How Business Analysts Can Use AI Agents | Practical Guide & Live Demo

0:06[music] So let's uh get on to the session and then we'll at the end we'll I'll ask the same question and let's uh answer from there. Right. So I mean I'm just I've just put two ways to use AI. It can be any other ways but you know for this particular uh discussion I've just put it this way. um you know mode A that is your version A you want to write faster some people want to write items faster like uh creating a conference page or creating a Jira uh ticket or whatever is it right um and some people want to think better and write faster I think I got one of

0:50the answer also from uh I didn't see the name but I got the answer so that is where you know how agents are going to help you to think better and help you to write faster. Right? So let's dwell into the session to see what we're going to achieve by end of this session. Um so

1:10this is what is going to be our agenda. Okay. Um how [snorts] agents can a agents can support your your BAS life cycle. Then we will look at these six steps like how you frame it, what you do for research, analyze, draft and critic and finalize the workflows. Please remember this flow through your whole process, right? We'll be looking at it in multiple uh uh steps but this is going to be your your whole point of discussion and always I say I think in my last session always keep human in the loop because um a can divert its own thought process and take it to a different

1:48level. So that's why I always keep human in the loop. We'll also look at some tools and techniques and then you will find out whether a tool is important or technique is important. You'll come back in in end of this uh session and then we'll do a live example and we'll take an agent. I'll share the link for the agents also. Uh you can try it out on your own and uh let me know how this is working out. uh offline of course uh will not take the whole time to explore it but I'll show you how to try it. You can you know download lot of agents like

2:23this or skills like this and you can try it on your own. So I'll not bore you with lot of theory um rather I'll jump into practical so that it's it's more interesting. So just have a perception of how u agents work in any any agent work in a BA life cycle right they'll help you to u prepare your questions um any conflict uh detection between any two source or data source anything and uh you know getting your artifact your first time artifact like let's say I want to write uh BRD or FRD or PR faq you want to do a first thing it would really help you and uh understanding

3:08your whole your whole coverage of your whole project by giving your history of what your project is all about. It's going to give you the full information about it. And the most interesting part of this is the critic. When I say critic, it is um now you're talking to two people, right? um person one is telling person person two and the person three is going to give a critic on the discussion that's happening between person one and two that's a critic right so there'll be always another guy who's coming and saying hey you know what what you thought is different you got to think little different right that's what a

3:44critic is all about and of course you got to shape it to your stakeholders you know who you are writing this document for BRD for different person for different person FAQ for customers or what not Right. Um yeah, the what agents don't do, let's be very clear. They're not going to do any of these. I'm not going to go read each one of them. And

4:06this is well known, right? So always I say this is a loop. You know when I said remember this is going to be your flow. Okay. So when you see the word human here, it's you who's going to do this. And when you see the agent here, that's the guy the agent is going

4:23to do that wonderful job for you. Okay? Of course. whatever we do also is wonderful. I'll not discount that. Agent has a flavor to it. That's how I'm saying agent does wonderful work. Um, so when you start it, you you have a problem, right? Let's start. I'm doing a BD or I'm writing a a scope document or something, right? You set your problem and you say this can be achieved, this cannot be achieved. You give a whole frame to the whole thought. That's what your frame is all about. Now, after your frame is ready, your problem statement is ready, you're going to ask an agent, hey, you know what? This is my this is

4:59what I'm going to write a document for maybe uh PR FAQ. Go research it in the market. Do all your research. Get me all the relevant documents and everything and then you know um give me an output. Research also includes your history data, your historical data. Let's say you have 10 BRDs, 20 FRDS or whatn not. Any documentations that you have, you can also add that to the pool and say, you know, do the research for me with all the context, right? And then here is where you guys combine each other, right? Um the research will give you certain documents and you will say some are

5:42really irrelevant to me. You need to have your own perception, your own human in the loop to verify it. Simple thing till 1 2 3. Let's think like this. You're opening a charge GPT and you're saying um write me a document for this. It would do a research. Correct? And once it research and give you the first iteration, you'll analyze it saying hey this is where I am, right? And once it says uh this is where you are and you'll refine it with your point saying that no I don't like this section remove this section for me or modify this section to something else or not that's what you

6:18will say and then the agent prepares the first draft for you a normal flow in our you know chat GP life cycle I'll call it the other important thing that really you need to do is a critic a second agent uh I'm going to show you how the critic works uh we'll not use an agent kind of an agent but we will I'll show you how an agent session is initiated so that you know um it's like till here it is agent one till step four is agent one you and him discussed a lot you both created a document now what would happen take the document you're going to another guy in

6:57another room and just giving it to him and say read it he's going to say dude these points are wrong you need to modify this this is not as per standards what not he's going to give you all the critics that's what this guy is going to do and very important is what I see we call it um LLM judge we call it LLM jury you would have heard about all these uh names and there's something called LLM council so it's like you're asking another LLM to go do a critical review for you for the document so what we all do is we do from 1 to four we do it

