Latest / Unresolved Dependencies / #12 Can GitHub Copilot Save Copilot Studio? 👩🏼‍⚕️

Transcript

#12 Can GitHub Copilot Save Copilot Studio? 👩🏼‍⚕️ · Unresolved Dependencies

• 0:02
  unknown: God fuck you.
• 0:04
  SPEAKER_00: Joe, good morning.
• 0:07
  SPEAKER_00: Joseph.
• 0:09
  SPEAKER_00: Good morning, Joe.
• 0:10
  SPEAKER_00: It's a new episode of Unresolved Dependencies.
• 0:14
  SPEAKER_00: Wakey Wakey.
• 0:18
  SPEAKER_01: Okay.
• 0:18
  SPEAKER_00: No, it's it's it's better than a dream.
• 0:21
  SPEAKER_00: It's Unresolved Dependencies.
• 0:29
  SPEAKER_01: Yeah, I'm sorry.
• 0:31
  SPEAKER_01: I'm a bit I'm a bit tired today.
• 0:32
  SPEAKER_01: Do you want to know why I'm tired?
• 0:34
  SPEAKER_00: Tell me, of course, why are you tired?
• 0:40
  SPEAKER_01: It is uh 7 a.m.
• 0:43
  SPEAKER_01: UK time here, and I stupidly agreed to do training for a client in Australia time zone, which means I've had to be up at midnight and between 1 and 5 am in the morning.
• 0:52
  SPEAKER_01: So I'm a little bit behind on my sleep.
• 0:55
  SPEAKER_01: I need my sleep to function.
• 0:56
  SPEAKER_00: Um literally, you you you already have your working day behind you now.
• 1:04
  SPEAKER_00: But you have a new UK working day in front of you.
• 1:08
  SPEAKER_01: This is the dilemma.
• 1:09
  SPEAKER_01: Do I do I do I power on through or do I go straight back to bed after this?
• 1:13
  SPEAKER_01: This is the this is the dilemma.
• 1:14
  SPEAKER_00: So it's my personal opinion as not as not as your help coach.
• 1:19
  SPEAKER_00: I would say go to sleep.
• 1:22
  SPEAKER_01: I think I might do, but you know, the pop but the podcast has to continue.
• 1:26
  SPEAKER_01: This is this is kind of mandatory, we have to do it.
• 1:28
  SPEAKER_00: So well, honestly, how are you doing?
• 1:31
  SPEAKER_01: How are you doing today anyway?
• 1:32
  SPEAKER_00: We're recording that early because you can't do it tomorrow, so so it's me.
• 1:38
  SPEAKER_01: Again, it's a me problem.
• 1:40
  SPEAKER_01: It's always a me problem, you know.
• 1:41
  SPEAKER_00: So what are you doing tomorrow?
• 1:44
  SPEAKER_01: Ah, well, we'll find out in a few moments when we do a feature of Joe's travel.
• 1:49
  SPEAKER_00: You're traveling, right?
• 1:50
  SPEAKER_00: I'm traveling.
• 1:52
  SPEAKER_01: I mean, it's a stupid question, of course.
• 1:54
  SPEAKER_01: You just assume that I'm gonna be traveling, and you're probably gonna be correct, yeah.
• 1:58
  SPEAKER_00: So yes, right, so it's somewhat your fault.
• 2:02
  SPEAKER_01: Exactly, exactly.
• 2:03
  SPEAKER_01: But yeah, speaking of Joe's travels, let's kick that off.
• 2:05
  SPEAKER_01: Um, so yeah, last time we did um uh I guess again, we the inbox has been absolutely smashful with responses.
• 2:14
  SPEAKER_01: Um, you know, people find this feature so popular.
• 2:16
  SPEAKER_01: So if you were playing and you just you know didn't have the time to email in, you know, why not?
• 2:21
  SPEAKER_01: You know, just email in.
• 2:22
  SPEAKER_01: Um, the answer was uh Lubeck.
• 2:25
  SPEAKER_01: I was in the lovely city of Lubeck.
• 2:28
  SPEAKER_01: Um and uh it was for work, of course.
• 2:31
  SPEAKER_01: Uh but a good but a shout out to Bjorn who gave me a great tip.
• 2:36
  SPEAKER_01: He saw I was there, he said, okay, you need to go and check out the Marzipan.
• 2:40
  SPEAKER_01: Lubeck is actually famous for Marzipan.
• 2:42
  SPEAKER_01: So I stocked up on quite a bit of very tasty, lovely marzipan.
• 2:47
  SPEAKER_00: Are you a marzipan fan?
• 2:50
  SPEAKER_01: I like anything that's got you know chocolate, sugar in, you know.
