:44especially with flash content, was incredibly important.

00:12:46that essentially the entire flash arrow was loaded,

00:12:48because otherwise, if you released the user on the half-loaded flash arrow,

00:12:52then this website partially worked,

00:12:54if it worked at all, or produced very, very strange things.

00:12:56Therefore, you have to get them to wait a little bit,

00:12:59so that everything you need is also loaded.

00:13:01And then sometimes we just incorporated Pong like that.

00:13:03While you were basically watching the loading screen, you could

00:13:07at least play a round of Pong against the computer using the keyboard, and that was kind of

00:13:12our approach back then to make loading times a bit more pleasant.

00:13:15It's like the mini-games on LinkedIn, where I always think, hey, business network,

00:13:19yes, and then they offer you, are you playing Snake and Mayon here now and no

00:13:23idea, not Snake and Mayon, but something else where I think,

00:13:27okay, I'm not sure now, but well, I had that with games too, by the way

00:13:32with the loading times, that was one thing and the second was, I played World of Warcraft, like

00:13:35a diode you have to say. And when new add-ons came out, you also had the

00:13:40Waiting time, because then it said 128 players ahead of you, similar to the Hotbassers principle,

00:13:45after the motive must be crazy, still 180 players ahead, press continue now,

00:13:49because the servers were completely overloaded, but that was also the topic and yes,

00:13:54when you play in these funny things, you don't know how long it takes. And in front of the

00:13:59problem, we are now as well.

00:14:00This waiting time you just described with these innovation queues, that's known,

00:14:04yes, one or another perhaps, who deals with the topic of AI, also with video generation AIs,

00:14:09where I perhaps didn't buy the very expensive packages in principle, then I will

00:14:14also be put in the queue first when videos are generated.

00:14:17Well, I've seen that before.

00:14:18With blades or something else, there are different video generation tools,

00:14:22where one is then aware, yes, okay, not everything has to happen immediately

00:14:27in principle.

00:14:28also a bit the topic, if I consciously deal with this, so consciously

00:14:32handle that something takes time, then it is perfectly fine if the AI

00:14:36finishes four hours later or even if

00:14:39it's tomorrow. Not everything has to be instant in today's world. But there we are

00:14:42of course also spoiled, right? I mean, with emails you said, okay, good, the

00:14:46sending email and then will be read at some point. With messages it's

00:14:50really bad, like with iMessage, Words, Telegram Teams, how the

00:14:54whole grammar calls itself and you press send and you actually expect, even if you

00:15:00might not express it and also perhaps don't assume it yourself, but if

00:15:03you don't get a response relatively soon, then you sometimes have colleagues and you

00:15:09admit, I've done that too, who then call and say, did you see the message

00:15:12I wrote. Even better is, you write the email, send the message

00:15:15afterwards and then retrieve it. Yes, so along the lines of maximum overload. Yes, okay, but

00:15:21But let's please get back to.

00:15:23Exactly.

00:15:24Because we just touched on this topic a bit, is that actually

00:15:29productive, when we look at screens?

00:15:32Ah, back in our youth, when we were waiting for loading screens again

00:15:37it was actually part of the tension curve, finally

00:15:40being able to play that game or finally getting the new update.

00:15:43But in the modern working world, if I say I have a tool that brings

00:15:49enormous efficiency in many, many use cases, whether it's programming, whether it's

00:15:56research, summarizing things, creating things for me,

00:16:00generating graphics, generating videos, whatever. Is it smart for this AI

00:16:06to generate this content for me?

00:16:08That I then look at, just like in the loading bars of the past, at what the AI is doing while

00:16:14it produces the result. So I actually think it's good to see what it

00:16:19is doing. Yes, you can read everything afterwards that it does.

00:16:25Yes, I actually think it's quite okay that you can read what it does,

00:16:28because depending on what task you assign it, you already have a

00:16:32bit of a feeling of where I can trust this good piece and where not.