7:29beautifully no issues at all right the fifth one is where we miss it. We just take the document, put it in an email, send it. Oh, my work is done. My boss has asked for me and I'm done with my documentation, my work. And then you're done with it. But I would say this step is very important and I'm going to show

7:46you how to do it. It's very simple. Okay? Not not at all complicated. And the last thing is you're the one who's going to finalize the whole loop to say that yes, I agree with what you're saying. Now you want to take the final critics's um reviews put it back in the first agent and say you know what this is what I got a critic can you modify the document as per step 1 2 3 and then get a finalized verification. So this is a whole loop you do actually we all do this in a regular day but we tend to forget the process that we do for uh you

8:22know start to end and how an agent can help you to do this work right now why always I say keeping humans in the loop right which is an important factor according to me is agent does everything okay even you don't need to keep your gates you don't need to keep your guardrails and

8:42anything. Agent does everything for you. Yes. Excellent. And it will it will get your materials. It'll get your source of things. But what you need to really do is you need to shape your business context correctly. For instance, you might say build a BRD for automating something in Jira. That's your B. I'm just cooking it up. I may not be right also. But yeah, so the business context might be less. It might have understood something different. So please please reiterate that and this is something the agent will not have right how a stakeholder is going to look you know your stakeholders very well you know your counterparts very

9:24well right how they are going to look at this document agent will not know so you are the one who's going to have your gate and watch out for those pitfalls right and take a look at this SLS cost it's not going to have it you're going to have it you know your wallet you know your data volume you just want to feed in the input and ask look my volumes are let's say 100 transactions and for 100 transaction I need an SLA of whatever um split seconds response and all those are in you know your information which the agent again will not have and then your

9:59final sign off you want to give so just understand what the agent will really do what you need to do so once you understand your role agent's life is it's a bliss It's simple but yes when I say agent agent uh what is that agent we'll just see in a bit what is that agent know when we go in our demo um this is what I was saying tools and techniques right let's look at each one by one um when you're asking your problem statement give your real artifacts your old brs your old uh uh PR faq if you're building a PR faq or whatever whatever

10:39the document you're building it right because it understands the context very well it understands your language it understands a business domain very well if you don't have it you don't have it nothing can be done or even a Jira story would really help so that is very important for it to understand a context don't think the agent will understand on its own and you need to keep reiterating it no always it relies on your context it might be trained on 20 billion parameters 100 billion parameters what not but don't rely on that it needs to understand this the center of your problem statement and and then is actually splitting your

11:17whole uh context. know for instance I told you uh you know sample here so let's say you are in step one you came to step three and then you and him are doing a analysis on let's say point number four if point number four is not good don't ask you to rewrite the whole document it's going to waste time it's going to waste your tokens it's going to waste your limits or kotas or whatever and you're going just going to waste time in refixing things which are already fixed somewhere else so just Say look at point number four let's sit and uh solve that problem that relates to my

11:52NFS non-functional requirements uh help me solve those B point iterate iterate it step by step don't go by fix the whole document I don't like the English in the whole document right don't do that just go uh step by step uh block by block that'll be faster and easier right and uh I gave you another thing we spoke about critic um in the critic there is a small flavor which I'll show you in the demo is give a persona to a critic when I say persona now here if he says he's given a solution architect you are an architect who who has to build this where does it

12:32break so think of scenarios for your use case it can be you are the CTO of or CFO of this organization um go look at the excel sheet that I gave you which how can you find loops I'm just cooking it Okay, it might be something else also but whatever you understood the flavor.

12:51So give it a persona. So what happens? It'll start thinking in that terms. Now you just give a document to it and say hey you know what um critic about this or give me uh information about it's not going to do always started it as a architect how would you look at this particular document give your context to it what you want it to think as because as I said it's it's trained on multiple billion parameters but unless and until you tell its persona it will not understand what it has to really do with that action right and I said work one section at at a time. Don't worry about

13:28uh you know breaking the whole document. Go step by step and try to fix it at one place. So let's quickly jump to a demo because the reason why I shifted this time to a demo was last time the demo is taking really long because I want to spend more time on a demo. That's the reason you know I reduce the uh blur part the the story

13:50part of it. [snorts] Now, this is the question that you would type. Anybody would type um customers are abandoning checkout when payment fails and you're going to ask a PR FAQ for it. That's what we're going to do in today's demo, right? But if you look at it, there are a lot of gray areas in that question, right? Who is that someone and who wants this charge, right? When when someone is doing the transaction, who's going to take care of the second reiteration of that? Is it the company? Let's say for instance razor pay or whoever failed the gateway who's going to do that particular uh uh

14:26charge and when a retry happen is is it going to be like the customer is going to see oh I'm retrying this particular session or what is really going to happen you're not given that clarity at all you just said customers are abandoning things in the cart because the payment failed maybe a pay would have failed but you're not giving the clarity to the agent who write it in the right way so what um you end up with in this use case, right? Um you get a single pager from the agent and saying that uh this is what really happening and these are the um some list

15:03of FAQs you need to do when something fails and end with it. Right? And you need to see how the agent is correcting things for um when you have the critic what is really how is it modifying when a critic comes into play. Right? So these are the things that you need to keep in mind when you start writing a document. That's what I would say. Now let's look at uh the demo on book first and then I'll shift my screen to another screen and then I'll uh show you the agents first and I'll show you how the agents are read and then you know we'll go frame by each each uh

15:43moves one by one and then we'll see how this works. So I'll take a quick pause here. Any questions you guys think um you want to add to it?