• 2:53
  SPEAKER_01: Maybe, maybe if I had some sugar now, I'd be I'd be like that would be a good start in the day, right?
• 2:58
  SPEAKER_00: Very looks healthy.
• 3:00
  SPEAKER_00: But yeah, why why I'm surprised a British guy likes marzipan.
• 3:04
  SPEAKER_00: So yeah.
• 3:05
  SPEAKER_01: Exactly.
• 3:06
  SPEAKER_01: If it's got sugar in, I'm there probably.
• 3:07
  SPEAKER_01: So yeah.
• 3:08
  SPEAKER_01: So that was the answer from last time.
• 3:09
  SPEAKER_01: So if you are interested in taking it getting involved, then you let us know on the email.
• 3:13
  SPEAKER_01: Hello at Unresolved Dependencies.com.
• 3:15
  SPEAKER_01: Let us know on the socials, we'll give you a nice little shout out.
• 3:17
  SPEAKER_01: And there's also some vague promises eventually of some kind of uh merch.
• 3:22
  SPEAKER_01: Uh so yeah, so as mentioned I'm traveling again, so where am I going to be next?
• 3:27
  SPEAKER_01: The clue this time is I'm going to a city that is famous for a tea-related celebration, uh, and where a famous radio psychiatrist used to go to a bar where everyone knew his name.
• 3:41
  SPEAKER_01: Where am I talking about?
• 3:43
  SPEAKER_00: So you didn't say pop, therefore, so a bar and tea and England does not go well together.
• 3:55
  SPEAKER_01: Yeah, you but you kind of get him beyond the right thought, beyond the on the right thought pattern, though.
• 4:00
  SPEAKER_01: So, but yeah, but yeah, have a think yourself.
• 4:02
  SPEAKER_01: Let us know uh in via email hello ohmsoredpenses.com.
• 4:06
  SPEAKER_01: Uh, we'll be glad to hear from you, and yeah, we'll give you a nice little shout out.
• 4:10
  SPEAKER_01: Why not?
• 4:12
  SPEAKER_00: Awesome.
• 4:12
  SPEAKER_00: Yeah, I'm very curious where you're going.
• 4:16
  SPEAKER_01: Yeah, so it should be good.
• 4:17
  SPEAKER_01: But you know what goes really well with um uh beverages at parties, food, particularly Sebastian's food, Sebastian's eats.
• 4:25
  SPEAKER_01: So tell me Sebastian.
• 4:27
  SPEAKER_01: What have you been eating?
• 4:28
  SPEAKER_01: I I need to know what you're eating, it's essential information for me.
• 4:34
  SPEAKER_00: It's especially essential if you start your day and you want to just figure out what happens today and what do you want to eat.
• 4:41
  SPEAKER_01: Um, I'm on my last low needs of hierarchy.
• 4:43
  SPEAKER_01: I'm sort of like sleep, food thing.
• 4:46
  SPEAKER_01: Tell me what you're eating now.
• 4:47
  SPEAKER_01: I need to know.
• 4:48
  SPEAKER_01: Yes.
• 4:48
  SPEAKER_00: I'm paralyzed of decisions at this point.
• 4:51
  SPEAKER_00: Exactly.
• 4:52
  SPEAKER_00: Yeah, basically, so this time, um, basically it was curry.
• 4:57
  SPEAKER_00: Uh, it was sorry, rice, chicken, curry, vegetables, and uh, it was made by my girlfriend, very, very nice of her.
• 5:05
  SPEAKER_00: So the the only yeah, it was the only issue was that I didn't snap a picture for Instagram.
• 5:14
  SPEAKER_00: So my my food blocking style or my fluke blocking skills kind of vanishing since uh it is uh much more summertime here.
• 5:26
  SPEAKER_00: But yeah.
• 5:32
  SPEAKER_00: I I I directly stuck my hat into it.
• 5:35
  SPEAKER_00: No, but it was very nice.
• 5:36
  SPEAKER_00: The the fun part is because I'm I'm um how to say it is nicely privileged asshole.
• 5:46
  SPEAKER_00: I I had a chance as well to go to travel to India sometimes in the past, so I the curry was made with Indian curry, which was super spicy, and I'm not that spicy eater, but uh luckily my my late run to the nearest supermarket um saved us a little bit here because then we could as well um put on different spices because all the curry-related spices I had at home were like straight out of India.
• 6:16
  SPEAKER_00: So instead of like um still burning and um uh getting flamed by the by the meat, it was very, very delicious, so very nice curry.
• 6:28
  SPEAKER_01: My best tip if you for spicy food, get some yogurt, some sort of normal plain white yogurt here, and that'll be that would sort it out in no time.
• 6:36
  SPEAKER_00: Quark.