00:16:37Trust is perhaps a strange word, but where do I know or have I

00:16:40a good feeling about handing over this task, and where might I

00:16:43feel the need because I am also unsure whether I have expressed myself well. Yes, I mean,

00:16:48you saw it today, yes. So I'm not a born Requirements Engineer. I typed the

00:16:52thing while you were talking, I typed what you were saying to me. Tried to

00:16:57somewhat thanks to touch typing, slam it into the box and

00:17:02then just pressed Enter, and I would say, neither the sentence structure nor all the words were

00:17:06completely error-free. Plus, it wasn't formulated for the system at all. It developed

00:17:10only through the conversation in the factory. From this perspective, I find it really important to look at these moments.

00:17:16Yes, but it also has to be done right because you just mentioned a

00:17:20word that naturally resonates with me, the topic of trust.

00:17:24Triggers. Yes, triggered. This means, in principle, we also need to trust what is happening

00:17:30because if we look at the history, which isn't very long, how we

00:17:35interact with AI and the topic of reasoning, someone appeared, we have already seen in this

00:17:40reasoning that was displayed to us various interactions.

00:17:44So at the very beginning, the reasoning was, I don't even know if it was introduced by the Chinese

00:17:50AI first and then copied by CHPT, OpenAI.

00:17:54Diepsik was, I believe, Diepsik was the first there and they basically always

00:17:57a very long reasoning was then displayed that you could read along with and then came

00:18:01at the end.

00:18:02first Open AI came, and then there was this big shock that Psyk is being released and

00:18:06the question arose, oh dear, we have the Chinese doing this because they don't have

00:18:10the cool hardware that we have, we trained now in Tel, so is our training,

00:18:14the way we do it, good and right. I believe that was already the first reasoning at Open AI

00:18:18to see, then I was like. Yes, exactly, but it was definitely the case that it was very

00:18:22detailed at times because the basic idea was that if I show the human

00:18:27in front of the computer what I’m doing, then they trust this output of the AI, so

00:18:32the topic of justifying is also important again. So I can then better understand,

00:18:36how the thought processes of the AI are. The more complex the requests and the more complex the output of course

00:18:40becomes afterwards when there’s still research done in between, multiple things being requested.

00:18:44of course, this part that is displayed slips into the gigantic.

00:18:48So if I now say I'm programming a complete application and a whole

00:18:53case structure is being set up, whether it's local or directly on GitHub, and Python

00:18:58scripts have to be written, databases have to be created, then this happens at a

00:19:02speed that I couldn’t keep track of if everything just runs by me.

00:19:07And this has also been consciously designed by the designers of the respective

00:19:10AI providers, who say, okay, we bundle topics together, you nicely

00:19:15said that gives you a good feeling, but this summary has been decided by someone.

00:19:20This is already a conscious design of the time that has happened, because it keeps

00:19:27you further on the screen. It gives you this feeling of being a kind of human labored being, without

00:19:32actually still understanding what is behind all these messages, because it is

00:19:35loading approximately at the level of maximum malice, and in the background, the textures are being loaded.

00:19:40But sometimes there are cool surprises when the AI considers

00:19:43the maximum malice of the data jubilee opening up. Yes, that’s quite amusing.

00:19:46So, there has been a certain design happening that should continue to give you the trust that something is occurring.

00:19:54But they are not yet playing with saying,

00:19:58and how long will it still take for what you see. So this topic, I can’t yet assess as a human

00:20:04because I have this 98 percent loading bar, and I don’t know how long the last two percent will take.

00:20:09It could still take a moment, it could take two minutes, or even an hour.

00:20:12And I believe there’s still so much potential in this, if we further develop the AIs, that they at least could communicate yesterday’s etiquette.

00:20:20How much time would still be left, because then one could certainly use time more productively.

00:20:24Although using time more productively is also a topic.

00:20:27You start and hit Enter, maybe you have a point that you trust now.

00:20:31We just had that.

00:20:32So what do you do now?

00:20:34You can have a coffee.

00:20:35Well, at some point, the caffeine consumption will be high too.

00:20:37You might also want to be productive.

00:20:39You do have a work life where your employer has a claim to that,

00:20:43the coffee machine doesn't need to be used more just because you're working with AI.

00:20:47What do you do? Okay, maybe you have other tasks or further AI topics.