16:00If nothing then I convey it well. So I I praise myself for that. Just one question just wanted to understand uh I think you have said it before but just a thing uh when you said that suppose uh while giving a prompt when you don't have a draft uh normally it is better if you have a draft so you can put it in the lm to get a to get a better iteration or a better uh what do you say reiterate it better but suppose if we don't have a draft then in that case we should go step by step right is that a correct understanding mir

16:43that's right so you have a draft means when I say uh something previously already built in your previous documents that you have that's what I meant by if if you're meaning by draft yes so if you have some previous artifacts um whatever PDFs br SRS yeah whatever whatever the documents you have always start with that if you don't have it Like you saw the previous example. Uh let me just scroll my screen up. Customers are abandoning with check

17:11out whatever you're starting like this. But even when you start like this, start with the right narrative to it. Hey, I'm we are facing a problem in this particular gateway system and uh we see that customers are abandoning checkouts when payment fails. I would really like to run a PR FAQ for this and u what all things I can cover. So something like that. You write a story and when the story comes out right the document comes out there might be some sections where you know very well how your stakeholders would look. Now in this case your

17:43stakeholder would be a customer right. So how would your customer look what would we look for? So those are things again you need to look at your critic angle you need to look at your um customer angle and you need to come out you know the best of your product. That's what I'm trying to say.

17:58Okay. Okay. Yeah. Cool. I'll just stop the share and I'll share my other screen. Um, let me know if the screen is visible again. I suggest if not it should be visible now.

18:25Okay. So um what I'm going to do in today's session is very simple. Uh we're going to use cloud code uh sorry cloud desktop or you can use normal cloud browser also. I'm going to show you some um agents that are already available for u any business analyst which are already pre-built. You just have to embed them into the uh cloud desktop. I can even go a little more complicated putting in that last session I showed about anti-gravity. You can use any such tools that you like but you know I was just thinking let's uh use any agent that's already built and do it. So I'm going to show put skills

19:07I'm going to put plugins and I'm going to put agents into this and I'm going to ask the agent to do that magic thing for me. Right. So before that I would like to show you the agents and I will share the links to you guys. Um so the first thing that I'm using is this one business analysis agent skills. Don't worry I'm going to send the links uh in the chat. So you guys can take a uh it's easy to take a note of it. Don't need to write or you don't need to worry about it. Just uh there's some time and hear

19:38me out. Um so what this particular agent skill is don't worry about agent skill plugin and all those things. End of the day an agent is again an empty file. uh as we call it markdown file and that's all an agent is all about so don't panic about oh it's a skill this is an agent just don't worry about it you can just search internet for prq agents prq skills for cloud code prq skills for chat gpt you can just import them and use them I'm just giving an example so you can try the rest on your own so if you see here um this particular uh

20:18faq is Amazon terminology ology, right? Sorry, PR FAQ is an Amazon terminology, right? Um I think it's a global terminology. We also use it in our organization. I don't think it's an Amazon terminology. I did a search for that and said that it's Amazon related. I just I'm just taking up we are a pure FAQs press release uh FAQ

20:39frequently asked questions. Yeah. Yeah. I think it's it's used in most of the organization whether it's hail from Amazon. I'm not I'm not sure. I'm not maybe started in Amazon and then we don't maybe maybe Joe. I'm I'm I'm not a [laughter] very uh seasoned guy there but yes I I'll I'll learn that from you if that's the case. Now I was just asking because you know when I used to what I was like okay why is only this why I call it something else. So that's the reason it got me to okay no because why I took that as an example is because we recently did a

21:11BRFAQ in my organization something that struck my brain where my business analysts were working on it. So that's the reason I just took that as a use case but this is a good learning for me and everyone. Thanks J. Thank you. Thank you. Yeah great. So yeah, so this particular skill helps you to build FRDs, BDS, uh um BDS, uh and all the test cases and everything gap analysis. It's written here. It's clearly written here. You these are the skills that it gives. Uh it is a skill MD and they are the references. So what it would really do is it look at all these documents and then prepare a skill

21:45and then this agent comes into action. So some people say an agent is a combination of skills. Some people say an agent is just an empty file with an MCP connected to it. It's it's it's the way that you want to build an agent for. No, there is nothing right. There's nothing wrong. How you want to embed it into your uh ID. When I say ID, it's your uh like cloud desktop or anti-gravity. It's it's the way that you

22:12want to represent it in any session. Okay. And there supports how you can use it with codeex cloud code um everything right the other thing that I'm going to take here is uh business assembly again that's from AWS samples they're a lot like this right you can just uh get if you go to list of agents you'll see lot of architecture documenter I mean non nonBA related also is available in this so anybody can just pick up a QA engineer can be scrum master can be done like there's a lot of uh agents available right off the bat. you can just take it up and you don't need to