• 6:38
  SPEAKER_00: Speaking of quark, shootout to the best breakfast on the earth, which is quark, berries, and some oat oats and some milk.
• 6:49
  SPEAKER_00: Best best breakfast on earth.
• 6:51
  SPEAKER_00: Yeah, I don't argue about this.
• 6:53
  SPEAKER_00: I'm eating this since 10 years every day.
• 6:58
  SPEAKER_01: Is that is that good food for like weight training?
• 7:01
  SPEAKER_01: Would you say?
• 7:03
  SPEAKER_00: Yeah, so quark quark, so especially like the fat-free mark, uh fat-free quark, um, has some proteins, but also with the berries and so on, it's it's quite nice.
• 7:14
  SPEAKER_00: Oats, anyways, is there small protein source and as well.
• 7:20
  SPEAKER_00: Um, I don't know what the word is, it's ballaststoffer.
• 7:24
  SPEAKER_00: So I um it it's good for your your diet in general, and um yeah, then you can just add whatever you like to.
• 7:33
  SPEAKER_00: For example, I add some um whey protein powder just for the taste, so just a scoop of it, maybe one and a half, and for yourself, why not?
• 7:45
  SPEAKER_00: You know, yeah.
• 7:46
  SPEAKER_00: Well, well, well, actually, it just helps me to get to the proteins which I need to consume a day much more quicker and more free of choice during the day in terms of what I eat.
• 7:58
  SPEAKER_00: Um, therefore, I I tend to have a very heavy protein meal in the morning, and then as well as well, like it's nice because it just saturates you for like half of the day, so you don't get any hunger randomly at 10:30 at a day.
• 8:14
  SPEAKER_00: So super nice.
• 8:16
  SPEAKER_01: That's my problem.
• 8:17
  SPEAKER_01: I think I eat it's my own, but it's again, it's my own fault.
• 8:19
  SPEAKER_01: I eat too much carbs, and carbs then make you growly hungry.
• 8:22
  SPEAKER_01: So, and my my stomach is very, very audible.
• 8:25
  SPEAKER_01: I'll literally be sat there and it'll just be like making that growling.
• 8:31
  SPEAKER_01: I mean, literally, literally, and I'm just I haven't say sorry, it's it's not my fault, it's my stomach.
• 8:38
  SPEAKER_01: It is my stomach making that noise, just make that clear, first of all.
• 8:41
  SPEAKER_01: So, yeah, it's it's definitely it's good to have something.
• 8:44
  SPEAKER_01: I think porridge can be quite good in the morning just to give you that base.
• 8:48
  SPEAKER_01: You need a base to make you last until like lunch or whatever.
• 8:51
  SPEAKER_00: So what is surprisingly not that good in the morning is sugar.
• 8:55
  SPEAKER_01: Yeah, that I ate too much, I ate too much sugar, you know.
• 8:58
  SPEAKER_01: I need to, I need to, I I slightly I did quite well with a diet last year.
• 9:01
  SPEAKER_01: I need to get back into more of a uh a nicer thing, but yeah, summer is always quite tricky with what's happening and things like that.
• 9:07
  SPEAKER_00: I find yeah, especially also with traveling, to be honest.
• 9:10
  SPEAKER_00: Travel it traveling is for me like the the hardest part because on it so a lot of countries don't have quark at their hotel breakfast, for example, and I also don't want to rely on these like 50 grams of protein milkshake drinks you can buy in in supermarkets, for example, in the US.
• 9:28
  SPEAKER_00: So it's not a big fan of that.
• 9:30
  SPEAKER_00: But I also don't want to bring like my I don't know, my supplements to every place I go.
• 9:35
  SPEAKER_00: So I that's always traveling is always a bit tough, so you need to be very keep attention on it, and then you often go for for dinner or something like that, and it destroys your whole diet very often.
• 9:49
  SPEAKER_00: So that's like the hardest part for me because I'm I'm a I'm not a picky idiot.
• 9:54
  SPEAKER_00: I I could eat like the same things like whole day, whole week, whole month.
• 9:58
  SPEAKER_00: It's for me, this is like not a big problem.
• 10:02
  SPEAKER_00: I literally eat the same breakfast since 10 years, and and I have not an issue with that.
• 10:09
  SPEAKER_01: I had a great tip from and yeah, well worth a follow on YouTube uh Damien Bird, just lots of really amazing content.
• 10:15
  SPEAKER_01: So yeah, his trick is like the Huel Meal packs, you can sort of just you spoil it with some hot water, and then you get like a nice little curry thing.
• 10:22
  SPEAKER_01: Um, and yeah, if you particularly if you're managing your diet and you know, um set off by things like you know, gluten or other things, which are you because you don't if you don't want to be that awkward person, like you know, around the dinner table, just like, yeah, can I is it is it gluten-free?
