WSW- Podcast Knitwit
===
Speaker: [00:00:00] You're listening to the audio portion of Workshop Wednesdays. Workshop Wednesdays is a free live discussion about topics affecting accountants, bookkeepers, and business owners. You can join the ABBO group in Facebook to participate live Wednesdays at 10:00 AM Pacific Time. Just search for A-B-B-O in Facebook.
This podcast is brought to you by schoolofbookkeeping.com, where you will learn, grow, and build a thriving bookkeeping practice. We have hundreds of lessons with almost every aspect of the industry. Start your free month today at schoolofbookkeeping.com.
Dan DeLong: Welcome everybody to today's Workshop Wednesday. This is a do-over from last week's QB Power Hour, which we had a little bit of fun challenges there with the timing and everything and Joel was [00:01:00] scheduled to come on and there was problems with the invitation, so wanted to extend the invitation, for, the workshop to really come in here.
And this is this is the next step for what we had talked about with becoming a promptologist using Nitwit, which we'll, talk about the his, tool and solution that he's made available for people. And really give Joel an opportunity to say all the cool things that he was supposed to say, during the QB Power Hour.
But the the goal here is that i- Joel, j- let's talk about your journey and how Nitwit came to be as a, solution for you because you created it because you were solving your own problem. [00:02:00]
Joel Slatis: Yes, that's true. We didn't get a chance to do it last week.
We'll do it this week, and Susan- Yeah ... and Silvia, both of you guys get the benefit of of very personalized attention. Rya, Andy, and Mason are with me, and the reason that they're here is because we are learning how to introduce Nitwit to people like you. So your feedback is gonna be very, important for us as we learn, as we figure out how we're supposed to be selling Nitwit to people and understanding their problems, right?
Understanding the issues that you're fa- dealing with, and that's why I was asking.
How Attendees Use AI
---
Joel Slatis: Before I get going Silvia I asked Susan already, are you using AI for work, and if so, how are you using it? Just a quick answer is all I'm really looking for at this point, but I'm just curious to know where you're coming from.
Sylvia: I do use it sometimes for work. I also use the [00:03:00] AI that's incorporated within QuickBooks and I like it a lot. And other times I just use the AI that's on Google or ChatGPT.
Joel Slatis: This
Sylvia: morning I had to file sales tax, and apparently here where I'm from they changed the tax rate by a little bit So I logged in, I logged into ChatGPT and it helped me figure out the new tax rate and the new computation, and I filed the sales tax in five minutes.
So it was very helpful.
Joel Slatis: So it saves time. Okay, good. Good. Yeah. Okay. We're gonna talk about some things today that hopefully will filter through to your practice and show you the true value of a product like Intuit.
Origin Story Behind Nitwit
---
Joel Slatis: So basically to answer Dan's original question, Intuit was born out of a need.
As you may know, I'm also [00:04:00] not just the CEO of Intuit. Intuit is a new product. It's owned by Timesheets.com, and Timesheets.com is the company that I've been running for the last twenty plus years. And we're a competitor to QuickBooks Time, if you're not familiar. You're probably familiar with us by now, but just in case you're not.
Overtime Rules Testing Problem
---
Joel Slatis: And while we're... I don't know where you're located, but we were- we are in California, and in California, there are special overtime rules, including daily overtime and another rule that's very specific to California called seven-day rule. And the seven-day rule m- is a rule, a, a state rule that says if an employee works on the seventh day of the workweek, and that day, it can't just be seven days in a row.
It has to be the seventh day of the defined start and end workweek. So the last day of the workweek. If they work on the seventh day of the workweek, and they've worked all six days before then, [00:05:00] on the seventh day, they're guaranteed overtime. So there's a rule. There's... So we, our software needs to be able to com- compute overtime that's different from federal rules that's, that, that basically says, "Oh, this person worked Monday, Tuesday, Wednesday, Thursday, Friday, and Saturday."
So on Sunday, they're due overtime. And in addition to the seventh day rule there are wage orders for people that have nannies and other special circumstances where there are exemptions to that seven-day rule if you have someone, for instance, working less than 30 hours during the week and less than six hours on any given day.
So there's some caveat to all of that, and we wanted to be able to support it because we have some clients that are using these kinds of things, and they wanted to be able to, track the hours [00:06:00] correctly and have it calculate correctly. So we started building that feature into Timesheets. So we built the feature, and then the engineers came and said, "Hey, Joel the feature's done.
Now we need to test it." So how are you going to do that? So I started thinking about it and I was like we, we-- first of all we need scenarios, obviously. So we have to have all these seven-day scenarios in a row where they work seven days, and on the seventh day there's... But on the-- maybe there's a day missing in the middle of the week, and we need to test that and make sure.
Or maybe there's a day that was over six hours of work, and did the exemption apply? Or maybe there was a day where there was over where there's a week where there was thirty-two hours of work, and does the exemption apply there versus twenty-eight hours? So we had all these different scenarios.
From 50-Line Prompt to Nitwit
---
Joel Slatis: So I thought, "I'll write a prompt that creates this data file for me." [00:07:00] And I had to... It's a CSV file that has all the scenarios, and inside of each file there's a bunch of columns and and each column has to have specific information. Column A was the date of the work, and column B was the clock in time, and C was the clock out time.
And I had to define all of this for GPT for the prompt in order to get the prompt that I wanted. And by the time I was done writing my prompt, it was fifty lines long It was a very, long and complex prompt. And as a result, I realized that their, the ChatGPT interface which is basically just a a text area, is it's not going to work for us.
We need something more complex, [00:08:00] something that allows us to build and edit complex prompts. And from that, Nitwit was born. So that's where Nitwit comes from. Does that make sense, everybody? And again, I'm... Right now I think I'm speaking mainly just to Sylvia and Susan, so feel free to interrupt anytime if you have questions.