22:49build anything from the scratch is what I'm trying to say right and another thing that uh for startup business analyst you have these kinds of uh skills uh where um you can look at um it has market size analysis uh startup financing models team compositions um to build any basic doc starting of documentations these skills would be absolutely helpful I'm going to share these links to you guys and also So you know what I did when I downloaded it was not straightforward. So I had to do some little bit of augmentations to bring it to the format. I'll share the zips also for your easy quick start. So you can

23:27see how my zip is built and then you can recreate it yourself. Right. Um I'll just keep this aside and uh maybe quickly I'll share the link so you guys can if I have it handy. I have it. Just take taking a moment to um Madi the the chat is disabled. Can you enable it for me?

24:02Um anyone from Techanas can help me enable the chat?

24:13I'm not able to send it but yeah if not I'll I'll have the technos team share it to you guys. Don't worry about it. So yeah the next thing is I'm going to show you the zips that I have created. Forget these uh these are some HTML files I created. So this is the business analyst plug-in. There's a PR FAQ critic agent because I'm doing that PR FAQ to do the critic. I added that small scale for it and generic business agent uh uh plugin that

24:45please. Hello. Yeah, please. Uh uh. Can I um Can I uh uh uh just I I I would like the the link for for uh download the the uh the is what you're saying? Yes. I I I would like the the link for download the the file. Yeah, sure. Once uh we get someone from technos to enable the chats, I can upload everything at one shot and you

25:34can just download it from the chat.

25:42Sorry guys, can you go on mute if you're not speaking? Thanks. Great. Then uh so these are the agents that you know um I built don't worry about it's called plugin or agent or tools but just for discussion sake what you do is go here and uh go to plugin click on manage plugins you can go this way one or if you don't want you can just click on your side panel and click on your settings and go on plugins and you'll see the same thing both are same both the ways are same and uh you have add created cloud you can also create skills by the

26:24way uh you know and convert it to an agent so you can ask cloud to do that also for you provided you have good number of token limits so I wouldn't do that if I don't have token limits so I would do this so let's start with um you know this particular plug-in business analysis plug-in and do an upload right you see with a skill and it's

26:50uploaded. Let's add the rest also. Um I don't think I can add all together but yeah I don't mind doing it one by one. It's an agent which is uh PR FAQ critic agent. So the last one is generic business analyst agent plugin. That's an agent for you uh as a general purpose senior with an unless you have all those. Now that's what the whole description about this particular agent is right. So now you have all of them in your system. So how do you really start uh you know talking to this right? Um start with slash business analyst. You see the agent is already initialized. So you're

27:45particularly talking to this agent, not globally telling uh co-work to do your uh you know your heavy lifting. And uh let's go back to our runbook. I'm just trying to get that. Yeah, if you see here, this is our frame one. Uh you are helping me to write a PR FAQ before drafting anything. Ask me the

28:09six question you need to answer. Customers paying end to end at checkout. do not write the dock yet. So you're just asking it to analyze first and you both are interacting and trying to get the first uh basic understanding out. Right? So / business analyst let's run it. Um sometimes the progress works, sometimes it doesn't but anyways doesn't

28:36matter. So what is a customer pain point in one sentence? what are the new features? So like this you got some some set of questions. So what I've done is instead of me sitting and typing I just prepared them ahead. So um let me just paste them.

29:03Sorry. Give me a second. I think I messed up with pasting.

29:31Sorry guys, I'm just getting the six answers for this. Uh just bear with me moment. I kept it somewhere in another uh tab and I closed it but I didn't copy it properly. Yeah. So I have some maybe additional questions. It came out from a different prompt but I'm just pasting it while it's working. What is the next? You have six questions or you can even put 12 questions depending upon your number. Right? Then what you do? Uh here is where this beauty of research happens. You are asking the agent here are the three last PRF from my team and

30:09last quarter checkout support tickets. Pull out the actual format and do it. I don't have it at this moment so I'm going to skip this step and I'm not going to do a research. Of course, you have them. You can do a research on your documents. There's no harm at it. But I'm just going to skip it. Um I'm just

30:25going to go for a little bit of analyze. So again, remember analyze happens between me and the agent. So this is me, which who I gave all the details. The research, the agent is going to do everything to get the research information on my documents or my history or whatnot and analyze is it me and I'm going to fight. me and the agent are going to find between each other to get the results out. I hope you're here with me, right? Let's let's do it. So, you see already there is a draft that has started, you know, being prepared in parallel. Uh it says an immediate

30:59release uh blah blah blah. I'm not going to read through it, but yeah, you can play around with this. It's as simple as that. Um I'm just going to skip one step. I'm telling again because I don't have any history of PR FAQs and I don't want to mimic any PR FAQs. That's why I'm just skipping that step. going

31:14forward. So I'm asking which one is uh weakest are you reach again? So I'm analyzing with the agent on talking to this right. So this is where we are and once it comes out with a particular set of uh angles and decisions just maximize this um certaintity of payment status zero effort and auto recovery. The weakest angle is duplicate charge prevention guarantee. So that's how you're going to handle the duplicate of charge prevention guarantee. That's something it's telling that you're the weakest

31:53point there. Now let's ask you to draft wrap the pair using our house format and strong angle. Mark every number as assumption instead of I don't what do you mean by house format? It's the because in step two I didn't give the research document. So I really don't know there's no house format but I'm just doing it just just

32:13to explain you how the flow goes. You can debate here and say no the weakest angle is not the duplicate charge coration guarantee. The weakest is zero effort and automated recovery. You know better your product. I don't know much to debate here because it's an example a wild example I took a common example. Uh that's the reason you know

32:30I'm just going with this flow.