00:20:52I actually find that really difficult. Keyword Mental Outlaw, when you stand there and

00:20:56say, you better pay attention. I have a project that's still processing. Recently, we had that

00:21:00Topic, which interface it is. The interface that I have in the company now with Graphed Agents,

00:21:05is just a list of projects. You see, this is a project that is wild by you,

00:21:09a blue bubble, but if you start and say,

00:21:11well, okay, he's working here, so I'll let him write the

00:21:14documentation, this one can summarize, this one can do this,

00:21:18that, and the other, who knows, so you have about five, six, seven

00:21:20things, after you have many. In the best-case scenario, you get a call,

00:21:25come urgently, here, Mark, take a look at this, then you come

00:21:28back to your computer, all agents are done and you stand there and

00:21:32think, what have I actually commissioned? Where are we

00:21:36actually. Plus the theater thing, what you just said, you start something and say,

00:21:40is it even worth starting? Yes, even if you actively decide to,

00:21:45to start a second point and maybe a post is on that post,

00:21:49that you're stuck on the monitor, that you, what you actually intended, you don't even know how long

00:21:53you need. Yes, I think that's a bit of, well, that’s one of the challenges,

00:21:58in my opinion, in designing the future world of work, this interplay

00:22:04between AI and humans.

00:22:06Now let's leave this simple level aside and not bring in any more actors yet. The question is how can it basically

00:22:13be designed in such a way that I, as a human, trust,

00:22:17that the results produced by the AI are correct, that I have a sense of when I need to

00:22:25most likely interact with it again in my daily routine,

00:22:29to either be the human envelope, envelope, whatever it may be at that moment.

00:22:33Or to make corrections, realizing that perhaps the prompt, the skill, whatever I did beforehand, to instruct the other work step, was not entirely correct.

00:22:42Or basically just to make a review to say, okay, here is a bit of a step achieved, to then move forward in the moment.

00:22:48And I believe that's an essential part, because, as you mentioned with the moment load, when I start doing many things in parallel, I think the frequency of errors massively increases.

00:23:00massively. If I now start five, six AIs in parallel to be particularly efficient,

00:23:06there will certainly come a point where I am overwhelmed, when I no longer

00:23:10know where each individual AI agent is actually at. If I have a status

00:23:14of that thing, I might try to copy the prompt into the other,

00:23:18that’s also another copy and paste, which I use very, very frequently

00:23:21at the moment when I am moving things between AIs and the internet. I

00:23:25believe we should really see how we can manage this both from the user interface side,

00:23:31that the AI offers, as well as how the workflow, how the office must be designed, how the workflow

00:23:37during the day must be organized, how notifications happen, how messages come to me,

00:23:42in this already intense stream of emails, Teams messages, notifications

00:23:49from AIs, maybe from multiple AIs, that are overwhelming us at that moment, like

00:23:53it is still manageable and we can simply be that human envelope

00:23:58and not just the hurried one who merely glances at it and then

00:24:02only delights in the funny loading message instead of being able to properly comprehend what is

00:24:07actually happening in the system. I have memories of the past and I had

00:24:13a Palm Pilot back in the day and I think the software was called Agendus, which was software that

00:24:18managed tasks and calendars, and I celebrated it back then and

00:24:23I have never found anything as cool since, that if you have things that you do not

00:24:27tackle in a day, those to-dos were displayed in the calendar,

00:24:31but they moved with you until you checked them off. And I would really

00:24:35actually like, regardless of whether one can change that superficially per

00:24:39Sebas. We also discussed the topic of whether one could do so in game interfaces,

00:24:423D landscapes, I wish for two things from these interfaces. One,

00:24:46that in this whole discussion about memories and code in the thing, somehow brings together,

00:24:51give me a wrap-up of what I did today, where do you think

00:24:55I am not finished yet? And secondly, that we somehow

00:25:00manage to combine contexts. So, whatever, I spoke about the same topic in two different

00:25:04conversations, whether I did that out of intention or forgetfulness

00:25:09is beside the point, but that I can somehow say, hey look

00:25:12I have a concept here, I have a context there, just like one can with Claude Cote, you can also say,

00:25:16uh Claude Covac I mean, take my data from JetGPT and you get a prompt and you

00:25:21could sort of do something like that, yes those are two things that I find totally interesting and

