Optimize for Agents - 009 Agentic Thinking

本站添加: 2026-05-09

[00:00:01]thinking.

[00:00:13]Welcome back to Agentic Thinking with Matias and Mike. Hello everyone and welcome back to the show Matias. Happy to have you back for another Friday session with us today. We're excited to jump in. Let's talk about our main topic first and then we'll get into our details. Matias, what's on the main talk docket today? What are we learning about today?

[00:00:31]>> Um, right, we're continuing um with with our um exploration of using agentic workflows with a with a PowerBI project.

[00:00:40]Um, last week, if you if you followed um we got some mixed results, let's say, you know, we um we weren't quite happy with what the model produced there. So this week it's all about figuring out how can we um prepare the project in uh in in different ways so that uh the exact same prompt will actually give us um substantially better outcomes moving forward. Um so lots of learnings here.

[00:01:07]Um I it's going to be unplugged, right?

[00:01:11]So I don't really know what's going to happen. Um but >> real demos as Marco Russ would say.

[00:01:16]We're doing real demos >> indeed. But I I I have thought about this a fair bit. So, I've got a plan in my in my mind about it. Um, >> yeah. What's um what's up with you? You know, any any um uh any news items to share?

[00:01:32]>> Yeah, I I do a lot of thinking. Um the algorithms all know what I like. And what I like is everything about business, marketing, and AI. So, everything on my feeds, all of the platforms have learned this is what I like at this point. And it's all of them. It doesn't matter YouTube X uh you know LinkedIn like all of them are sending me feeds about this kind of information. So I'm I'm really engaging with this kind of content. Well one I'd like to share with the community here is just this uh it's a concept of in the age of AI which is I think what we're entering in now. This is this is a new era to build and design. There's this idea or concept around founder mode. And so in this short I'll share here in the chat window in case you want to go look at it yourself on your own time. Um let me copy the link here. I found this to be really um engaging from a concept standpoint.

[00:02:29]Um and it was basically thinking through if you're a CEO and you are using agents and you have a large company, it's going to be more difficult for you to adopt agents and move forward because there's, you know, we got to be safe. We got to move strategically. We don't want to upset our existing customers. There's this a bit of resistance to like doing things new and creatively. The age of AI is incentivizing, I think is the word I want to use, incentivizing the founder model. Move fast, build with agents, create quickly, um go and just build faster, right? And so this gentleman uh talking on this podcast, I think he works for Airbnb, he's reimagining his whole the whole company or or portions of it. I think that's where he's working from. I don't quote me on that. But the reason I like this this short a lot is because this is how I'm thinking. I'm literally rethinking my entire business. Like anywhere I have friction about something I am not wanting to do, answer emails, write statement of works, um you know, laborious time logging of tasks and things, I'm looking at this and going in the same way I had when I saw Power Query for the first time. I saw Power Query and I thought, "Oh, this is amazing. I don't want to go import from Excel anymore. I want to use Power Query for everything. Like even if it's like I want to make this tiny little table inside Excel. I don't even want to do that. I just want to go right to Power Query and then use this really nice experience and automate everything. And I'm having the same mental thinking processing that's happening now on top of agents and everything I do all my entire business.

[00:04:04]And so I really like this AI founder mode. I'm fully embracing it. I'm on board with it. Um and that's where the short comes from. So Matias, what's your reaction to this? the short >> um totally agree you know definitely the way I think about stuff as well what's interesting to me I um actually see insane analogies here between the business world and the engineering world um with respect to challenges we have now with uh you know those those new AI capabilities um when you've got a brownfield engineering project um and you are trying to AI enable it, you're going to find that it's actually so much harder than um doing the same thing with a green field project where you start off uh with the project um you know using AI practices um and and agentic workflows from the beginning right and it's exactly the same thing I believe when it comes to companies right Um uh anyone who starts a new business, a new company right now, uh in an AI native way, um >> huge advantage, >> a huge advantage over trying to convert um >> legacy.