32:38So till here one to four is something we all do every day in our business I hope right now it's trying to do another it's reading the file it's rewriting the whole uh document based on this there's no house format so it's going to assume there's a generic format and build it

33:12Yeah, I think it's it's updated. Well, you can see the difference how it was in the first version versus now. Uh the company name is not there because it was waiting for a format. I didn't give anything. So, you see the placeholder is already there. right now. What do we need to do next?

33:28Right. Let's look at the critic. This is this is the the the you know the critics that I really love about uh you know any discussion. I'm just going to download it. So I'm just going to check where it's downloaded. Uh where it download. Yeah, here it downloaded here. Now let's go there. So when it says do it open a new session always right when you say a new session always a new agent comes up a new session comes up it doesn't have any precontext of what you did in the previous session so that's the critic that's a small trick you can use the same model if you need or you want to

34:14change it to a different model it's up to you it's it's your choice I would say you know using a critic use a better model no need to be high the effort can be medium also it would work like charm. Now upload the document that was in downloads if I'm not wrong and I really don't know what is the file name if you yeah there you go and uh I'm going to say you're a skeptical head of payments regarding this for the first time what breaks what would you refuse to approve two ways you can just start the contact or no I'll use the same business

34:54analysis agent it's up to I will use the business analyst agents and I'm saying I'm giving it a context. I'm giving it a persona to say hey this is what you are and think it from that angle. Think from that angle and I'm giving a better model. Now let's see what it comes out as a critic.

35:30It's going to take some time because I've given a heavy model and uh so bat just wanted to interrupt for a second. Yeah. This persona is like the skeptical head of payment is the persona right where you mentioned. Yeah. Yeah. You can call it head of payment or your uh head of finance. If you are leading a financial document or you're head of a CTO, if you're doing an architecture review, it depends on your persona that you want to really have an eyes. For instance, I prepared an architecture document. I want my chief

36:02architect to look at it. So, I will ask some agent to think like a chief architect and say, "Hey, you know what? Think like a chief architect and tell me what are the flaws in my document." It could be as simple as I would have left this security section itself as null. Okay. And uh before that this business analyst is a plugin, right? You mentioned the business analyst backslash before the [laughter] it's an agent. So plugin is the way that cloud core calls it. It's an agent actually. That's why I showed you in

36:32this particular thing. Uh when you go to the plugin, you have manage plugins. You have something called general. It's an agent. It's it's assuming that as an agent. Okay, plugin is a feature they call it. I don't know they they don't keep as agents but in cloud code we have something called agent.mmd. When really

36:49doing coding we use agent.mds. Okay. Yeah. So it has sorry you have anything else? Oh wait yes. Hi I have one question here. Yeah. So who will design this plugins uh I mean agents for us? Sorry, who who will design who will uh create this generate this uh agent for us? It's it's it's it's it's I really don't know who does it. There are a lot of a

37:21agents that are available in the market. You can also build your own agent. Okay, that is already available and means the company will get the license of it and then make no license. There is no license. It's all freeware. These agents are also free. So you just go to Google or anything like Sapphire wherever is it and ask give me an agent for business analyst. It would give you an agent. I showed you the links right a minute before I showed you some uh uh let me show you this u these all our global GitHub which is available anything that MD if you see it's markdown markdown files are

37:55something that's an agent actually so it's will be available you can just ask for it um simple let's write give me an agent for a business analyst Just think like that. So here you'll see 11 business agents you can download and use um sorry here you have cross team reasoning agent uh for each one and one more thing that you can do is open a chat GPD session ask it to create an agent and give you the agent to you in a file it would give it to you in a zip like how I have it I'm going to share it to you guys. So that's another

38:40way to create an agent right or if you know something that you have written you can create your own agent. Yeah. Yeah. Yeah. One more thing means that nam I'm sorry you're cracking up. Uh the whatever the agent we are downloading. Hello. Am I audible? Yes. Yes. Can you please repeat it please? So I saw the isn't uh was available. So if we uh and used it in the cloud uh model but if uh miss the uh isn't whatever we have downloaded that is uh that can be used in any of the AI model

39:25or that is specific to the AI model. There is nothing specific to a model. A model is just a base for you. It's like uh you know I don't want to give this analogy but yeah I like it. Um a a model is think like a base for a cake and on top of it you keep multiple layers. So each layer is a skill or you keep it as an agent. It's up to you. So your base is your model. Any model can be kept as a base. It really doesn't matter. If you see here I changed from uh what is that uh I was in sonnet I changed to opus and