Dan DeLong: Had a couple more join in while while you were talking there. So we got Brian from CBG, and Gideon also joined as well.
Joel Slatis: Gideon is one of our guys. Hey, Brian. How are you? Oh. Nice to s- nice to see you. Okay, so anyway, that's where Nitwit basically comes from. And, so I'm gonna close the door here so that I don't have little critters flying in here.
Dan DeLong: Yeah, we don't want, we don't want a, a, a bird visit.
Joel Slatis: They came in here one time. And they were flying around, and one of them took a dump on my wall. I don't know how he managed to do that, but I was, like, having to clean bird poop off my wall. Anyway. I know, it's really funny. Anyway, okay. [00:09:00]
Live Demo Setup and Login Fix
---
Joel Slatis: So we built Nitwit.
And what Nitwit allows you to do, and I'll just give you a quick demonstration here, is I can add my prompt, right? So here is my prompt, and I can submit that directly to the AI. Now all my buttons are disabled. I wonder if that's because I'm sharing. Let me just refresh and see if they come back.
Why are my buttons disabled? Let's see. Test. I can't submit anything. I don't know if that's- You're out of credits. Oh, I'm in a... Let me log out. I'm in the wrong account. That's why. There
Dan DeLong: you go.
Joel Slatis: Yeah. I'm on the staging account, that's why. Okay. I shared the wrong window. Thank you, Dan. That's helpful.
Dan DeLong: Do what I can.
Good.
Joel Slatis: Awesome. Okay, now we're back
Dan DeLong: in. There we are.
Joel Slatis: All right.
Stage Prompts and SOC 2 Agent
---
Joel Slatis: And let me create a new... This is an example of a long prompt, by the way. This is the kind of prompt you can [00:10:00] do when you... And I loaded this intentionally early so that you could see what it looks like. But this is the kind of long prompt that you can see.
You would normally add your prompt here, and then you would you can either submit it directly to the AI, or you can add it to the stage. And this is the stage. All of this is the stage here. And the stage exists so that you can create a prompt in pieces. Now, this particular prompt is a really interesting one.
You all are familiar with SOC 2? So this prompt is a prompt that I built and I'll get into more stuff later, but this since it's on my screen, I'll explain it. This is a SOC 2 prompt. It's an agent that I've created. This is all of the content for an agent that I can run that will help me go through the SOC 2 readiness so that I can be SOC 2 compliant, which is something that we're working on with Nitwit.
[00:11:00] Okay? Does that make sense? As I was saying, you can type your prompt here. This is my prompt, and I can add this instruction to the stage by just clicking add to stage. And when I do that, if I scroll down, let me collapse this and go down to the very bottom, and here you see at the very bottom is...
And I can edit this, and I can add more stuff to it. This is my prompt. Here is my really big prompt, and I can save that, and I can edit my prompt in pieces. I can drag pieces around. I can do all kinds of things to make my prompt exactly the prompt that I want. And that is the wisdom, the beauty of time of Nitwit.
Yes, go ahead.
Why Staging Beats Chat Memory
---
Dan DeLong: Now the and the challenge that I think I think we're experiencing as we interact with the, the [00:12:00] large language models, AIs that are, out there is this whole idea of like this chat going back and forth and what it does with that existing chat. This, this whole idea of memory is a, new thing that's coming into coming into the forefront.
But the problem is that, AI is having this goldfish brain where it you, end up having to essentially repeat yourself in order to- for it to have the same context, and that's what this staging is solving for, right? Is that it's, allowing you to create an on-deck circle of a prompt, so to speak to be able to send it once and done
Joel Slatis: type.
Is that- That's exactly right, and we're, gonna get to that. So let me let me try to get back on track here though- Yeah ... because I'll, let me-- [00:13:00] I wanna get to that, but in, in, the right context. So- Got it ... so basically, Nitwit allows you to create long and complicated prompts. So now how does this apply to your everyday work?
So what I'm gonna do is first I'm gonna show you the prompt builder. So you don't have to be an expert. This must look extremely complicated to people, and I get that because it's a more advanced thing. But let's take a step back and I'll show you how it can be simple.
Prompt Builder for Client Cleanup
---
Joel Slatis: So let's say you're a bookkeeper, you're an accountant.
You have a client that you need to do a cleanup for. So let's start with something simple like the, the client cleanup, okay? Which is something I think a lot of people experience often in this group, right? So the first thing you're gonna wanna do is Nit- Nitwit has conversations and topics, and basically you can think of these as folders and subfolders for your discussions.
So you-- when you go to ChatGPT and you start a new conversation, that conversation is in that window, and when you close [00:14:00] that window, the conversation kind of disappears. You have it in a very long-running list along the left, but after a week or two, it's gone. It's water under the bridge and it's behind you.
Here you have the ability to just categorize things. So, we're gonna start a new topic under the timesheets conversation. Or I could change the conversation if I want, but I just click New and and I'm gonna enter my topic. So I'm gonna call this Client Cleanup. Cleanup. All right? And we'll give the client a name Johnson Machining.
Okay? Some kind of a small local machining shop, and they've got a few people, and you guys do the books for them. And, so this is gonna be the client cleanup prompt. So I click Create, and by the way, to get here, let me just cancel this really quick. All I did was I clicked New, okay? To create a new topic.
So this is [00:15:00] the starting point. And once again, client- Clean up. And Sylvia, since you're definitely we this is all very new. It's, this, we only launched last week. There was a lot of interest, but for the workshop obviously people are busy. Thank you for coming. Come on off mute so that you can ask questions, 'cause I'm really talking to you and I wanna be able to interact with you, if that's okay.