• 10:36
  SPEAKER_01: Is it have you got the C that action?
• 10:38
  SPEAKER_01: Have you got uh have you got have you got have you got a pesca pescatarian option?
• 10:41
  SPEAKER_00: You know, you don't want to be you know you you feel like you don't want to be that person sometimes, particularly just be the awkward person that brings us chicken and rice to like every occasion and just eat it from your tub box.
• 10:53
  SPEAKER_01: Yeah, I mean that's a it's everyone knows you're like oh it's a chicken and rice guy, chicken rice, chicken rice.
• 10:57
  SPEAKER_01: Yep, there we go.
• 10:58
  SPEAKER_00: That's the dream, and then just eating from your bowl.
• 11:01
  SPEAKER_00: Yeah, that's a dream.
• 11:02
  SPEAKER_00: Also, what was a dream was our last episode and was a little bit awkward because on the day I think on the day, right, when we released the episode, or like just a few days later, Microsoft changed a little bit like how we think about Copolit Studio, right?
• 11:18
  SPEAKER_01: There was a nice um announcement, you know.
• 11:20
  SPEAKER_01: It's actually it was pretty good to see the announcement because it was a nice video announcement.
• 11:24
  SPEAKER_01: I think um Ryan did the whole uh video demo stuff like that.
• 11:27
  SPEAKER_01: It felt like that there was quite a bit of good, although we'll get into maybe some of the annoying aspects of it, perhaps um annoying inverted comments, but uh the the the way it was announced, the energy, I think it was I think it's also quite interesting.
• 11:40
  SPEAKER_01: There's gotta be, I think, the feature fastest to GA from public preview that I've seen ever.
• 11:46
  SPEAKER_01: And of course, we're talking about the a month, literally a month, and of course, we're talking about the what we can now call the GitHub Copilot Harness experience in Copilot Studio, you know, it just rolls off the tongue, doesn't it?
• 12:02
  SPEAKER_01: It's uh yeah, so beautiful.
• 12:05
  SPEAKER_01: So so basically is so everything from our previous episode is out of date, then pretty much.
• 12:09
  SPEAKER_00: Oh no, so because so most of it um is still true.
• 12:13
  SPEAKER_00: So everything in terms of the license consumption, how core work works, how GitHub Copilot itself works.
• 12:18
  SPEAKER_00: The only little thing that changes is that you now have two different experiences in Copilot Studio, and potentially the billing is a bit different.
• 12:28
  SPEAKER_00: So maybe we start with that reconciliation part of our last episode, and then you may can explain to the folks um what the actual new experience is, and I can explain the hard part right now.
• 12:42
  SPEAKER_00: Um, so what we told last week is essentially true.
• 12:47
  SPEAKER_00: Um, with Copolit Studio, you still have the classic interface, which is more of like your conversational agent, so a conversational structure.
• 12:55
  SPEAKER_00: So this is as well still in the same consumption rate as before, it consumes Copolit credits, unless your user um the end user is an M365 Copilot licensed user, then this is within the license.
• 13:10
  SPEAKER_00: So advise at this point if you do need to create an agent that is a very chatty experience, potentially about some documents, some internal rules, or something where um people need to reach out, and then there is an um agent uh or multiple agents that sits there and provide information, you still want to go with the classic experience because that chat experience definitely sits good there, and as well, it still supports the uh topics feature, which is um one of the features that didn't make the cut to uh the new GitHub Copilot Harness experience.
• 13:51
  SPEAKER_00: On the new experience, the licensing looks a little bit different, so there um because it is built on the GitHub Copilot Harness, um the general execution of agents will be consumption-based.
• 14:10
  SPEAKER_00: So it's pretty similar to covert, or more similar to covork, let's say it's like this.
• 14:14
  SPEAKER_00: Not like just every action you take, every step you make um will consume uh credits at the point.
• 14:21
  SPEAKER_00: The the system, like what consumes credits, is a bit easier, so it's a bit easier to follow like how much the consumption is now.
• 14:30
  SPEAKER_00: So, in terms of, for example, estimation, I think it's a bit easier.
• 14:33
  SPEAKER_00: Um, while there is a bit more control over that.
• 14:38
  SPEAKER_00: Still, the important thing is um all the agents you have built so far, um, and the agents you built since the announcement of GA, which is like the 7th August or something, um everything that was built till the 6th August will now be consumption free till I think 1st of September.
• 15:00
  SPEAKER_00: Um, and then this will be as well charged.
• 15:04
  SPEAKER_00: Um, everything which you built right now today in the new experience will be charged right away.
• 15:09
  SPEAKER_00: This also includes um the creation of an agent.
