blake1to1

0:02

S… Speaker 1 (blake1to1)

uh hi can you hear me yeah can you hear me yeah sure awesome um well thanks for thanks for thanks for i know it's a holiday thanks for joining um let's see um i think maybe maybe what we can talk about just very high level is like

0:29

S… Speaker 2 (blake1to1)

The current vision, both from a team perspective and also like a product perspective. I can I can tell you I'll just like be candid about what Dan is talking about and, you know, just know that this conversation.

0:50

S… Speaker 2 (blake1to1)

should just be you and I just chatting like don't let's not for right now let's not take anything to heart because I just kind of want to give you the picture that I'm seeing from other people and Dan and the kind of ridiculous things that he's seeing you know he's the CEO right so AI can solve everything and we have to be the ones to filter out what it actually can solve and what where we can use it so

1:18

S… Speaker 2 (blake1to1)

Very briefly, I'll try and keep it concise. The way that I'm looking at AI at AccuSource HR is both for the development team and to accelerate our velocity. So basically boost what we can get done in a development sprint. Two weeks ago, I presented to the board.

1:40

S… Speaker 2 (blake1to1)

about our AI strategy. And you and I can talk about that a little bit and maybe we can find ways to make it better. And then on the flip side, I think, did I already tell you about compliance source?

1:54

S… Speaker 1 (blake1to1)

No, you talk about a new system from Azure, but without details. So then there's the product side. And aside from the development team, the product side is legitimately putting AI into our products. And I'll skip the fluff.

2:20

S… Speaker 2 (blake1to1)

There are things that we want to do in Source Direct that have AI involved. Honestly, that's kind of on the back burner right now. It's not a priority. The highest priority right now is this new HCM product, which I think we've talked about, payroll, ATS, right?

2:47

S… Speaker 2 (blake1to1)

And I know I talked to you about this because I wanted to give you credit for it because it was kind of a blend of your AccuChat chatbot and then my on the cuff or off the cuff comment about clients interfacing with a chatbot. And that's what drew Dan to connect those things. So compliance source is a interim name for.

3:15

S… Speaker 2 (blake1to1)

This new category, what we're trying to do is get into a different line of business or not a different line. But so what we're calling it is team lifecycle management. And it's the full spectrum of hiring, onboarding, payroll, time and labor.

3:43

S… Speaker 1 (blake1to1)

all the way through to compliance of a workforce. So being able to know, hey, I'm going to hire somebody in the state of Florida. What do I need to know before I do that? How does it fit into my current company landscape? Yeah, I understand. And Yelena mentioned that during her meeting with Gurov, he showed her wireframes for a new iDriving system.

4:11

S… Speaker 1 (blake1to1)

Maybe could you possibly show them to me? Yes. Because I have already started working in the MCP for AkuSource, but it's just my initiative and I have no road map. Yeah. So can you see my screen? Yeah, sure. Okay. So there's been a lot of development on this, meaning not development as far as coding, but as far as like what...

4:43

S… Speaker 2 (blake1to1)

what the direction is so here's what's crazy we we've been working on and i can't show you the other project because i don't have yeah i don't have the login for that but

5:00

S… Speaker 1 (blake1to1)

We acquired this .NET project, Microsoft project, and we kind of want that to be the driver of this, but then Dan shifted because of our ideas in AccuChat, and he vibe-coded with Claude this, essentially. That's true.

5:23

S… Speaker 1 (blake1to1)

You know, it's essentially just a fancy dashboard with here's all the background screening stuff. This is safe source, tax credit we might get into. And then here's people. And this is more like the HCM side. You've got benefits, all of that.

5:44

S… Speaker 1 (blake1to1)

But this is the big piece is in this window over here. This is the AI chat body. So I stood this up. I basically took his JSX file and put it in AWS and connected it to Bedrock. So this is the big piece. So if I ask it, like, here's Priya Sharma. If I say, what's the status of Priya? Oh, yeah. Hold on.

6:19

S… Speaker 1 (blake1to1)

What's the status of Priya Sharma? You know, it goes off into AI world and brings me back, you know, Priya Sharma, employee status, all of her stuff, background checks in progress, right? It's a full status upload. So then I could say something like, one, two, higher.

6:54

S… Speaker 1 (blake1to1)

So Virginia, we currently have zero employees in Virginia, zero candidates in Virginia. Salary history ban, ban the box, e-verify is required. So it gives me a full picture of if I want to hire somebody, what am I? So anyway, you get the point. I don't need to demo this to you. Here's where it gets interesting. And this is what I am.

7:23

S… Speaker 1 (blake1to1)

This is where I'm going to be candid with you. We have an extremely aggressive timeline. And I'll just tell you the timeline and then I'll give you a quick backwards thinking of where we're at. And we're still, Gaurav and I are the two main points on this project right now. So we are still, every day we're back and forth about like how we're going to build this. Because basically by August, we need to have some form of

7:54

S… Speaker 1 (blake1to1)

background screening thing. And then by December of this year, we need to have a fully functioning people management, safe source, screening, compliance management tool. Those are the aggressive timelines.