But anyway, client cleanup, we're gonna create it, and that kicks off our prompt builder. Now, this is a really cool feature that basically interviews you so that you can tell the GPT or one of the models, whichever one we choose to use, what you're trying to accomplish. So I'm gonna say, it says, "What do you want?"
And let me make this bigger And Sylvia, like I said, if you can see it, yeah, I always [00:16:00] forget to to make these things bigger. So I'm gonna just zoom in a little bit. And I may zoom out if I need to just because of screen sharing here. But it says, "What do you want this prompt to accomplish?" So I'm gonna say, "I want to clean up a client's QuickBooks."
QuickBooks. QuickBooks Intuit will sue me if I actually type QuickBooks without the- Without the B ... capital Q and the capital B, so I had to go back and fix that. All right. "I want to clean up a client's QuickBooks. They are a machine shop." Okay? So we'll send that off. You just click send here.
Now, the AI in the background is thinking and it says, "Okay, what is the primary focus of the cleanup? Chart of accounts restructuring, inventory tracking, bank and credit reconciliation, or job costing setup?" And hey like I said, Sylvia, this you're, my single test case here, so I hate to be [00:17:00] picking on you, but out of these are any of these items resonating with you?
And if so, which ones? 'Cause we'll put them into our prompt
You're still on mute, by the way, if you're speaking. I can't hear you
Dan DeLong: Maybe she's feeding the peanuts. Propo-
Joel Slatis: There we go. Banking Friday reconciliations. Okay. So did you say Friday reconciliations?
Sylvia: Bank. Bank.
Joel Slatis: Bank reconciliation. Bank and credit reconciliations. Okay. So I can actually just click on these. I don't have to type them in, but I could also type in and P&L analysis Okay? So you can, click on stuff, but you can also type.
So I'm gonna send that off and it is l- working now, and it's gonna ask me its [00:18:00] second question: What is the primary deliverable format you want the AI to generate? A sequential action checklist, a diagnostic review, a step-by-step SOP guide, an executive remediation plan. I'll just pick a couple of these to save Sylvia, poor Sylvia.
And and then I'll send it one more time and and then when it comes back, I'm gonna click Build Prompt. Okay, so it's checking. It's doing its thing. And now I'm just gonna go ahead and build the prompt because it says the provider is uncertain, so it's not sure what to do. So we're just gonna go ahead and build the prompt.
Build and Edit the Cleanup Prompt
---
Joel Slatis: Now, here you can see the prompt was built for us. Let's look at it. Now this you should think of as a scaffolding, right? It is a beginning point to our prompts. It's not the end point. And here you can see the goal is to audit, reconcile, and clean [00:19:00] up historical QuickBooks account records for Precision Machine Shop, focusing on cash accounts and income statement integrity.
Your role is you're a senior forensic accountant. Let me back out of this one tiny bit so we can see more of it. You're a senior forensic accountant, certified QuickBooks ProAdvisor specializing in small to medium-sized manufacturing and machining enterprises. Key points: target business is an industrial machine shop with typical manufacturing operation overhead, prioritize multi-period bank account and credit r-card reconciliation, and so on.
So you see how it puts all this together, and then it adds constraints. Focus strictly on bank and credit card clearing, account balance validation, and income statement reclassification rather than full inventory system rebuilds. You can change any of these if this is not correct, or you could delete it.
So you could just click Edit, and then you could go into this instruction and say where it says, "Focus strictly on credit card clearing, account and income statement reclassification," [00:20:00] I could just say period here and delete the rest if for whatever reason the rest was irrelevant. Or I could add additional information, right?
So basically you just save that and now my constraints are updated. And then finally the output. A comprehensive cleanup workflow detailing specific audit steps.
Attach P&L and Reuse Monthly
---
Joel Slatis: So now you have your prompt for your machine shop, and you can upload And attach the P&L or whatever download files you get from QuickBooks, and you can put those in.
So if I had an... I, let me... I, think I actually do have a P&L that I can attach here in my downloads directory. So I'm gonna click upload file, downloads directory, or documents, downloads directory, and profit and loss by month. Here we go. So it's gonna read the file, and it's gonna add it right down [00:21:00] here at the bottom of my prompt.
Now, here's the power, one of the cool things about Nitwit. This cleanup client topic that has this prompt is saved forever, and I can always get to it by going to my timesheets conversation and then looking for my cleanup prompt, and it will pull this right up. And next month, if I wanted to... Let's say I was doing a profit and loss analysis on a monthly basis for a client.
What I'm able to do is I'm able to go in and delete this profit and loss statement and re-add the next one next month. Does that make sense? And then all I need to do is resubmit my stage to the AI, which is this button down here, and it will evaluate the new file with this instruction set. [00:22:00] Does that make sense?
So, that is the power one of the powers, one of the superpowers of Nitwit, is it is able to take a prompt that you've created. First of all, it'll help you generate the prompt, and then you edit the prompt yourself to change it so that it is basically ready to go. These are the things I care about.
This is what I'm interested in. And then every month, all you have to do is delete this file And then go back in and upload the latest profit and loss that you downloaded from QuickBooks, and then run your prompt. So now I'm gonna run the prompt. So to submit it, you just click Submit to Stage. It moves over here into the threads, and I can view my prompt.
This is what I submitted. It included my file. I'm gonna collapse it so that I don't have to see it. And it is using GPT [00:23:00] Mini to fetch the response. Okay? So let me just take a minute and Sylvia, I just wanna ask, does all this make sense what I'm explaining so far?
Sylvia: Yes
Joel Slatis: Okay. Do you have any questions?