[00:05:27]>> Legacy. Exactly. Isn't it funny? Yeah, this is exciting >> and and I think Matias, both of you and I resonate with this really well because we are both like not only are we founders because we have our own companies, but we're like AI founders as well. Like so we're heavily invested in like AI is changing how we do things >> and as we talk offline, >> a majority of what we spend our time thinking about and building through and creating with is heavily how can I get the AIS to do this? How can it's not just can I solve the problem it's how can I solve this problem with an automated AI thinking system where does the AI need to reason about things where doesn't it need to be >> where does the AI create things where does it where does it need to be in investing in like thought >> so anyways really cool topic great short I think I would I just would share that as well um I do want to be mindful of our time we don't have a ton of time today so Matias maybe we just jump right into demos >> sure absolutely because there would be so much more to say about >> I apologize. You had one more thing I thought was really really important. I It slipped it slipped my mind. You had one more announcement that I need to go pull up a link for.

[00:06:33]>> Yes. Two days old.

[00:06:35]>> Um >> so I'm a huge Cloud Code user, right? Um and so Enthropic announced um something which is quite unusual these days. Um they they've been giving their users higher usage limits. you know, generally and we've talked about what what GitHub copilot has been doing lately. Uh things are moving in the opposite direction. Uh two days ago, big announcement on the anthropic blog, higher usage limits due to their um uh presumably significant compute deal with SpaceX. So they they removed um peak hour limitations. they doubled um uh budgets within um five hour usage windows and um I can definitely say I'm you know this is not just uh a blog post uh with an announcement this is real right I've I've really um and in fact um I'm really struggling right now using up u my max um claw code subscription it's actually really It's a real challenge now um spending all those tokens. Um so there we go. That's exciting news and um it just shows how competitive the market is.

[00:07:54]>> Yes. And I would agree with you right there and I this is the kind of thing we this is uh this is what we've been saying since the beginning of the Agentic Thinking podcast is everything is about tokens. It's going to be it's going to be a token war at the end of the day for all of this stuff. So very excited to see that as well. Awesome.

[00:08:11]Very good. All right. That being said, to our desktop.

[00:08:15]>> Absolutely. Let's do that.

[00:08:16]>> All right. Let's jump over to the desktop here. Uh M Matias, pick us up where we were before. We have a PowerBI model and where we can do next.

[00:08:24]>> Absolutely. Okay, cool. So, um that's where we landed seven days ago. Um I still have my uh chat session open here.

[00:08:35]And um this particular prompt um not very well written um uh was basically saying um from the uh model improvement recommendations uh the agent had created for us previously please tackle two of them display folders and measure descriptions. uh create a new local branch uh commit changes um and make sure that the GitHub issue um which in this case is GitHub number three or number four um is reference. Um so it then um kept working in fact took quite a while um and ultimately um we got a nice report as you do. Um but when you and I reviewed what was done you know there were quite a few things that you know we weren't so happy >> very junior it felt very junior >> indeed so first of all um just a little recap uh something which is really amazing because um uh we've already committed all those changes obviously right so and um so looking through my git commit history at this point I wouldn't necessarily know what exactly the agents actually produce but um in the vs code chat that window I have um this alternative div view which for people need to understand this is different from our git diff right this is the the div that was produced as part of the session output and um it basically shows us how um 72 lines were added across those five files and as you can see um it does look exactly the same as a git diff in VS code you know where we have green and red color coding for additions and removals. Um, and so yeah, I want to actually hang on here just a moment. Also notice there's some different icons in the folders window as well. So I want to highlight this as well. So >> yes, >> so when you change a file versus when there's a new file versus when there's a agent edited file, >> there's like three separate icons now.

[00:10:40]So, I believe the square with a circle in it is indicating to you that that file has been changed by the agent, but it actually is not. It's not. Yep. It's pending a Yeah, perfect. Love that.

[00:10:54]>> Um, it's it's highly useful, but at the same time, it can be really confusing as well because >> Yep.

[00:11:02]>> if you switch from your explorer view to the source control view, >> you're going to see very similar things.