39:57I call the same agent. There is no difference. It really doesn't matter. If you have Gemini free, go to Gemini, do the same import and call Gemini Pro Flash. It really doesn't matter. Agent is agnostic. Okay. Thank you. Yeah. So great. Another question that you asked, can I write an agent? Yes, you can write your own agent. You can ask cloud to create your own agent. Um, so once it you type your prompt, ask it to create the whole agent and you can say convert this to an MD file for me. it would create an MD file. You can download it. You can go as a plug-in and

40:34manage plug-in and you can add it to an agent. As simple as that, right? So, that's another way for you to uh create agents. But anyways, let's get back with the flow. So, if you see here, I asked you the head of payments and what really breaks it's going to give refuse to approve. There you go. So, this PR FAQ is really bad. And what actually breaks? um I you know what a CI in MIT but yeah I'm I'm reading with you guys so these are that particular domains technical stuff I may not know much but anyways so it's going to give these kind of breakages and uh it's

41:15going to give why I can't approve you see right it thought from a different angle whereas the previous agent told everything is wonderful everything is beautiful go ahead man you you are amazing now what you can do is copy this whole thing and as simple as that go back to your old session and say uh this is the the review I got from uh another agent as simple as that or another model or another session it would understand so it will start redefining your whole documentation on top of this you can take it and give it

41:57to the critic if you're not confirmed. So it's always having a second second agent in your line to understand and do your work. See that review is correct stronger than my own pass key points hold up. Now it's going to ask you how do you want to proceed? You want to do this revise with all the

42:15findings just log in open open items. I'll say revise it cuz today I have my cloud kota so I can fire it and that's how you know we come back to this stage. I let it run. Uh I'm not going to kill it but we'll come back to that but meanwhile let's look at this. Um so if you see here we did the critic and then you know um you edit yourself you get back from critic you finalize a document. to see how fast just in this session we prepared a document. But if you ask me about the accuracy at this point, I don't know the accuracy of this

42:50document because I'm not a subject matter expert at this particular uh use case. But yes, it's as fast as getting things done. So back to my question, uh write fast or you want to think better, right? So you want to think better and write fast is what I would say and that's what I think someone also said the same thing. So now you are able to relate how we started off our session right. So um yeah as

43:21go ahead Akash. Yeah so uh Barat thank you uh for uh you know showing us this demo. I I have couple of questions. Uh first is that although this looks pretty much fascinating you know to to create a document while providing the a pretty uh artifacts right with with respect to your work that you do on a day-to-day basis but what is the actual use case

43:49here like what we are trying to achieve. So suppose you know once we provided all the steps right we got the document and that document uh let's say have a lot of discrepancies in it right with respect to the actual business that that needs to uh cater to right so in in that sense what should be your approach towards that document um I'll just reiterate what you said just say you're asking uh we got the document the document is good the document is that we reiterated it, fixed it. We're asking what is outcome of that uh document. Is that what you're asking

44:28Akash? Yeah, correct. That's right. Like basically what we are trying to achieve while you know having this document ready. Um okay let's think different way now. You might be writing multiple BRDs or FRDs or PRS you might be writing right on a general day how much time do you take to prepare these BRDs or FRDs or let's say a Jira ticket you're opening a Jira ticket for a new epic or something you would take good amount of time right what would you do at this current situation is you'll ask GPT or any other GPTs or in at least you'll use a row and build a uh a use case asking this is my

45:08uh epic and this is the problem statement generate a document for me. Period. That's what you're doing, right? But if you see the whole step and the way we use the agent, it's going very granular to the problem statement. We are going very close to the problem statement and trying to solve it. Now, what would you do is you would iterate the document multiple times and would take at least days for to complete your

45:32whole documentation at this point. Right? Whereas we are giving all as much as information to the system and trying to finish it as fast as possible using an agent. So the simple outcome is time time saved with an enriched u you know outcome document. So, so uh I have a follow-up question that we got the document uh we got the document that is enriched uh with all the pros and cons or uh you know with respect to uh the work that we need to deliver right uh now uh because in in MNC's or in in the professional environment right there there there would be a security concern

46:19uh for the data that we're providing to our internal systems, right? That would so in that case you know how you will justify that the document that you have created is up to the mark in order to resolve that particular problem statement and that would be the right solution to it to it. Uh very good question Abash. Um really nice that you're thinking in that particular context of your security concerns because we all face the same

46:48problem I think most of us in this call. So now let's look at the document that is created. Now, now first of all before we go there, I'm sure we all are using any cloud or charg that you're using right can be gemini or anything we all are using the organization approved channel local correct so let's be clear there or it should be a local model and a local referencing because we don't want to do any public don't take your public documents and upload it in a uh personal uh chart GPT or any account and get a document generator that's absolutely Totally not the right standard and I