This is my first-- One of-- This is not my first demo. I've done probably eight or ten demos by now for Intuit to different individual people. But this is Is, this path that I'm on making sense to you? Is it resonating with you? Is it what you're expecting to hear in this demo? Is it making sense?
Are you liking what you're seeing? I just wanna understand where we are so far to make sure I'm on the right track
Sylvia: Yeah, it makes sense. I understand what you're saying.
Joel Slatis: Okay, cool. By the way, have we met? Do you go to Scaling New Heights? Have we met? Did you... I'm wondering. Your voice sounds familiar.
Sylvia: No. [00:24:00]
Joel Slatis: Okay. You never know. All right.
Compare GPT Claude Gemini Outputs
---
Joel Slatis: I submitted to GPT Mini, and of course my answer came in. Now I can just drag this.
All these bars are draggable, and so I can drag this over and I can see my answer, right? So here's everything I need to know. Now, what if I don't like the answer? I can also submit this to Claude. So all I have to do is, I'm on GPT Mini, all I have to do is click over to Claude and I can resubmit. And in fact, I could also go to Gemini and resubmit.
And so Gemini and Claude are both working on the same prompt And pretty soon I'm gonna have three answers. I'm gonna have the GPT Mini answer, I'm gonna have the Claude answer, and I'm gonna have the Gemini answer. And okay, here we go. Gemini came in, and so now I have these two answers, and I can even compare them side by side.
This little icon right here allows me to [00:25:00] see all... And Haiku finally came in. So now you can see I've got my GPT answer, I've got my Claude answer, and I've got my Gemini answer. Now, you may not need to do this kind of thing on every search that you do, but if you're looking for... Not this is a client cleanup.
But let's imagine a situation where you're more interested in tax law, the new no tax on overtime law. And so you create a prompt that asks about no tax on overtime, and you want a second opinion. That would be an exce- excellent example of where getting three different opinions from three different engines would be useful because you don't wanna rely on just what GPT says because, as everybody w- who uses AI knows well by now GPT and other LLMs can hallucinate, right?
They can invent answers out of thin air. And so by [00:26:00] by getting a response from more than one engine, you can compare. And in fact, you don't even need to compare because this is a lot of reading. So we built a feature for that too.
Knit Insights Summary Comparison
---
Joel Slatis: That's called Knit Insights. That's this one over here. So I'm gonna run the Knit, and that's gonna work.
And what that does, while it's working, we can talk about what it does. What it does is it reads all three answers, and it's gonna give me a comparison. And so anytime you need a second opinion on some tax law or something that you're looking up for a client when you're doing your bookkeeping work, you don't have to go to GPT on your own and then go to Claude on your own and then go somewhere else on your own.
You can just come in, and I ran these one at a time, but you can actually run these individ- you can run these all together at the same time and then run the Knit. And the Knit will compare it, and it will tell you where they agree, where they disagree, and where any one of these engines came up with a unique answer that the other two didn't have.
So you'll have [00:27:00] agreements, you'll have disagreements, and you'll have comparison and you'll have what's unique between them. What it-- Let me see if the Knit came in. Sometimes... Oh, here it is. Okay. So you see on the Knit. Pardon me?
Dan DeLong: It just finished. I was-
Joel Slatis: Yeah. So on the Knit you get an executive summary, shared areas of consensus, right?
So it flagged zero COGS, uncategorized expenses. This is coming from my file. Remember I attached a file. And so it shows you where they all agree, and then unique contributions. So Claude said these things that no one else said. That's all from Claude. Gemini said these things that no one-- that no other engine said.
And of course, GPT Mini had these insights. So you have all kinds of insights, and then conflicts is the last [00:28:00] step, and you can see where they disagreed. So Claude said seventy, eighty percent direct labor, twenty to thirty percent overhead. GPT Flash and GPT Mini did not specify exact percentages, and the resolution is blah, blah, blah, right?
Utilities and anomaly. Claude flagged a hundred and eighty-seven and twelve dollars is unrealistically low. GPT did not specify it, and the resolution is Claude's observation is valid for a machine shop with three-phase power and CNC equipment. However, if this is a small job shop or other data represents only partial month activity, the amount may be reasonable.
Recommend verifying it against a twelve-month utility history. Is this information, Sylvia, useful? Would you find this Knit Insight comparison useful if you were doing some kind of a, a analysis or research to, to understand something? Is this something you could see yourself using? I'm just curious[00:29:00]
Sylvia: Yeah, I think so. I think this could be useful s- for cleanups because it basically tells you where you need to work on, right?
Or what you need to work on. And, Sometimes, let's say if I go over a profit or loss and balance sheet, maybe the first time I scan it, I don't see all the things that I need to fix.
So it would be nice to upload the P&L and the balance sheet here, and then go over what ChatGPT or Claude might think, and then- start doing the cleanup work.
Joel Slatis: Okay. And also, by the way, sorry, I had to get up to let the dog out because otherwise he'd be barking at me during our presentation.
I don't know if you heard that. [00:30:00] Okay. Thank you for that, by the way, Sylvia. Okay.
Save Prompts to Library and Share
---
Joel Slatis: By the way, now this can also work well for not just cleanups, but let's say, like I said, if you had a profit, a monthly, a profit and loss that you wanted to analyze for an executive every single month.
Or if you ha- And now all you have to do, again, is just delete the profit and loss, add the new one, and et cetera, and you have the save. And you can save this prompt to your library, where you can pull it up any time you want. So the way you do that is you click save to library, and you give it a name.
It already has the topic name, so you can stick with that. Or you can say whatever you want. You can add to it. And we also have the ability in Nitwit for you to create a profile and add your picture and a little bio. Now, I know what you're asking. Why would I do that? The reason is because you can make your prompt [00:31:00] public.