[00:11:10]um you know about modified files and their respective changes and uh over there the change would say was it an addition was it a deletion or was it a modification right sure so uh and then the um uh the similarities go even further you know when when you look at the diff view you're getting here so for people who are new in that space and you know who are not um uh heavily um used to using source control it can be quite confusing and sort of just um >> understand one scope is what git sees in terms of what's changed in your folder the other scope is what your chat uh considers a change um that one also allows you to explicitly keep or undo. Now let me call this one out specifically right because uh obviously >> as I said before all those changes that were done as part of the session were already committed right which is why looking in the source control bit over here we don't have any pending changes they're already in my git commit history in fact they're actually all in here um now um this keep and undo which we have here for each individual ual granular change and which we also have here for all global changes and in addition to that we also have here at the file level. So five you know three different um scopes at which I can apply keep or undo that is related merely to the the chat session right and so I can for instance I can now say keep this one and if I were hypothetically well let me actually do it if I click undo here >> Mhm. it it undoes it which then means obviously so this the undo relates to what the chat did but it it does mean that with respect to my source control I now have a change which obviously is the removal of that particular line that had been added previously right um so folks just need to be very very mindful um that we have two highly overlapping diff and uh sort change control systems here and um uh they both make sense if you really understand them but they you know you can get highly confused as well. Um >> sure makes sense.

[00:13:46]>> So obviously I don't want to undo this.

[00:13:48]So uh which means I now use the git undo to undo my undo my chat undo. Right. Um yeah there we go. Right. Okay. which also means down here uh you can now see I no longer have four files showing up as change rather than five as before.

[00:14:10]>> Okay, >> but let's get to the um to to the uh gist of what we actually wanted to achieve here. So >> sure >> uh our main critic well there were there were two main criticisms I would say right one is you know and and that's the key one we notice that whilst all those measures have had um descriptions added automatically by the agent um those descriptions were not very useful from a semantic point of view. those descriptions were mere rephrasing of what the underlying DAX function does.

[00:14:51]Uh those descriptions uh are not achieving a business focused uh description of what the measure you know from a semantic model point of view actually does. And um uh in that respect, you know, I I said um a week ago, this is the kind of stuff I would expect from a very junior modeler, you know, who was tasked to provide descriptions for measures because intuitively they may well do exactly that, right? uh read the DAX formula and and just describe what the formula does as opposed to actually understanding you know what does this measure do in the context of a dashboard or report for instance. Um so that was a criticism we had and um I wanted to explore different ways of how we can mitigate that uh and how we can um uh uh sort of get to to a better place. Um and uh so my idea was um because all those changes are basically sitting in this one git commit here which by the way was our second uh criticism but that's uh a lot more minor than the other one. All those changes sit in this particular um git commit and in fact they sit in a dedicated branch as well. Right? So remember um down here we created this branch. Uh in fact that was part of the um prompt uh for the session. So I'm going to um uh I'm going to uh create uh a new branch um with which doesn't have any of the those changes done. And I'm going going to try again. And I want to try a few different methods here. Um, okay.

[00:16:47]>> So, let me create a new branch here. Um, let's call this, um, I don't know, demo. Um, uh, model fix attempt two maybe.

[00:17:08]>> Yeah, like that.

[00:17:09]>> Okay.

[00:17:09]>> Mhm.

[00:17:10]>> Cool. So create branch right. So um and whilst we're here um I I I don't want to I don't want to lose anything. So I don't want to switch to this branch branch immediately. I don't want to lose anything in this particular window. I don't want to lose anything um uh you know for comparison purposes. So what I want to do instead um you know I've got this particular branch here. I want to um create a rock tree um which um again is a relatively new git concept uh and it's become quite popular when uh cloud code actually introduced native support for ro trees. Um so rock tree in short allows you to check out multiple branches of a repository in parallel um without explicitly having to create um separate um clones of of that repo um yourself. Uh it means that the uh uh the main cloned folder actually has a reference of all those additional work trees um that you've created and you can easily switch back and forth between them. Um so I've just said on this particular branch I just said um open in rock tree and as you can see it's doing two things. One it's creating a whole new uh VS code instance for me. Uh which means you know the uh other window I can keep open in the background. I can have that chat still there. So I can always easily go back. Um but then also we can see that down here very nicely. Um this is this is my new folder. This is the old folder. Right? So um >> love it.

[00:18:58]>> We can also look at the actual um uh file system folder it's created for that. Um and we can see down here right um so by convention um so the original uh clone I created was basically called PBI modeling MCP copiloted because that's the name of the repository by convention. It then creates a new subfolder where it appends rock trees and then it appends the um uh folder the branch name I've selected um on top of that. Um love it. So there we go. So that's how work trees work. Definitely a concept you should um be familiar with because it's >> I'll pull a link from something about work trees and just kind of get a 101 of work trees because I think they are very powerful and especially when we're building with agents, this is a a skill that's worth learning.