• 15:15
  SPEAKER_00: So before in the classic experience, and this is maybe the biggest lapdown at the at the point of being right now, that when I create a new agent in the new Copilot Studio experience, which is GitHub Copilot Harness.
• 15:34
  SPEAKER_00: Harness is by the way, the already word of harness harness harness.
• 15:41
  SPEAKER_01: Um put some emphasis in it, harness.
• 15:44
  SPEAKER_01: Yeah, you can have some fun with it.
• 15:46
  SPEAKER_00: I really hate that word.
• 15:48
  SPEAKER_00: Um, but this new build agent this will consume credits from the very first execution, so not from published, no, not after publishing, how it was before.
• 16:01
  SPEAKER_00: So that you build your agent, you test the agent, and you make adjustments, which is totally natural.
• 16:06
  SPEAKER_00: No one of us builds just agent right away that works in the first try.
• 16:11
  SPEAKER_00: You do not test it.
• 16:12
  SPEAKER_01: I usually I usually I usually do.
• 16:14
  SPEAKER_01: I mean, it's uh maybe a bit hard, so yeah.
• 16:17
  SPEAKER_00: Okay, you you you may be very gifted in this uh agentic world.
• 16:21
  SPEAKER_00: We can say, yeah.
• 16:23
  SPEAKER_00: So before it was that, like after um pressing publish, that will then consume credits if the situation appears that the end user is not M55 co-pilot license.
• 16:37
  SPEAKER_00: Right now it is the moment you click on create and you test your agent, this will consume credits.
• 16:45
  SPEAKER_00: And and this is like the the biggest caveat of that, or like the biggest like laturn of the of the new experience that like there is uh like a consumption cost for the makers, for the deaths, for people just trying out what the new experience can do.
• 17:02
  SPEAKER_00: Um so this is already like the the most important point.
• 17:08
  SPEAKER_00: And I I have not seen in the license guide any exclusion for developer environments as well.
• 17:14
  SPEAKER_00: So um it is just in general, the moment this agent um goes into testing, um, it doesn't need to have to be published, it will consume credits, it will cost, and you need to be aware of that before you um build that or make this hardness decision.
• 17:33
  SPEAKER_01: Okay, so that's a huge box you've just dumped there.
• 17:36
  SPEAKER_01: There's uh loads of stuff there to unpack.
• 17:38
  SPEAKER_01: Let's just focus on a few little bits there.
• 17:40
  SPEAKER_01: So you mentioned the word classic.
• 17:42
  SPEAKER_01: Now, classic, we usually we use to talk about anything that's kind of old.
• 17:46
  SPEAKER_01: What I think is kind of interesting when we're looking at the learn articles, we're talking about it being uh the standard harness.
• 17:52
  SPEAKER_01: And in fact, there's three harnesses that it mentions it's the GitHub Copilot Harness, the standard harness, which is what we sort of are used to there, and then we now got this concept of GitHub copilot chat harness, uh specifically for extending M365 copilot.
• 18:06
  SPEAKER_01: But I think it's interesting that they haven't used formally the words classic harness because usually, well, in our world, classic, what does that usually mean when it's being used?
• 18:16
  SPEAKER_01: It means it's old, it's old, and probably is going to be eventually for like you know the chop type thing.
• 18:23
  SPEAKER_01: So I mean, is it safe to build standard harness agents?
• 18:27
  SPEAKER_01: Do you think should we be telling people to do that?
• 18:29
  SPEAKER_00: No, so there is no official information or as well.
• 18:32
  SPEAKER_00: Honestly, there's also no NDA information about that.
• 18:36
  SPEAKER_00: So at the moment it's still safe, and there is also no immigration between your standard harness, as I just learned, to uh GitHub compiled harness experience.
• 18:46
  SPEAKER_00: So everything would you need to uh what you already have built in the in the standard experience, let's not call it classic anymore.
• 18:54
  SPEAKER_00: Um, and you need to gather to the new experience, you need to rebuild that.
• 19:00
  SPEAKER_00: But there is a we can potentially put this in the in the show notes, there is a um a skill for that.
• 19:08
  SPEAKER_00: Um basically that can do it with with GitHub Copilot.
• 19:12
  SPEAKER_00: So you can use that skill and then just connect to Copilot Studio via the GitHub or the Visual Studio connector, and and then just pull it from there and redeploy it with the skill.
• 19:25
  SPEAKER_01: So these new agents are also stored as YAML definitions similar to the old style agents.
• 19:31
  SPEAKER_01: Yeah, yeah.
• 19:32
  SPEAKER_00: So you can just interpret that interpret it and migrate to basically a new experience.
• 19:37
  SPEAKER_00: So but there is no button in Copolis 2G to click on and then oh yeah, do this to the you can type Copolit harness, please.