So if you have a particularly useful prompt and you wanna share it with the world, you click make public. And so when you share it, and you go into... And now this, I don't know if this is working, because this is a feature that we're still in beta, right? But you can go into the library up here And this is my private prompt library, and I have folders here.
Prompt Libraries Tour
---
Joel Slatis: In this Business Agents folder, this is where you can see my Instant SOC 2 expert, but I also have a GDPR compliance expert and a PCI compliance expert. And I have three saved prompts inside of my Business Prompts directory. And I also have the-- This is the saved prompts over here. I also have the public prompts, and this is where it would save if it were public.
And you see here it says, "Work in progress." within the next few days, this will be working. But anyway public prompts are where you can share prompts for your friends. [00:32:00] Maybe, Sylvia, you tell your friend about Nitwit: "Hey, this is great. You can do all this stuff. It saves me a bunch of time. I'm sharing my prompt.
You can go into the public prompts and pull it down from there into your account." Does that make sense?
And I wanna welcome Nancy. Hi, Nancy. How are you?
Nancy: Hi. Doing well,
Joel Slatis: thanks. Nice, to see you. Better late than never. Okay. We were talking about-- we were in the composer, and we were looking at a real-world example of a prompt that we used to analyze a business. And so Nancy, you kinda missed the beginning, but what I'll do is I'll offer you after the fact, because I want you to see all this, I'll just sh- I'll-- we'll work out a time, and, we'll show you the the stuff that we already covered so that you can stick [00:33:00] stay with us.
A-absolutely.
What?
Dan DeLong: What I'll do- Oh, yeah. No, go ahead. Sorry ... what I'll do, Jules I'll create this as a recording and mail it out to to the folks that, joined us, as well as those that ha- that are available. We are gonna repurpose the workshop and make it available on the workshop listing, so you can always go back and review that as well.
But, Okay.
Joel Slatis: Perfect ...
Dan DeLong: so the, the general-
Joel Slatis: Okay, yeah, I just wanna make sure everybody got the benefit. Yeah. Okay. So
Dan DeLong: now- Yeah,
Joel Slatis: absolutely.
Recap Multi Model Knit
---
Joel Slatis: So what we were looking at was we added a new prompt through our prompt builder. It created this scaffolding for us so that we didn't have to do it ourselves, and then we submitted it to the AI.
And just a quick recap, we got answers from GPT, Haiku, and Flash, and then we did a knit to sh- to-- that compares these for us so that we don't have to... And I'm gonna show you the full screen. We have our mini answer, our [00:34:00] Claude answer, and our GT answer, and here's our executive summary, our knit, which compares these three so that I don't have to read all of this, and all of this.
I have this knit comparison that saves me a bunch of time. Okay. So now I'm gonna show you a really cool use case. So Nancy, you're just in time for the use case.
Client Context Prompts
---
Joel Slatis: When you use KnitWit because you can store prompts, you have the ability to create what we call context prompts for each of your clients.
And this is another example use case of how it could be useful for you. So here's what I'm gonna do. I'm gonna create a new prompt new prompt, and I'm gonna call this Client ABC, okay? So this is for a client, and this is the context, and I'm gonna click Create. Now it's gonna ask me to start the prompt builder [00:35:00] again.
I'm gonna skip that Okay? And it had our old stuff in there, but we'll skip it for now. It doesn't matter. Now, this is client ABC. Now, what is client ABC? Nancy, do you have any current clients that you're working with right now that you wanna use as an example?
Nancy: I would say, Hold on a second.
Joel Slatis: Who's your biggest client?
You don't have to tell me their name, but just give me what, what type of business are
Nancy: they? A nonprofit. Okay. A nonprofit that is helps kids that have, or people that have heart difficulties.
Joel Slatis: Perfect. Watch this. So I'm gonna create a prompt now. I have a client who's a nonprofit that works with kids.
Okay? I'm gonna add that to my stage. Now, that's saved for later, for when I later submit the stage to the AI, and I'm gonna build out this [00:36:00] prompt. They are... And I'm just gonna make some stuff up now, but you'll get the idea, okay? They are they are located in California I'm gonna add that. They do about $150,000 or they raise about $150,000 per month.
They help special needs children with XYZ, right? And what I'm doing is I'm building a profile for this client. Now why am I doing that? What is the value here? The value is I'm gonna save this prompt. Client ABC is always gonna be under my clients, right? I'm always gonna have this. Anytime I-- let's suppose now there's no tax on [00:37:00] overtime, and I wanna know if that applies to this client.
So what I can do is I can say, I can pull this cl-- I can pull this prompt in from the library anytime I want, and that gives me my, context for my client, who they are, what they do, where they're located, what type of organization they are. They are a an LLC. I know that. We'll add that to the stage. They whatever else, whatever useful context.
They use QuickBooks, right? They use QuickBooks for accounting, X version. I can even just put essentials if I want. I'm gonna add that to the stage. And anything I can think of they offer the following services This is all relevant, right? They-- And I can do even a bulleted list here.
Service A, service [00:38:00] B, service C. Okay? And I'm gonna add that to the stage. And here you can see I've got my bulleted list. Oops, I included an extra bullet down here. We'll go back, and we'll just type in service D. And we'll save that. And we fixed our... And in fact, we'll just add A here so that we're consistent, and we'll save that.
So now we've got our prompt, right? This is context that now when I ask a question about client ABC, do they, does this client... So here's my question for this client, and I'm gonna make this a H2 so it stands out. Does this client have to pay overtime, et cetera, under certain circumstances?