[00:19:47]>> Absolutely. Okay. So which means uh with this one because again I I created that branch uh one commit ahead we're back to where we were before. Right? So um we've got the semantic model checked out here.

[00:20:03]We've got this original research um item that um actually creates an inventory of everything that um could be modified.

[00:20:13]Um, and so I wanna let me just do a little bit of an um uh so let's call them prompt um improvement um methods. So I want to show three maybe four different ways how we can um change that. So the first one would be um to use a richer prompt. That's the cheapest and most intuitive one. uh well not cheap, it's actually the most expensive one, but um it's the the easiest in terms of getting it done quickly, right? Um >> so the second one would be to use um agent instructions um at the >> Can you zoom in a little bit on that? Um yes, >> control plus on that document there.

[00:21:07]Yeah, perfect.

[00:21:08]>> Instructions. Um so that would be for the repository.

[00:21:12]>> Um the third one would be to use um skills and then the fifth one would be to use a custom agent. Those are all valid ways of achieving what we need to achieve here. And um uh it's not the case at all that you know one is better than another. Um anyone who actively works with agentic systems needs to be aware of all of them. Um >> okay. So, um let's just go back here and um the goal is to be able uh to use the exact same prompt um but to achieve better outcomes. So, I'm going to paste the prompt in here and then I'm going to add something to the prompt. Right? So basically all the things that uh we um want to be done uh separately uh all the things that um we want to be done uh differently I'm just going to add to the prompt here. So what should I say um when uh composing um uh um uh measure um uh descriptions um do not and capitalization generally is a very good way of highlighting or emphasizing uh instructions you're giving to an agent. do not um merely uh describe um what the measure uh formula um formula does uh technically. Uh goodness, I'm really bad at typing today. Um um instead um uh think about this uh from a uh business user perspective.

[00:23:06]um and um describe in uh instead um what a uh Excel um user for instance um can expect as um output from that measure. Right? So that would be one um uh additional sentence or paragraph um to provide a richer prompt here. Um the other thing Oh, sorry. Uh I um um I sent it off already, but I think that's fine.

[00:23:46]Um we we don't um um have to worry about um everything else. We just want to see that um that it um uh well how it behaves now and and you know whether um it uh it produces um uh you know an output that's much closer to what we're looking for. Um in the meantime while yeah sorry go ahead say one of the things that while you're in that text box this is um you accidentally not accidentally but you hit enter and it sent the message. So, it's funny when you're in that little text box window, it kind of feels like the DAX um editor or when you're in like making DAX formulas, right? You need like to hit I think it's like shift enter or hold control enter uh in order to get another a new line to kind of make new lines in there. As soon as you hit enter, it just goes.

[00:24:37]>> No, no, no. Actually the real the thing that really um happened here is that I constantly switch between different coding agents and it happens to be that um in claude you on a on a Windows machine you need to press control enter if you want to enter a new line. Yes, correct. And in most other systems you need to use shift enter for a new line.

[00:25:01]And I just accidentally um because I've got the muscle memory from from so many CL sessions, I just press control enter which unfortunately doesn't mean anything code. That was my mistake. It actually happens a lot to me unfortunately. But there we go.

[00:25:19]>> Inconsistent tooling or inconsistent harnesses I guess is what that is is causing the problem with. Yeah, it it gets even worse um because um when you use claude um in the browser um it actually accepts shift enter rather than control enter. So you have to be extremely mindful of it. Um okay. So um this is actually something I was hoping for. Remember um uh permissions is always a re a key thing for you to configure correctly. Um uh one you don't want to um overdo it and and give too many permissions because you know you can easily get into trouble with your agent going um um too far. And uh on the other hand, if you're giving too few permissions, this is happening, you know, you're constantly going to be >> prompted and um the agent is not autonomous and you get disrupted, right?

[00:26:17]So, finding a right balance is actually key here. It turns out um that this is not something um you know um I'm struggling with. It turns out that um the uh VS code guys, you know, who who built this harness um have provided us a really good way um around this. They've got this autopilot mode here, which means >> um you don't have to go for default approvals, which uh you know, which prompt you a lot. You don't have to go for bypass approvals, which never prompt, but basically allow everything.