47:26wouldn't vouch for it. Number one, so I'm just taking 10 steps back before I answer your question Akash. Um now that you have your office approved uh GPT or office approved you know your tools that you're using um we are using the agents that's a skill okay skill doesn't exposees your data because already you have your NDA and all your documentations done for relevant tools. Let's call let's take an example in this use case as cloud right cloud desktop or a cloud UI now you're going to use that you're using a skil or an agent what we just did and you prepared your documentation so your

48:05question is coming towards how do I prove the authenticity of this document now if you are if you see the whole flow right in this whole flow you were the one sitting and talking to the agent when I say you It's us humans. The one who's sitting and talking to the agent and trying to get the right facts and

48:27figures. Mhm. So you gave all the history to the agent. You gave all the history of all the information that it needs to prepare the document for you and you you at this place you know you technically analyzed with it. You fought with it and you read all the documents. Of course you're going to read you're going to read it. You're not going to just generate a document and send it to your uh engineering team and say hey guys build this particular product. this my B or FD and build this product. I don't want to

48:54do that, right? You are building it. You are contemplating. You are fighting with the agent and getting the right out. On top of it, you are also putting a critic hat to say whatever is a particular persona, you're asking it to fight back against to say that whether the document is right or wrong. Now you have done so

49:11many iterations to get one document out. Understood. It looks big. The process looks big. But you see while just talking between us we did this whole process in this session. I think I understood that. Okay. So basically you have so many uh uh models available. So you you basically keep pushing yourself to get the right answers in a in an accurate uh manner that that you want to achieve. Right? So you trying to to fight and analyze again and again when when you feel that okay the document that you got is not 100% accurate but at least meeting the criteria of reaching to 95% or 90% of you know what you're trying to

49:56achieve. Absolutely. You nailed the outcome of this. Understood. The only thing is what what is really saving than what you are doing is you're doing it very fast. Yeah. So yeah. Okay. The time consuming is something that you know we we are avoiding basically. Absolutely. And time consuming quality and thinking from different angles. You're going to wear different shoes at this time. It's not only who's going to wear his shoe. You're going to wear a shoe of a a CTO or an architect or whoever is it, right? You're going to give a different persona and wear that shoe also and think differently because the models that you're using is already

50:29trained for all the data set that you need. Yeah. Perfect. Perfect. Thank you. Thanks for explaining this in detail. Oh, this really good question actually. I might have not have put that in that uh angle but your question really nailed it down. Thanks so much. Great. Yeah, I just got this somewhere this code. So I just thought I wanted to include in this session but yeah as I said uh this get you write a document really strong with a very strong critic and uh you know thinking from different angles that's what this whole exercise that we did now for [snorts] and what really changes for us as I said the time

51:13that changes from days to hours and u yeah we can do it as fast as possible and you are covering every aspects in that document. You're not losing out on security. You're not losing out on um QA. You're not losing out on whatever the aspects, right? Uh a developer should think when you're building your any document. Um again, you know very well because you're using an AI, the the quality of the document will definitely be good. See, many people again say, "Oh, you know what? A will replace my

51:46job. This is a standard statement I get. Will it?" I'll be like no it'll help you to accelerate it do it faster and better with the quality outcome. Yeah it will it will basically amplify your work in a in a better way. uh you know if you if you get a breeze with the with the with the stack right with the AI stack with the AI tooling right so I think yeah it will amplify your uh or it will enhance your uh working the the way you work earlier working style and working model definitely you know as I said in my one of my sessions also uh I have a product

52:22person um who actually you know builds wipe codes builds a dashboard for me or builds that product in a smaller version and shows hey this is how I want it to look. Yeah, it it it gets it gets better to get approved from the different different stakeholders from a bullet point of view, right? Because like you have something already

52:45already made up, right? True. It's already ready. I don't need to break my head. I don't need to ask where do I put this button, where do I do that? It's already there ready. I just have to uh you know understand what he has maybe his template and just start building on top of it. I just need to wire it back. So you see how fast he helped me to uh think faster and he helped to get the UX approval. He helped to get the development approval. He

53:09helped to get uh the architect approval. Now that's what I'm seeing an agent and all these things would really help right. So it brings your quality, it brings everything into and and of course I don't want to repeat again and again the time is the fastest thing that you can really achieve out of this. And yes uh I just request you to uh please uh paste the uh the I would love to do it but but uh the I think someone someone needs to enable the chat right it is disabled right they don't have anyone uh they are mute you can find this agent online right why

53:43do you why does necessary need to paste it u no the links yeah you can google it yeah absolutely fine you can just search for the yeah so why that's not a problem like just for the purpose too like when when you have already already figured out the the the the demo section, right? Why you want to again you know again again you know put your efforts in in that somebody does anything you should always take it from there. So I don't go build an agent. Somebody has already built it like uh in the stack. I just download it and reuse it. Always, you know, build on

54:16top of somebody's build, not build from scratch. That's what I see. Right. But see, you're saying that you know getting buying from the architects or you know the SME or any other sponsors or domain experts. Okay. uh once you make this through using the using AI the AI agent I mean you need to actually share it with those people right to get their buy in how can you just say I mean I know it even if it mimics an architect mimicsme or a sponsor still you need to tell them hey this is what we have got based on the well maybe you not tell that you did it