00:25:25Last point on this, even if we say, okay, you can't start something in parallel maybe

00:25:30or start too much, in such times when you're waiting, one could also

00:25:34make a very good review of topics where someone else might need a human review

00:25:39review, yes, if I give something to the AI and you give something to the AI and we're both waiting,

00:25:44then it could well be a good idea for future process design at the workplace,

00:25:49if you have or need a human in the loop or some human expertise, because you

00:25:56might think, oh come on, the market is crazy, the Apple user, I need some Apple feedback here

00:26:00and here's the crazy UX person, who can explain to the market how this works with these

00:26:05time surface so that it doesn’t get boring, and then each expert can also

00:26:10step in and maybe give approvals, write opinions, send prompts out again

00:26:16and also, in a professional context, you could streamline processes like a workflow not

00:26:22only by chasing agents, but also through a horde of people, let’s say, and utilize the

00:26:27time that the colleague might otherwise have spent waiting in front of the

00:26:31monitor and not have to get so stuck, because the system

00:26:36manages your time. The system takes you by the hand. The problem is not,

00:26:40that you are working on something else. The problem is that you forget it and

00:26:44that you might need incoming assistance when a disturbance comes in,

00:26:49which it then channels. Yes, but for that, you’d actually need a

00:26:53simple step like a simple chatbot with attached simplifications. But

00:27:00first, there has to be a basic understanding of time because I don’t know if you’ve noticed

00:27:04that when you’re just chatting normally or prompting with an AI, it simply has

00:27:09no sense of how much time has passed since the last prompt input.

00:27:14Oh God, when he then says, I've researched for you for two hours, and you

00:27:17say, no, it wasn't even two minutes, you fool.

00:27:21I find that quite terrible, it's never been seen before, but I

00:27:24had examples like this where I say, I was writing about something

00:27:27and then said, well, I'm not continuing with that over the weekend, and I had someone

00:27:31come Monday evening to sit down again and talk to my bot, and then he says,

00:27:37yes, nice, that, or no, he says something like, he says, yes, you want to

00:27:43go into the weekend now, we’re not continuing anymore, and I say, the weekend is

00:27:48long gone, please get used to checking a current time

00:27:53regularly or querying from the internet, for me or wherever else, so that

00:27:58you know what time period I'm currently talking about, where are we now? So simple

00:28:03things like, if I have a travel plan and say, let's check the place, what the

00:28:07weather is like there next week, then the AI should first check, okay,

00:28:12what day is it today, so I know what next week is? That is something

00:28:15you would do in such a moment, but if you reference that two or three prompts later

00:28:19and a week has passed, it hasn’t understood that a week has passed.

00:28:22So at least in the system prompt, asking for the current time

00:28:30and date at least once, has somehow been forgotten until now. So I’ve

00:28:33rarely experienced this, I always have to explicitly say it.

00:28:35I also have to remind him that when I say, give me, what has happened that’s new

00:28:39and such, that I have to explicitly tell him and remember, today is this and that,

00:28:43this and that time, the time is usually still irrelevant.

00:28:46But if I don’t tell him when something happened, then he tends to come up again

00:28:50the corner and say you, fascinating, today GPT-4 was introduced and you think, first of all,

00:28:56you are yourself GPT and secondly you may notice that you are perhaps one

00:29:00version ahead and thirdly, this is, let's say, in times of IT

00:29:05time is very fast-paced, but in times of AI it is even much worse.

00:29:11That feels like an eternity ago, I was still young then. Yes, I mean, honestly,

00:29:16we need to, I mean a simple question of time is not,

00:29:19something that would somehow consume a lot of processing power, so the fact that this isn't built in from the start is an absolute

00:29:24mystery to me and incomprehensible because, as I said, it works if you say

00:29:29research current news from the last three days, because then, basically, that seems to be a trigger for the AI to really look up,

00:29:36what is current in the background and then actually verifies things out there. But this, this simple functionality,

00:29:43of course is lacking to even have any awareness about it,

00:29:47is the person perhaps even still there at that moment? Is it perhaps night?