[00:26:54]Autopilot means um every time um the model would prompt you, it's actually using um another uh model uh to determine how severe that particular tool action is. And that model then makes a decision on your behalf, right?

[00:27:14]So that's autopilot.

[00:27:16]>> Yeah. Reading something is less impactful than like writing uh or you know creating something new or even you know delete. it would be like on the highest end of that threshold like so deleting things is probably gonna like oh this is a severe action we should really confirm with the user a delete is is required here >> exactly so when I use claude um I'm generally very good at um predefining the permissions in in my cloud settings files and normally it it's it's very good uh at then not um prompting me um when I use co-pilot and vs code chat uh that's not so straightforward unfortunately the declarative permission system is not as great as the one in Claude, but um autopilot is actually really good. So there we go. So I'm going to switch that. Um it's warning you. Um there we go. Uh but I still for this particular prompt I still need to explicitly confirm it. So what's this doing? It's looking at um Oh, there we go. Right. So it's it's using the GitHub CLI. So that you know that's GitHub CLI that allows you to interact with the GitHub API in a very um high level way and obviously it wants to list um the issues I've got on this particular repo because I referenced the issue as part of the prompt. Right? So of course I want to allow that. All right. So whilst it's doing that, I'm going to go back to my original window because obviously I promised you four different ways of tackling the same thing. And so it can go on, you know, with my modified prompt now whilst we look at um alternative ways of um achieving that. So I'm doing the same thing here. I'm going back. I'm creating um a new branch.

[00:29:00]Um I'm going to call it demo. Um uh the agent instructions.

[00:29:10]There we go.

[00:29:12]Uh create branch. And then I can go here and I can do the same thing. Open in rock tree. Got got yet another window.

[00:29:23]Um this time I'm going Well, first of all, I'm making it bigger so you can see.

[00:29:30]>> Thank you. Appreciate it.

[00:29:32]This time I'm going to switch into autopilot immediately.

[00:29:36]>> Love it.

[00:29:37]>> Unfortunately, this is not something which is persisted across uh VS code sessions. So, you always have to do that explicitly unless I haven't discovered it yet. So, if if someone knows something I don't, then let us know the chat if you if you know what this is and how to keep it on all the time.

[00:29:51]>> This is collective learning here, right?

[00:29:53]Yes. Okay. Right. So, I've switched that. By the way, you know, I'm using uh set 46 here. uh you know your choice whatever model you want to use but uh that one um is very capable and at this point you still get it for 1x um premium requests unlike OPUS 47 which is 15x as you can see so that uses 15 times more um of your allowance. Okay so agent instructions what does that mean? Um um we can at the repository level um define um uh standardized documents um that um will be consumed by any agent session automatically. Um and um unfortunately there are various uh conventions um what that document is called and where you place it depending on which harness you're using. Uh so in claude code for instance would be claude.md. Um thankfully outside of enthropics claude world um many harnesses have converged on a um convention that's called agents.mmd. Um, and so I'm going to create an agents.m MD file here.

[00:31:16]Um, and uh, copiloted will read that automatically. Um, and in fact, agents.mmd is a convention that also works um, uh, in a scoped fashion. So you can create agents.md at the root of your repository, which means um, that one will always be read in any new session. You can create nested agents.mmd files in subfolders and those ones will only be read and only be applied when the agent actually does something in that particular subfolder.

[00:31:48]So very powerful concept. Um >> okay. And so this one basically means you're putting um let's call it static um additional context um into all sessions that run within that particular repo, right? Um so um and so this is something which will automatically and completely um hidden from you uh be hidden to any prompt you're giving. So for instance I can say um whenever um um you annotate um uh semantic um u models with um descriptions.

[00:32:42]Ensure you do not um merely uh describe the underlying uh DAX functions.

[00:32:55]Um goodness, sorry about my >> right there. Go back to right there.

[00:33:01]Instead, focus on explain the business logic and the purpose of the measure.

[00:33:05]>> Exactly. So yeah, so this >> tab through that one.