• 19:45
  SPEAKER_00: So yeah, this is not there, yeah.
• 19:48
  SPEAKER_00: But yeah, for right now, it is not going away.
• 19:50
  SPEAKER_01: Yeah, it feels like it's gonna, and it's because it still feels to me, there's still gonna need to be those scenarios where okay, I need to have you know determinism.
• 19:57
  SPEAKER_01: Deterministic is what I need in my life, I need to have full control over what my agent can sort of do.
• 20:03
  SPEAKER_01: Um, so that option is gonna be there.
• 20:05
  SPEAKER_01: So I you know, I would say I'm still maybe saying to people, okay, try and have a look at the GitHub copilot harness over the standard harness, but yeah, I think we're still going to be probably talking about well, both, if not all three of them, for quite some time now and into the future as well, perhaps.
• 20:22
  SPEAKER_00: Um I mean, definitely there, as I mentioned, there are use cases to use the um the standard harness over the GitHub copilot harness, which is very simple in terms of I want to have a conversational driven agent and I want to have the lifters in M55 Copilot or in Teams, and I have a bunch of licensed M55 Copilot users.
• 20:44
  SPEAKER_00: So if this is your use case, I would 100% go with the standard harness.
• 20:50
  SPEAKER_00: If I have something that, anyways, runs autonomously in the background, or is something that is as part of a um more complex model um of multi-agent or in terms of um needs to do some some action in the background in terms of updating the system or something, I would may think to go with the GitHub Copilot harness because it supports skills and memory and has some recent new features.
• 21:17
  SPEAKER_00: And as you know, skills is maybe something that is all over the um all over the bubble in the past months and weeks.
• 21:27
  SPEAKER_00: So meanwhile, I think with skills, everyone everyone had some touch points so far, and you can just use those skills now in in custom agents, which is great, and you can uh build a lot of automation based on your business processes, and you don't need to um you do don't need to know coding in terms of um having everything tracked, you can just um define it in skills perhaps and um give this to the agent.
• 21:57
  SPEAKER_01: It's what I think is also interesting in the article they do like.
• 22:00
  SPEAKER_01: of a table.
• 22:00
  SPEAKER_01: We'll link to this in the show notes and stuff like that.
• 22:02
  SPEAKER_01: But you know, it talks about like as you say skills and memory.
• 22:05
  SPEAKER_01: Okay, yeah, GitHub Copilot Harness uh supports that.
• 22:08
  SPEAKER_01: But then for the standard harness and copilot chat harness, it's kind of like this almost like you know, come hither sort of teasing sort of language like uh oh not a focus not a focus as if to sort of say like okay maybe maybe if you're good in the future.
• 22:23
  SPEAKER_01: So I'm just I'm just wondering this is this perhaps maybe the the biggest example of like testing in production that we've ever seen with this Google Copilot harness?
• 22:32
  SPEAKER_01: Is this going to be still a thing in six months or a year time?
• 22:35
  SPEAKER_00: It's I don't I don't think that is like something testing in production because the um what we have seen on GitHub Copilot in the past is already quite good and it's um uh follows a good adoption not only in in our organization but as well across multiple organizations and GitHub Copilot actually works very very well it works very well with agents with skills with different models so I there there is definitely they have done something right in the GitHub slash max of third quarters here um and there is already a success story with GitHub Copilot so so why not learning from what they have done or how they have implemented and why not have feature parry with um or for example what happens on foundry and what happens on GitHub so I so there is definitely definitely a need to to at least try that but the the past success of GitHub Copilot um gives them a huge favor and I I I don't think it's like something like testing production it's more like oh yeah from all the things we have done in the past 12 months or 25 or 24 months this architecture um has proven that works very good.
• 23:57
  SPEAKER_00: So we know adapting this for other scenarios as well and we need to change that and it's better sooner than later to do this change before the whole world just runs on agents and the internet is just full of agents and we're um very very close to to Terminator movies.
• 24:15
  SPEAKER_01: So making this change right now is perhaps enjoying the summer break yeah because no no piece working we're all just asleep aren't we so yeah yeah they they could have just announced it during the World Cup and no one would have noticed that exactly we would be still here on the same day talking about an Indian could it be could it be GitHub Copilot saves Copilot Studio could we be saying that in 12 months time you know this you know now we can we the meteoric rise of Copilot Studio above and beyond where it is now of course you know we we we can we can basically think GitHub you can definitely make the case because um if we if we think back like six months or so my impression of Copilot Studio was like yeah it's a nice tool to play around but um as soon as we we received like real business cases and um more complex scenarios and um different kinds of user inputs where it's just not streamlined always the same prompt or always the same situation.