00:29:52If I were to come around the corner with my answer or the notification that I'm now

00:29:56finished with my research, finished with the programming code, so what we have in

00:30:01our communication, so time is one of the essential themes, that our

00:30:06world operates in conjunction because we can arrange meetings at a certain time,

00:30:09because the airplane lands at a certain time, because the supply chain is maintained in that way,

00:30:14they were there on a certain day, so this whole complex world we are now in and in which we can live and in which we can implement AI, only works because we have a concept and an awareness of time.

00:30:28That’s exactly the other example, I don’t know, but it was about time. The car wash is two minutes away.

00:30:36Should I be driving or walking, running after the car only makes sense if the car is already there, driving makes sense because you want to wash the car.

00:30:44Here comes the AI or depending on which model you have, it also likes to get mixed up.

00:30:48from the side, you can see that too, but while you were just speaking, it occurred to me

00:30:54that you have also been using your crab, so Open Clan, for a few days now

00:31:00and you mentioned a bit about how you set something up

00:31:04that it communicates with other services, with other models, and that it

00:31:09will certainly also encounter a waiting situation.

00:31:12So let’s switch back to the fact that it’s not just people on machines

00:31:17or even people. If there are people who are late, you should do so in times of

00:31:21Mobile phones have even increased; in the past, you would meet up, and if someone wasn't there, they lost.

00:31:25Today, you plan to meet and a second later you get a message saying, I can't make it today,

00:31:30Sorry, but it doesn't matter. Machine to machine. We sit there waiting and managing time?

00:31:35But there's a similar problem in the AI world; since it happens in interaction with

00:31:43humans already. So I enter my prompt, I wait for the

00:31:47Research, it can be done in 20 seconds, in 1 second, or in 3 minutes.

00:31:52You can also see this when AI agents give each other tasks.

00:31:55But even there, this concept of time has not really found its place yet.

00:31:59It's somewhat tied to Open Cloud, like a hard beat.

00:32:03So there is an internal clock that reminds me of something, that

00:32:06I initiate something new. That is there. I would like that too.

00:32:09Yes, so it's a bit of the habit I have, but you have the habit, that's what drives

00:32:13me forward, but it does nothing else than to say, now another beat

00:32:16has passed, let’s do something new, but this moment when one AI asks another AI something

00:32:23or needs something from another AI, time is not yet a topic there,

00:32:27it’s just waiting at the moment and just like it used to be in IT

00:32:31a lot when I would request an API, I need a microservice.

00:32:36for something to design something, if it hasn't responded in a certain number of milliseconds

00:32:41then I just canceled it and maybe requested a different service,

00:32:46so that I could still produce the content. If I imagine this in the AI to AI communication

00:32:51then it means, yes, as long as I don't do it consciously, things just happen

00:32:58that might not be beneficial at all, because it can certainly

00:33:01be the case that this one AI might also not have sufficient server capacity to

00:33:06really answer perfectly or due to its reasoning or due to the task that is posed

00:33:11it might just really take a while.

00:33:12That can then simply take a few hours.

00:33:14And the other AI also has to deal with that waiting time.

00:33:17It could certainly use its time better instead of just waiting for the result of the other.

00:33:22To wait for AI.

00:33:23Instead of waiting passively, I once found a nice quote from someone I

00:33:26met. I wait. Oh, you’re waiting? Yes, but I’m waiting actively. Okay, because just now you

00:33:31said, like, it sounds at least like it used to, right? Back in the day and in your hometown of

00:33:36APIs. It wasn’t that long ago, depending on how far you count back in time.

00:33:39Namely, before we started pretending to be agents here, I was a proponent of

00:33:46N8N. And with N8N, I felt like I built all sorts of things and at some point I

00:33:52realized that it’s actually a really great idea to build Vibecoiling frontends

00:33:57and attach N8N workflows to them, and then you find yourself and you give some

00:34:02complex tasks to the N8N workflow and then you realize, oh yes, damn it, you can’t

00:34:07even set the timeout, and somehow just one minute before, the thing just does

00:34:11closes the channel and says error, the workflow continues to operate but cannot deliver its feedback

00:34:16to the frontend anymore because the channel is closed, you're standing there thinking

00:34:20and so this ominous modern technology. I'm here, have agentic workflows, have great

00:34:26interfaces, but I can't seem to teach this thing, just wait a little longer.