[00:33:08]>> Absolutely. So this is AI um actually right so if you're looking at the evolution of AI this kind of autocomplete is the very first you know >> um >> iteration that we had before chats and agents uh came about right but but nowadays they are so much faster because they can use very low latency um mini and nano models um uh for you to uh basically provide those kind of subsecond um completions and they are just extremely good nowadays right two three years ago >> you know could have been gone one or two one of two ways nowadays extremely good at inferring your intent >> and usually they suggest you stuff that's way better than what you would have written so it and in this case it actually turns out right I've given it this and it completely inferred what I wanted and gave me a much better version of what I probably would have typed, you know, with various typers. Now, so there we go. So, I'm happy with that. Um, this will help users understand the intent behind the calculation, how it can be used. There we go.

[00:34:25]>> I want to pause while you're going through that one right here as well. One thing I want to note here as well, a lot of what I'm doing right now pattern-wise is I'm doing things like this where I'm building some semblance of, you know, an agent or custom agent or I'm building specific things here that are going to help me generate better output from the agents. I'm also to this end, this is just a file just like markdown and everything else. I'll take a first swing at this >> and then I would go back to an agent and say opus or something more reasonable like with more reasoning and say look the intent here is to make it not just write the DAX formulas to give real business user value look at this look at this prompt what should I what am I missing >> add something to it so I'm actually using the agents to help me generate better prompts to be used in the agents so that again it knows what it wants to from text to get a better output. So I just describe more around the outputs I desire and let the agents figure out the wording it needs to be able to to run better. So even this I will cycle through a couple times letting agents modify this code here just to get a better output.

[00:35:33]>> Right? And to add to that, you know, my my per personal um experience, if you want to say, is uh those um repository scoped agent instruction files, they should never be static. They they need to live with your project. And in fact, I frequently, you know, in in all live projects I'm working on, I frequently um have uh like a weekly, if not more often, review session with my agent where I'm asking the agent, have a look at agents.mmd. Um understand that this is a grounding file that will be um included into every single agent session in that project. Um >> is that still accurate? Does it still reflect um where the project has evolved to? Does it still um uh point out you know the the the the key dos and don'ts?

[00:36:32]Um does it miss anything? You know I I do that very frequently. Um y and um particularly when you have fast evolving projects you know that go uh you know that do sort of substantial uh evolution you know um quickly it's very important you know this is not something you write once uh and also the other thing is um uh claude and other coding agents have an init function right uh where uh when you uh go into a new project with them you can you can do slashinit and then it will actually autogenerate um those kinds of files for you and you may well think well this is um a job done now but it's highly highly um recommended and highly important for you to actively review what's in agents.mmd or clot.md because um it turns out the quality of those files has a huge impact on the quality of all your agent sessions you're running later on.

[00:37:36]Okay. So, let's do the same thing here where basically I'm copying that exact same prompt from before and um with no further additions, with no further modifications.

[00:37:49]>> Um I'm giving that here. I'm going to let this run. And I know we're probably um >> we're at time right now.

[00:37:58]>> At time. So, let's let's just do a quick review of where our first attempt got to. Um and then maybe we can pick up things uh here next time around if that's okay. So this is the one. Um >> so prompt uh this >> I believe this is richer prompt. I think >> this Oh, you know what? Um obviously um my prompt included the instruction to say please create a new branch for this, right? So um that that's why I got confused. So Right. So this but we can see it here.

[00:38:37]Right. So um attempt two is actually showing as the um as the uh um root node in my explorer. So there we go. So that that that's that one. Um if we go up it's the longer prompt. And let's just look very quickly. What has it done?

[00:38:56]Customer.tindle.

[00:38:58]Oo.

[00:39:02]We're not getting any comments.

[00:39:05]>> Oh.

[00:39:07]Oh dear. Right. So that's a fail.

[00:39:11]>> That's a fail.

[00:39:12]>> Welcome to the real world of real demos of real coding. Like this is why we do this because we we we modified something but we weren't clear enough. It didn't understand what to do.

[00:39:22]>> Okay. So let me have a quick look. Issue three is about descriptions. Issue four is about display folders. Okay. Um uh now let me make all the edits. So for some reason it's decided not to make any For some reason it's decided. Oh here business user measure descriptions added. Okay. It was very selective.

[00:39:48]>> Uh sales amount margin cost. Um >> so go to the sales TMDL.