Under, under whatever circumstances, right? [00:39:00] Now that would not normally be what you typed. Normally you would actually type the circumstances. So I can add that to the stage, and now if I scroll down here, I can see... And I'm zoomed in a little. Normally I have a big wide screen. When you're on your computer and you zo- and you zoom out to normal, you can see much, much more, and also these are draggable, so you can, see more.
But I'm zooming in to- so that you guys at home can see everything Okay?
Rocket Ship Context
---
Joel Slatis: And so this is my payload at the top of the rocket ship. That's the analogy I always like to use. When you look at a rocket ship, 90% of it is fuel, 10% of it is the tiny little bit at the top, that's the payload. This is the fuel.
All of this. This is my context that explains everything that the LLM needs to know about what we're doing. And and then this is the question. So now I can submit that question- That, that's a perfect
Dan DeLong: analogy Joel, [00:40:00] because as you, just mentioned that, right? I'm thinking of a of a the Apollo rockets, right?
That they fell off the, the stages fell off as it was going up into space, but the capsule is what they were doing to get to the moon, for example, right?
Joel Slatis: It's all the point of... The, the whole point is the capsule, and the whole point in this case is the question, does this client have to pay overtime under whatever circumstances?
And you we, submit that to the LLM, and then here's our answer. We get our answer. Now you can go directly to ChatGPT, and you can say, "Here's all this information." You could copy and paste this into ChatGPT. What we do at Nitwit that makes this useful for you is we give you the sc- we give you the ability to save this.
It's ready for next month. Any time you have another question, you go in here and you just edit the question. Does this client have to pay overtime? Under [00:41:00] what circumstances? That's no longer the question. Does this client get to take a tax break on XYZ? And then you save that, and then you submit to stage again.
And so now, And so you can repurpose and reuse a, a context prompt. So the way you're gonna w- use this is you're gonna have a context prompt for every client, right? In here it would say client ABC, here it would say client XYZ, here it would say client whatever. That's just one use case. Ob- obviously using it for different things here.
But if you had a conversation that was titled clients, and then each topic underneath the client is a different client, and e- each, each topic under this conversation is a different client that you're working with and you have a context for each one, any time you have another question that relates to that client, you're gonna be able to come [00:42:00] back to Nitwit.
You're just gonna have to change the question and resubmit the stage. That's all you'll need to do Does that make sense, Nancy? Is that making sense? It does
Nancy: make sense. I had a couple of questions. You may have already covered this.
Joel Slatis: No, please. We're here for you.
Knit Insights Explained
---
Nancy: So you can take from the prompt or you can input it through these three through Claude, GPT or Gemini and then your Knit insights will summate everything that you've just stated?
Joel Slatis: The Knit insight is a comparison.
Nancy: Okay.
Joel Slatis: We- it's our fancy word for saying compare. Okay. So if, you had a question like, does this client get to take a tax break on XYZ? That answer, here's ... I'm on Gemini right now. Here's my answer for Gemini. I did not submit the question to Claude or Mini, but I can.
Great. So all I have to do is click the Query button. So now I'm comparing- Also wanna mention
Dan DeLong: the the Faster and Smarter toggle there at the top there.
Joel Slatis: Oh, okay, yeah. Right now while this is [00:43:00] working the Faster and Smarter, we ... People always say, "Do I have access to the latest engine answers?"
Yes, you do. The faster models are defaulted to the latest Mini, the latest Claude and the latest Gemini and if I switch over it's gonna be the latest GPT, the latest Claude Opus and the latest Gemini Pro. These are the flagship. Probably I'll be honest with you, most of the stuff you do you're not going to need the smarter models.
But if you're doing some really in-depth analysis you can switch over to them. If you really want to, y- get a really great answer and you really wanna be able to depend on that answer and you want deep thinking on that answer, then go over to the smarter models. But the faster models are really gonna be probably pretty good for most everything you do and it saves money.
This request only cost me one credit [00:44:00] on, Claude. I'm on Claude. Same thing on GPT, same thing on Gemini. Now I have three answers and I can actually click this little icon here and view these answers side by side. Here they all are and to answer your question about Knit, it's just a comparison. So when I click Run Knit, which is the Knit insights, all it's gonna do is read all these things and tell me where they agree and where they disagree.
That's basically ... It'll, also highlight unique contributions. So in other words, whatever one says but the other two don't, it'll say GPT said this but the others didn't." So here is the Knit insights and let me go back to the non-compared view and then I'm just gonna make myself some extra room here I can collapse that.
Okay, and you'll see here I'm-- Oh, let me go to the knit. There we go. Critical findings the problem, available tax breaks, treatment on expenses, critical verification steps, and the bottom line. As currently structured, [00:45:00] the LLC receives no special tax breaks. This is the insight you're getting from it comparing all of these three different answers.
And of course, you can go into each answer. And the other cool thing that you can... There's a lot of little cool things you can do, but one of the really cool things that you can do with these answers is, let's say, you really like this GPT answer but you don't care for the... Or maybe you like the short answer at the bottom.
That's really all you need. You can edit the answer, and you can delete everything you don't care about And then just save And now you've got this little note reminding you that you edited the response from GPT, but now you have a short answer, and this becomes curated knowledge for client ABC that you can return to anytime you want because it is saved in your conversations.
And-
Nancy: Two questions.
Joel Slatis: Ex- pardon me? [00:46:00]
Nancy: Two questions.
Credits And Toggles
---
Nancy: If what is a credit? And the second one is the security. I know Claude is supposed to be the most secure AI out there right now for the general public, but what are the others? What's the status of the others and for Nitwit?
Joel Slatis: Okay, fair enough. So a credit is basically just our word for token but we calculate it slightly differently.
W- we, want it to be a to- are you familiar with the term token?