• 25:16
  SPEAKER_00: We faced a lot of difficulties not only in publishing but also in like maintaining the agent and we had like very weird setups that we changed the whole architecture of like all the agents we build which is now again obsolete of course but never mind uh but like well and then I was like I I was saying and uh I think you can you may hear it like in one of the first episodes or something or at one of the the conference um sessions I I did in the earlier this year that like okay Copoly Studio is fine for like a certain case of of situations but if you think about more complex scenarios you definitely want to go with foundry today I would may revise this and say okay you you you have a case of Copol Studio now again where you have a um yeah like your low code first approach when you can write skills when you do have an idea how this works uh with all of the MCPs and so connected uh you would be able to to create your more complex agent as well in Codepoly Studio Microsoft Foundry of course with the Microsoft Agent Framework or any kind of other kind of frameworks still sits there and is still getting like this super super complex cases or code first approaches where we for example connect the agent call to to a website or something like that.
• 26:38
  SPEAKER_01: So basically what you you what you basically said is you you you friend zone copilot studio you're like oh you know you're really nice you're really great but I'm sorry it's it's it's it's not you it's not me you friend zoned it fast forward six months you sort of see copilot again you're coming to Copilot Studio view again and you're sort of going oh hello hello that's you basically the glow up is real I'm with the music box in the backyard of Copilot Studio and backing for my pardon so yeah the glow up is real okay well I hope that you succeed in that in in all of your endeavors there is all I was saying but let's let's talk about the final thing but build to create it seems it seems controversial but I don't see given the way the consumption costs are I don't see how there could have been any other option at the moment for that so it's kind of like we were it was a nice ride while it lasted is what I'm saying basically well yes but also no so on on the one side so far so from a from a economical point of view absolutely right decision it's just like okay there are so many people running this on on their def environments there they they potentially have all those numbers um on like a very fancy dashboard and and then they say okay we need to bring this we have so many costs and we have so many thousands of agents running in context of a deaf environment or they're not published and they just consume um messages during the creation phase and this is a high cost for us.
• 28:13
  SPEAKER_00: So from from that point of view it definitely um makes sense from now from my personal perspective and from like your maker perspective or for someone for example very different breed but doing workshops and trainings this is a challenge now because um suddenly everything what we do in our session or we do at a workshop of course this is just a very slow tier on the list of uh people consuming messages on on the power platform but this also now costs us money so um we will double down on like um what are the scenarios and how complex they are potentially um as well if you want to just explore co-poly studio and this was already my critique on covork you you need to find your use case on covork but to find that use case and to find that use case for the new co-poly studio experience you need to have enhanced on and try out there is no no simulation to get there and say oh yeah this is my perfect agent this is not existing so you need to have enhanced on you need to try you need to um to excuse pardon my friend you need to fuck around and until you figure out yes yes of course exactly yes it's a classic thing yeah yes and and this is how it works with like new technology and new stuff that happens and you can't do this without generating costs now and how how do you how do you argue for this cost?
• 29:44
  SPEAKER_01: We had like the this this month we had like one agent that consumed within four days a hundred thousand credits if you're not aware of that if you don't check the telemetrics on PowerPlatform admin center which is um bit lacking right now in terms of availability or for example the Microsoft admin center with agent 365 which is again a paid license uh you potentially generate an invoice at the end of the month if you don't set up your limits um that you don't want to see but Sebastian you know how badly Microsoft is shocking you saw the earnings report last year you know how can you morally be expected to take such a free ride moving forward you know you have to pay for this stuff you know Satya's children haven't eaten for many many months now I've been reliably told so you know you you have to pay now that that is just the way of the game we have to pay speaking a bit more seriously though um I just wonder whether there will be and we have we've spoken about this kind of uh not on the podcast specifically but whether there will be more of a push towards going local because everything always feels kind of cyclical in everything to do with IT you know we you know we there's we you know there was a whole demand you know back in the 80s okay let's go to have these remote machines that we connect to have these dummy terminals oh let's all go back local again oh let's all go back to the cloud and stuff like that you know yeah with these rising costs and these large language modeling could it be a point when we start to get to see this run on our local machines you know are we just building and developing locally do we just build and test and develop our agents locally and then just get them onto these cloud platforms after the fact so that that's a part of your character because I was just exploring last week's a lot of around um using local models and and deploying and what are differences and what we can do and um how for example things um do you run?