[00:39:54]You're picking the store one right now.

[00:39:57]>> So the the sales one maybe has it in there.

[00:40:00]there. Oh, there it is.

[00:40:01]>> There we Oh, look at that.

[00:40:03]>> So, it was very >> It's done collectively. It's done it only for three.

[00:40:08]>> Um and interestingly enough, um Oh, right. Yes. So, we didn't see it because um I didn't scroll down far enough. Um so this one says so this is sales amount year to date the accumulative net sales revenue from January uh 1st of the current calendar year through the last date visible in the report compare this against to see whether okay that's actually not too bad >> I like the description >> quite verbose but um I think that's fit for purpose here right because we would not assume a technical user for that right that's not the audience um So margin percent um can we find margin percent? Uh >> control F.

[00:41:04]>> Uh it would be called measure. Oopsie.

[00:41:08]Measure.

[00:41:10]>> If you just go to the if you're on the sales file and not do the search icon there, just control F on the file. It'll just pop up the search window in the file.

[00:41:20]>> Well, here. There we go. Margin >> measure margin percent. Here we go. Ah, here. How many cents of every dollar?

[00:41:27]Oh. Oh, I love that. How many cents of every dollar in sales revenue remain after paying for the product sold? A result of 30% means that for every I like this hundred pounds of $100 of revenue, 30 is gross profit. Compare this across product. Love it. So, okay.

[00:41:44]Very good. So, >> that was a good description.

[00:41:46]>> Fantastic.

[00:41:47]>> You even gave an example. I love that.

[00:41:49]>> So, okay. So, mixed, right? Um, >> it better it definitely did what we wanted, but for some reason it's decided not to do it um everywhere presumably because it was missing context and it wasn't able to produce something. Um, but we don't know, right? Um, I appreciate we're at time, but this is exciting and um I can't wait to see what the other attempt um has given us, but we'll have to wait a week to get >> So, so regardless though, this is a like I wanted to kind of end on this final thought here. This is a real process like what you're seeing here of like multiple attempts, refining a prompt, doing it again, doing it again, doing it again. This is where the real work happens. So we we said you know AI will take your jobs. No, we've just made a brand new job which is prompt engineering where we're now rethinking like what do we really want? Is the output really what we needed? Do we need to change it again? Do we let it live as a highly customized prompt or we want to reuse it at the agent level? So these are the considerations that we're now redesigning. We're rethinking this whole process to get better output. Now, why the the question may be like, why would you do all this work here to figure this stuff out? What you I think you'll see here in in a couple sessions time here is once we get to something that works, now we have a solution. Now, we have a more um catered prompt or engineered prompt to get us a regular output that we're more happy with. And kind of to our point earlier in the session, which was you're going to build skills, you're going to build agents and things like this and they will need to evolve with your process, with your skills, with the repo that they're a part of. And it will be constantly you're going to need to be like maturing these things as you go through this process. And so this is directly what I've been finding as I'm doing. I'm doing a lot more of like pause, hone in, build a prompt, get what I want as output, and now I'm thinking about building process to make sure that that prompt regularly gets reviewed and improved and over time gets better and better. And and so this is I think really the core principle here is we're showing you how you make good prompt engineering elements and then we'll have to refine them over time as the project evolves and gets better. Anyways, >> let me if I may, let me use another analogy here. You know, last time I said the the outputs which we weren't so happy with were, you know, sort of what we what you might be getting from a very junior model, right? And so the process we're going through right now is effectively the same process you would go through with your junior employee.

[00:44:31]>> Yeah.

[00:44:32]>> By explaining more, by providing more context, by training them. Um so that ultimately you know after uh having invested a bit of time they become senior and and they are able to produce the right kinds of outcomes first time whenever you give them a new task and that's very similar to what we're doing with our agents here.

[00:44:55]>> Love this. Very very good. All right everyone thank you very much for joining us for another working session around agentic thinking. We're going to keep going down this path. There's so much to unpack here. We're just scratching the surface. We hope we enjoyed our demo today. Matias, thanks for screen sharing today and just diving in and building some amazingly cool things. We're going to keep going down this path. Thank you all so much and we'll see you next time.

[00:45:16]>> Thanks everyone.

[00:45:27]Hey, thinking