Nancy: Yes.
Joel Slatis: Okay.
Nancy: No.
Joel Slatis: So tokens are, like, approximately three to four characters long, and that doesn't really make a lot of sense to people, right? So we said, "Let's make sense of it." So we're in- instead of calling it a credit, a token, we're calling it a credit, and because it is not exactly the sa- we don't wanna do the exact same thing as a token, where it's three to four words.
So a credit [00:47:00] basically is one penny. It's easy to understand that way. Okay. This search cost one penny, and that's what a credit is. So if it costs five credits, it was a five penny search. And by the way, you can see when you're using a fa- the faster models, all of these basically cost just a penny each.
So you're basically gonna be able to do tons and tons of, searches. It's very... It doesn't cost a lot when you're using the mini models. But like I said, if you're doing something more complicated or that you just wanna really get an in-depth, long answer on, you can always switch to smarter.
And we actually have quite a few cost saving and cost control things. You'll notice there's other toggles here, web search, compare, and context, and these three each do different things, which I can get into later if you'd like. But basically they're all cost saving, but also they control and tell the models how you want the search to occur, and it helps you get better information that way.[00:48:00]
And these are toggles that you don't have with Claude or GPT or Gemini. And they're just, either not available or they're turned on by default. Context is turned on by default on GPT. So is web search. So you can't turn those off, and that way each search that you do costs more than if you were to maybe you don't need web search.
On this prompt you do because you're asking about tax law. However, on other prompts you probably don't. If you're a- if you're asking it to analyze a- A profit and loss statement and give you an analysis. You don't necessarily need web search for that. If you're, And now context is when it takes into account the previous questions, and you see it, it-- we have this little orange bar here to, to show you when context is on or off.
And in this case, this question is [00:49:00] unique, and it doesn't care what previous questions I asked. So I can turn context off. That saves me money in my request to the LLM. One day, this is gonna get expensive, right? You know how they figure out a way to raise prices on everything. So eventually you'll be very happy to be able to turn context off because in a regular ChatGPT window, the longer it gets and the more questions you ask in that conversation, every time you click submit, it sends the entire history of the conversation as context, so it knows what you're talking about.
But maybe you don't need context all the time. So in Chat- in Nitwit, you're allowed to turn it off.
Security Roadmap
---
Joel Slatis: Now, the second half of your question was about security. Now I cannot speak to the security of Claude, GPT or Gemini. That's not really in our purview. We are providing them throu- as is. So whatever their security is is what you're getting.
[00:50:00] Unfortunately, that's just the way it is, Nancy. We don't have any control over the security. Now, Time Sh- Time Sheets. Nitwit being a new company, we are working on our SOC 2 certificate. We're planning on having. We do not have it yet. We only launched beta last week, as you probably know. We, do not yet have our SOC certificate, but we're working on it, and we will have it.
That is our goal. We're gonna be able to-- Because we're gonna be dealing with a lot of accountants, they're all gonna be asking about security. Now the other thing that we're doing is we're gonna be installing what we call secure folders. So if you have conversations like these client conversations that you wanna keep secure, these folders that we're gonna be installing for to- for certain topics, 'cause it doesn't have to be for everything the, they're gonna be, they're gonna have extra encryption.
They're gonna have much higher level of encryption than just standard. Now, that isn't to say that we're not secure. We're using the-- all the industry [00:51:00] best practices to secure our data and our website. We don't allow training. We're not allowing third parties access to our database in any way unless it's absolutely necessary for the operation of the website.
And we're, we're, gonna be, as I said, implementing all of the necessary steps for SOC 2, which means we'll have multi-factor authentication installed for everything, and we're gonna have processes and contingency plans for breaches and policies for all that stuff. It's all coming. It's not done yet, but it's coming.
And in addition to that, you're gonna be able to enable 2FA on your account. So when you log in your Netwit account, you're gonna be able to turn on 2FA multi-factor authentication if you want, and you're gonna be able to store your client conversations in a secure folder that has extra encryption on it that makes it very hard for anyone to get in.
So the security that we have is going to be SOC 2- and secure folders. Today, [00:52:00] it's just the re- regular security of the website, but as I said we're in beta, and we're not really a target for anyone at the moment because we're we've got five customers. And eventually we hope have to- hopefully have thousands, but one week after launch, and-
We we don't have a lot going on yet. The best security in our case for now is obscurity. No one knows about us, and no one cares about breaking into us. Because we're so small, it's not worth anyone's time. But by the time we get anywhere near being big enough for anyone to care, we're gonna already long ago have installed the features that I'm talking to you about, secure folders, SOC 2, and and the ability for you to enable MFA on your account if you so choose.
Does that answer your question?
Nancy: It does. The concept is great, by the way, 'cause I have several times gone, and I'm sure others have too, where you put something in one, and then you copy it to the Gemini, and then you... Because you're trying... You don't like the first answer, or it doesn't give you as much information.
So this is great.
Dan DeLong: Yeah. [00:53:00]
Joel Slatis: Cool. And the context- Dan, yeah, the- ... is repeatable. Sorry, Dan, you were gonna say?
Cohort And Sharing
---
Dan DeLong: No, I was, I'm just gonna try to wrap up here 'cause we're coming up on the top of the hour. Wanted to make sure that we have everything that we wanna say today. What I truly appreciate about Netwit is that it allows you to stage your prompts cr- craft your prompts, it outside of sending it and then having a chat to go back and forth to, to refine it.
And then it allows you to send it to multiple models because, as Nancy said different models give different results, and allow you to stitch them all together. And you didn't even talk about how you can share, Ted your, response- Oh, I haven't even gotten into that. Yeah ... With, your client.