00:34:32These are the moments when you sometimes sit there and think, yes friends of the night, what is

00:34:37this again? What would be a typical human interaction when I

00:34:41sit in a room or if my work would take a bit longer, that I would then

00:34:45actively, so to speak, tell my boss, my colleagues, or whoever is at the checkout, who is just

00:34:51waiting, that it takes a little longer, we have a lot to do right now, or something like that.

00:34:55Or I look forward to it.

00:34:56When the boss comes next time, greets as he goes out, comes in and says it takes

00:34:59a little longer, so that we don't work in a timeout now, I turn around

00:35:03and just keep working.

00:35:04I should try that, let's see how it feels in the real world with handshakes and acknowledgments

00:35:09and water in the protocol world, how that feels in real life.

00:35:13Exactly, but we have created these complex systems in the working world, where we have dependencies with each other, because we rarely do everything alone.

00:35:24And we manage that by making an agreement on how long something takes, when a project is finished, when a handover point is, whether I am now working with a sprint logic, whatever methodology I apply,

00:35:35so that in principle this issue of time can be estimated better and better and others can rely on the fact that at a certain point something will be finished.

00:35:42and this reliability. So this trust, now in this case regarding the time

00:35:48when a handover takes place. This has not yet been really designed in the agentic world, or I just don't see it being really designed yet.

00:35:57And if we expand this issue further, it can become a problem because there is also simply a huge potential for error in a

00:36:05communication when I imagine that 10,000 agentic systems have to communicate with each other to fulfill a certain task.

00:36:12When one of them waits so long and the other 9,900 are there in the

00:36:18Idol-time sinking and having a joyful holiday and not being able to continue working,

00:36:22because no feedback comes out of the system, that's just not in the spirit

00:36:26of the inventor probably. No agent under this number. That is very nice.

00:36:31When we planned the episode, so in a public round we had some

00:36:40food, drinks, and thought about what we discussed or want to discuss

00:36:43today.

00:36:44We said we want to ask at the end of an episode, to better engage our listeners,

00:36:49what we learned from the episodes because also

00:36:55we notice every time that when we hold episodes, the input that either

00:37:02arises from the conversation or simply due to the fact that when

00:37:07we agree on topics that either come from current events or from the

00:37:12The professor of each individual, that you learn things there.

00:37:15On the page, you brought in a point today and I want to take the opportunity

00:37:19first, I'll let you speak. Jens, did you learn something today or was

00:37:24it actually quite boring for you and are you just happy to tell me something?

00:37:27explained? That would be the emergency exit if you don't set anything up,

00:37:30but what do you take with you today? So because I am also a

00:37:34talk thinker, I always take something with me in our conversations. Now, it is of course

00:37:38correctly the case that I've been dealing with this topic for a while now.

00:37:42But of course this is within our discussion, this topic that one must also consciously design time in the

00:37:50workplace, in the interplay with agentic systems,

00:37:56so that there is actually no conscious loss of productivity. Because I am on

00:38:01on one hand, I can have an incredible increase in productivity when I delegate certain tasks to the AI,

00:38:05but if it just leads me to sit mindlessly in front of the monitor,

00:38:10then part of this productivity gain will be consumed again.

00:38:15And I believe that's simply an important aspect. I had sensed it a bit beforehand,

00:38:18but somehow it became clearer to me through our conversation,

00:38:21that you can't just address this from the UI side,

00:38:26but you really have to look at how work will function in the future,

00:38:31and organizationally. So you really have to take a systemic view of how it can be designed consciously, because we can't assume that every UI we will see in the future will inherently consider this issue. So we create situations, and I believe we also have to enable the people who work with this system at the same time. We need to handle this consciously.

00:38:53Sure. Also, the question back to you, Mark. Did you learn something?

00:38:58I suspect a lot, though I didn't really calculate it.

00:39:00So I'd be lying if I said I knew the term beforehand.

00:39:06You had suggested the topic before.

00:39:09I was already a bit anxious inside,

00:39:11because I thought, oh my God, what is that? I won't be able to say anything about it!