• 31:37
  SPEAKER_00: I would say not yet maybe for personal productivity maybe in terms of like you have a local agent um or you have a local setup where you can just quickly chat with it or something that runs locally on your machine that just um considers things which are on your machine um something like open claw in that direction and you kind of hook it up to messaging um and different mcps but still the the model needs to run as well locally so it does not only have the need to have your agent locally deployed but also in the model and this is I think where it falls short still a little bit the there is um I think you shared this there is a as well a model from from Germany that is quite good for a local model um there is well Gwen models which are quite good um Google has a own breed of local models I would as well for example um yeah I would like to see some local models for Microsoft Nvidia has them so there's a wide uh wide variety of local models but I think still like for that organizational wide rollout or for this organizational wide use case I think it's maybe still still too early and that whole hardware and infrastructure still costs that much so that's also why we have that high cost prices because all of these light energy water consuming data centers do consume so much energy and do cost that much to run.
• 33:14
  SPEAKER_00: So it's not just that the costs of AI are that high that everyone earns a ton of money and um builds up generational wealth but also because of the infrastructure costs are that high and you would need to carry out this um costs by yourself now internally and you need as well to be um you are responsible for the uptime you're responsible that everything is up to date you're responsible that everything is maintained over the the sake of time so these are all factors you shouldn't ignore when we're talking about local for personal productivity uh I I totally agree that this is definitely something to explore um in terms of um the the own if you like to do it because there's definitely some some barrier with the with the setup but with Olama for example it's quite easy um and you still can for example there's there's something called like something called there's called like uh the Hermes agent which you can spin up to like a telegram or something and then you can chat with it and I can response it you don't need to go for the console or the terminal.
• 34:27
  SPEAKER_00: So there's definitely options and I and I think this is the the models are good enough to do that for personal productivity.
• 34:36
  SPEAKER_01: Yeah I'll link to the one that I sent you um um uh made in Germany Sufi S you know a a true uh Wunderwaffen the best thing that Germany has done for quite some time I'm sure but yeah it looks it's receiving well let's see who knows maybe Germany saves the whole AI race you know in terms of making it affordable making it better you know all this sort of good stuff but yeah this is I think where we're increasingly going now um I think in terms of you know because I think there's gonna we're gonna reach a point where um you know things are just gonna break you know in terms of consumption to cost in terms of costs being too high all this sort of stuff I just feel like that we're gonna hit a point where people just go right I'm not paying that I still want to use it so I'm just gonna go local instead I get I get maybe 50% maybe of the of the same sort of benefit but I'm still able to manage my costs better.
• 35:27
  SPEAKER_01: So yeah definitely definitely worth exploring indeed so quite a week what what a what a week or what a day between publishing episodes you know happens.
• 35:39
  SPEAKER_01: Usually this is the this is the time of the year where we can just sit back and relax and just not worry about anything sort of um you know production breaking hitting our sort of news feed.
• 35:49
  SPEAKER_01: So I have to say props to Microsoft in terms of just the the way they approach the whole announcement the way it was announced uh the energy behind it as I say it's it's just it's just crazy that you know the the time from it being just public preview to GA is just phenomenal.
• 36:05
  SPEAKER_00: I've never seen anything come come go through that sort of life cycle as fast before it's uh astounding you have to say just nothing like project operations no you're we get you're gonna be you got you're gonna be waiting two years for things uh I'm sorry but I may but what one more thing maybe at the end to ask maybe you should kindly explain the word harness because this is not an AI word harness just harness harness it is it is basically uh it is basically just a connection of different things that that make it up together to like one thing so um if you think of agents a harness is basically like all of the tools triggers um skills all of that all of that combined is the harness with the agent together um there you can also apply this this word to for example um qa so a test case a test suite or um the scripts you run on your automation testing this all is a harness so I really don't get intimidated but as like a p it's like a picnic basket you know an inside picnic basket I've got my my agent my tools my knowledge sources maybe a nice little nice little dessert for afterwards you know so it's all just yes so it's a nice presented and lovely so yeah so so please please don't get intimidated because everyone knows harness and you need to understand what a little and you have no idea so harness you have to get the face right when you say it harness harness yeah it is it basically in old terms we just call it a solution I think yeah exactly exactly so yeah it's it's nothing new it's it's it's been there for a long time and since I think the 80s or something so it was just not used and someone just thought oh yeah this is like a non-used word we we didn't use a lot in the in the past decades so let's call it now and and make it very important in the age of AI.
• 38:00
  SPEAKER_00: Retro is the new vibe you know so I'm all for this let's let's bring let's bring back all the great stuff from the from the 80s all the great stuff from the 90s yeah yeah so because retro is the vibe also being offline is the new vibe and therefore um thank you for listening but no please go back and touch some grass and go enjoy enjoy the summer um you're still here till we resolve this thing here um so thank you yeah i'm gonna i'm I'm gonna go back to bed and see if that helps go back to bed how this goes I will head down for breakfast um and therefore thank you so much um new episode of unresolved dependencies every second Monday and uh yeah last words to Joe uh keep on vibing enjoy it and uh I hope you uh really really enjoy your newness goodbye