You can share your response. What we're, planning on doing with Joel and Netwit is having a live cohort where we... It will be much like what we have here [00:54:00] today, where it's a, where it's a workshop. We'll have a, a general topic du jour for, each of the live sessions that we have.
But it's more of a, a conversation where we, have this going back and forth to, to really craft a- and get the most out of Netwit. And I'm putting that in the chat here. There's a link there to, to pre-register for it. It is going to be starting at the end of September. So- Coming up soon, a couple weeks here.
We'll have four live sessions that will be recorded like this so that you can review them later in in School of Bookkeeping. It's a paid cohort, so it's $147 for the four live sessions. However, if you have a paid Nitwit subscription it's free. So this is a great way [00:55:00] for Joel and their team to to gather feedback about about Nitwit, make it better because they're in very early stages of the launching of this, product.
And they wanna make it as useful as it can be. But they've already done such, great work up to this point. We just now need to, Now we need use cases. And Joel and their team need use cases of, okay what would people use this for so that we can... They, can... We I'm, like, part of the Nitwit team now.
So Joel and the team can, make it even, better. So that's the purpose of the cohorts. And we'll, we'll get more information about, that cohort out to the School of Bookkeeping family and community, and we'll [00:56:00] make those make those available. So Joel, any, any last parting words that you wanna say about about Nitwit or things that are coming up on the horizon for you?
Joel Slatis: Yeah. We're adding a bunch of features. We're continuing to work on this. I'm more interested in hearing from Nancy, to be honest. I wanna understand her needs and stuff, but we've run out of time.
Dan DeLong: Yeah ...
Share With Clients
---
Joel Slatis: Nancy, I didn't get a chance to show you the share, so I might as well since Dan brought it up.
Dan, is that okay? Do I have a minute for that?
Dan DeLong: Yep.
Joel Slatis: Okay. Basically imagine you wanted to share this answer with your client. Now you just click Share, and you can fill in their name Dave, and you can say dave@clientabc. And it says, "Joel has created the share." And you can even give him a note.
"It looks like you don't have to pay taxes," right? And then you click Send, and this will send an email to [00:57:00] Dave with a link. And when he opens the link, he will see the prompt.
He will not see the stage. So all this area on the left is not, is hidden from the client, but he will see this thread and he will see the answer that you shared.
And if there's answers on multiple chats, he'll have the links at the top of the page too. I don't know if I can... I think I'm only sharing this window, so I can't show you what a share looks like. But in the cohort you'll be able to see it because we're gonna be- Yeah ... A little bit more organized and we're just learning.
I'll be honest with you, I'm just learning how to explain this to customers right now. That's why I'm asking you so many questions, Nancy. But anyway you'll be able to share this with clients directly and and they'll get a very pretty screen that's designed to look like a nice share and explain everything and and so that becomes a useful tool in your arsenal for [00:58:00] communicating information to your clients when you're working in AI and you've got an answer that you wanna share with them.
Gotcha. Okay.
Sign Up And Wrap Up
---
Joel Slatis: I don't have-- I could go on for another hour, but, For, for now, that's a quick... Nancy, do you have a, a Nitwit account already? Did you sign up for a trial?
Nancy: No, I just signed up for what y'all sent out, so
Joel Slatis: All right. Okay ...
Nancy: just signed up on that website that you sent
Joel Slatis: out. Perfect. So can you go to...
If you want, you can do it right now. You can get a free account on Nitwit to play with this. All you have to go to is Nitwit, with a K, like in the logo up here in the top left, nitwit.ai, and you'll be able to sign up for an account. You could sign up with Google, or you could create a username and password, and that would be, And then you'll be able to actually start using this functionality right away. It'll take you through a little welcome tour, or it should, although that's been a little bit buggy lately, so it may or may not fire. We'll see. But I would love to hear [00:59:00] your impressions, and you can always reach out directly to me as well as Dan.
You know how to get a, a hold of Dan, but I'm easy. I'm Joel@Nitwit. And I wanna hear from you directly. I wanna hear anytime you have the smallest little problem, I wanna hear all about it.
Dan DeLong: All right. Oh,
Nancy: I appreciate it. Thank you.
Joel Slatis: Yes.
Dan DeLong: All right, Joel. Okay, well- thanks for joining the workshop today.
It was really useful to get a little under the hood of what you can do with Nitwit. Looking forward to having you help us with the with the cohort i- coming in September. And then next week on the workshop is we're gonna be talking about some the ani feedback for for QuickBooks, and the topic is the matching in the bank feed.
There's there's some changes, with, the bank feed that we wanna talk about next week on the workshop. So hopefully everybody has a great week, and we'll see you next time.
Speaker 2: [01:00:00] And that wraps up another insightful episode of Workshop Wednesday, brought to you by schoolofbookkeeping.com. We hope you enjoyed today's discussion and took away some valuable tips and strategies to enhance your bookkeeping practice. Remember, if you want to stay ahead in the world of bookkeeping and accounting, be sure to visit schoolofbookkeeping.com.
With a wide range of courses, resources, and expert guidance, you'll find everything you need to sharpen your skills and boost your career. Don't forget to subscribe to our podcast on your favorite platform so you never miss an episode. And if you enjoyed today's show, please take a moment to leave us a review on your podcast platform of choice.
Your reviews help us reach more listeners and continue to bring you the best content every week. If you have questions or topics you'd like us to cover, reach out to us on social media or send us an email at [email protected]. We love hearing from our listeners and are always looking for new ideas to explore.
Thanks for tuning in to Workshop Wednesday. [01:01:00] Until next time, keep learning, keep growing, and keep excelling in your bookkeeping journey. I am Dan DeLong, and we'll see you next
time.