00:39:14I learned today that one can be intentionally sent through areas at the airport.

00:39:18I found that very cool right now.

00:39:20Thank you.

00:39:21But you are satisfied.

00:39:22Yes, it's great.

00:39:23I can already feel the happiness for the next time in advance.

00:39:26That's basically the anticipation.

00:39:27Anticipation. It’s like, once you know it, then it’s not there anymore. It’s

00:39:29like, sorry, everyone listening to the episode. It’s like, by the way, the Easter Bunny exists.

00:39:33not. Since then, the anticipation of my children is also different. What I actually still

00:39:37I will take some time to delve deeper into this after the episode, it is a topic, what can one do against

00:39:42do this mental overload while working. Although this doesn't really have anything to do with time management

00:39:46in that sense perhaps, it does relate to the fact that one

00:39:50can also be tricky. I'm trying something out, trying something out, trying

00:39:52something else. So I would like to take another look, so that I can refine it a bit more

00:39:54and the second thing is really to go deeper with the glasses on when

00:40:00we talk about interfaces because maybe in the future corporate AI is, you know

00:40:06always a keyword, right? So how do you integrate models so that

00:40:09people in the company can perhaps work on a large scale tailored to their

00:40:13work through skills and I don't know what to work with.

00:40:17what can you do if there are feedback loops, how do you get

00:40:21experts involved, how can you make waiting time into active waiting time, so that you

00:40:28certainly, okay, I'm looking at what the machine is doing, but I rely, as I said, on

00:40:33what I've calculated now. Before people start doing five different things, it's important that they

00:40:36maybe get directed into a control function or something else, those are points,

00:40:41that I think I would definitely carry with me, where I will definitely

00:40:46lose a thought or two. So, then let's not grumble, as bad as when

00:40:54we're remote, but I have to say, what I like about being present is that we feel

00:40:59less interrupted because the latencies that usually occur over video disappear when you see your

00:41:05counterpart in person with much less latency. So I enjoy it more this way, but well,

00:41:10okay. We can check if we can do this again. Then it would also be

00:41:14It was fun to see you in person while we recorded this episode. I think we should

00:41:18take along the idea that conscious time management, time is a totally valuable resource that we have as humans

00:41:25. I believe it is negligent if we, I don't even know how to say, at the beginning of this

00:41:31AI era, well we are almost right in the middle of it, not to address this topic from the start.

00:41:35This is really a conscious approach that can increase productivity,

00:41:42prevent mistakes, and generate higher satisfaction. I think that is the

00:41:48thing that we should tackle, and we can conclude with a quote. Yes,

00:41:57you need to read it. You know, I can't see well with my eyes right now. You need to go ahead

00:42:01and then I will follow with the explanations. So Benjamin Franklin, I believe, said back then,

00:42:06it's on his monitor. I once said that it’s always nice.

00:42:10Yes, I believe I have that feeling. I remember, in my deep education box

00:42:15I'll just check. Right after the bio basics course.

00:42:18So, I did pay attention in school back then. No idea if that quote ever came

00:42:22up in school. But he said, Lost Time is never found again. I find that

00:42:27very beautiful. And with that, I would like to invite everyone who enjoyed this podcast and how we

00:42:34shaped this time to share what we are talking about here.

00:42:41We would be super happy to receive comments and likes. I told Jens already,

00:42:46yes, I was recently proud to have a cyclist stop next to me who said,

00:42:51yes, great podcast, and then he rode off. At this point, greetings to the great

00:42:57unknown, yes, with the bicycle, with the sporty bike, with the racing bike, so much

00:43:01has been said. It’s anonymous enough because I really don’t know who it is. Thank you for being

00:43:06there, Jens. Thank you for letting us record this beautiful

00:43:10podcast with a nice cold drink and until next time. Ciao. Until next time.

00:43:16Welcome to ThinkDifferent, ThinkAI, the podcast by Mark and Jens.

00:43:24Two technology-loving minds who not only talk about artificial intelligence, but live it. Here you get

00:43:29clear categorizations, real practical insights, and a fresh look at what is possible.

00:43:35Understandable, critical, and always with a wink.

00:43:39AI to ponder, to smile about, and above all, to discuss.