54:53using an agent because they might not like it or they might not No no no you you have to tell you took agent help and you have to do it. Even a small write up and let's say my boss is asking I'm giving a write up. I'll say I took help from this GPT and it helped me do it. There's no there's nothing harm in nowadays thing. See the problem only if it's not just BS and just putting some dumb data in it and then saying hey

55:17review it. Nobody's going to review it. If it's really accurate to the facts and because we have done so much of effort in doing multiple iterations the quality is definitely going to be good and and by just seeing the document anybody would know it's generated by GPTs for information. Tell me in the organizations are not GBD

55:37generated. Mhm. But there are some you know there are some very few now I mean few but still there are I mean at least in the publishing world um you know they kind of try to you know block or prohibit extensive use of it but um nowadays yes you cannot do without it for the most for quite a few

56:02places. So if it is anything academics or anything related yeah that would be a nogo. Okay. But I'm saying like getting say even if that AI agent replicates you know the mindset of the architect. Okay. You still need to share it with the person and say okay do you agree with this? Right. You have to you have to always take it with you. See anyhow now let's say you you mimic the architect you mimic the CTO or whoever is it. Um in your documentation you'll have approvers list. Correct. So that approval list the sign off should be there from that person. I'm not going to sign because

56:37when I'm an engineering director, I'm not going to approve it without even reading it. My version of my flavor of critic will be there. So that is also going to add to your harnessing of your uh your whole documentation, right? Because the thing is um I can't I just can't hand it over to you know make it final without showing them because and say that hey you know the agent created agent is mimicking the whoever architect or whoever or Elon Musk mimicking them grock is mimicking him so I don't need to I this is equivalent to that person I don't need to go to no no you you should you should

57:14definitely get a physical buy in from the relevant stakeholders no doubt about it okay cool makes sense what I wanted Thank you. Yeah. Yeah. So basically you know basically I think uh what you are telling right is is a is a perspective mindset right when you share with your colleagues or with your team members that you have taken the help from AI to generate something good right some might take it as a positive note some might take it as a negative because like look there's so much hype is going on people are really scared uh scared or worried or I don't know something or or insecure you know

57:58uh that they thought that whatever the skill set they have built over the period of time will be will be obsolete within within within hours right by anyone else who who know how to tweak or fine-tune from the AI agents right who know how to how to get the things done uh in a in a in in a way right it doesn't doesn't make sense that whether you have taken the right approach or wrong approach. If you have delivered what is required right and you get the done basically you know you you have you have made it that is what is happening with the AI uh models and AI

58:38agents and then generative AI augmented NLP so many things are going on so yeah I agree so I think u that's again you know perception going on there a lot of things going on with uh agents and I'm sitting and we are building agents but yeah just to sum it up because we are on time so I didn't want to take others time we'll just uh put together what all we spoke today in the full session so remember these are your steps you frame it you research it you analyze you draft critic and then you finalize right now you have this just one last question but you have this agent right

59:18which is doing all this what I mean can it just be done on chat jig instead Yeah, I was just using cloud as an example. Chat GP also has the same plug-in. You can click on it and you can do it. Even Gemini, you can do it. It really doesn't matter if your company

59:34allows uh any other like dream club. No, without the plugin. I'm saying just basic. Can it be done without the plugin? Just the basic ji. It would do but it would not wear the hat of that particular analyst. Now for instance, why we say an agent? Okay. Um as I said from the beginning also models are are trained on let's say two 200 billion parameters or 100 million parameters. Now when it wears an agent hat it thinks in that

1:00:04particular angle. Now I'll say code reviewer agent just for discussion sake. It thinks very well like how I have to do code reviews. It's separate instructions are given. If I say a BA, BA has a certain roles and responsibilities. It thinks in that particular angle. You can definitely go ask the chat GPD do your iteration the same steps without the agent also you can do. No doubt about it. The quality that you get with an agent versus the quality you do it directly. I would say at least 40 to 50% difference because we build agents and we build the real agents not the skilled agents. So I've

1:00:41seen a tremendous difference between using an agent and doing the same billion parameter model versus non agent one. Make sense? Yes. Thank you. Great. Thanks. Uh back to the critic you know you have uh use a separate session as I said so that it thinks in a different angle. Not only this anywhere else also let's say you are you're you want to buy a car also and asking one session you did a full history open another session and say see this is what the analysis the first session we you know do your analysis and it would do differently it would think differently because it doesn't has your previous context of

1:01:21what you thought and uh give it a persona as as in this case let's say you're a industry car advisor I'm sorry I'm talking about cars because I love cars and uh ask it to attack. When I say attack, brutally hit your reviews, brutally hit your documents. Uh about it so it starts giving all the negative things. It's not going to do like hanky panky and be very soft about your documentations and of course battle with it. So keep your guard rails, you keep your business context, what are the judgments and uh your numbers are yours and the final signoffs are yours. So you have your

1:02:00guard rails, you have your leverage and this is the method I'm giving you. So that's pretty much uh with me.