8:24

S… Speaker 1 (blake1to1)

Where Gaurav and I kind of go back and forth is like, I want to build this as a completely separate compliance source project, Vue.js UI, and then we go to different backends for data. But the big problem is both systems, source direct and...

8:50

S… Speaker 1 (blake1to1)

The HCM system that we acquired are both kind of legacy in their own ways. Just like, you know how Source Direct uses Volt files to deliver pages and, you know, we have a lot of Volt forms and stuff like that, Falcon forms. The Microsoft project is predominantly web form.

9:13

S… Speaker 2 (blake1to1)

.NET web forms. So there's no API layer for us to stick a nice front end onto, if that makes sense. I don't understand. We have access to database for SourceDirect for the new HR system, so we could build API and MCP for every system and connect to this dashboard.

9:39

S… Speaker 1 (blake1to1)

Well, the problem, I think it's clear. Yeah, I agree with you. So here's kind of where I'm at with that. So if we build it, let's say we created a compliance source project in Git, Vue.js, it's all the latest Nuxt, Vite, it's all the latest Vue.js stuff.

10:00

S… Speaker 1 (blake1to1)

yeah and then we have a couple different options right i was thinking the like bff approach which is where you have a back end for the front end so we have a very very very thin small back end that is used as the main driver for uh orchestration so we go to different apis source direct safe source emcentrics which is the hcm um

10:29

S… Speaker 1 (blake1to1)

But the problem, the biggest problem right now, and I'm working through this, is like HCM. There are 531 pages that are all web form pages. They have no APIs. It's all web form. It takes source direct and times it by at least six. That's the size of this project. It's huge. So, you know.

10:59

S… Speaker 1 (blake1to1)

AI is my best friend now, so I've been having conversations with it to figure out how we can best migrate. And there's a couple of different approaches. I think we might be able to... There's a couple of different approaches. One is we create a complete new Vue.js UI for the HCM project. And that becomes this, right?

11:29

S… Speaker 1 (blake1to1)

Part of the reason for that idea is roles. The HCM project has so much permission. It has like six different roles. Like it has like admin, employer, employee, manager. Like it has a ton of different roles. So thinking to keep that the core.

11:54

S… Speaker 1 (blake1to1)

and create a new UI on top of that, and then use that UI to call source direct if we need to, when we need to. Anyway, let me pause. Let me pause. What are you thinking? I'm throwing a lot at you. I'm just curious. I think I understand the main problem with the forms, with the front end, but...

12:19

S… Speaker 2 (blake1to1)

You go, as I understand, to create a new system based for all of our systems and plus a new HR system. As I understand, I need to create a wiki with a full of functionality. Later...

12:47

S… Speaker 2 (blake1to1)

get analytics from previous companies and get information about... Because I understand that usually users use 10% of this system.

13:01

S… Speaker 2 (blake1to1)

Every day. For example, setting form using one time when you create a new company, onboarding another. And later you don't need this form. And it's a big problem for the legacy system because every day you create a new form, new form, new form. But users don't use it every day. Users use just 10% of...

13:30

S… Speaker 2 (blake1to1)

of forms and other. And I think you need to ask Guru or different, I don't know, to use this system and get the main solution, the main pattern. That's a good point. That's a good point. Like maybe what we could do is just target the client-facing...

13:56

S… Speaker 1 (blake1to1)

workflows and just migrate those to that's true that's true that's true um that's a good way to pare this down um okay and then

14:31

S… Speaker 1 (blake1to1)

So then there's this whole AI layer. So the two big things for me are roles, like user permissions, and the AI piece. The user permission thing, where I think it complicates this project is, like if I go to screening, yes, SourceDirect has API v2, but it knows nothing about my users. So if I'm a client user, and let's say I've got parent...

15:00

S… Speaker 2 (blake1to1)

companies and I'm only supposed to have access to two of those companies and then on top of that I'm only supposed to be able to see the profiles that I create. So we have to bake in to this new project all of the existing role RBAC in Source Direct and the RBAC that's in the HCM project.

15:24

S… Speaker 1 (blake1to1)

Yes, I understand. Usually we are using Gord Rails. For example, right now in Source Direct we have a Gord Rails to check system role and send only open information for this user, right? There we have a similar approach. Can I show you something?

15:53

S… Speaker 1 (blake1to1)

Yeah, it's about my MCP server. I have already shown... Do you see my screen?

16:17

S… Speaker 1 (blake1to1)

Yes. Yes, okay. Here is MCP toolbox for databases made by Google. Here, I prepare...

16:29

S… Speaker 1 (blake1to1)

tools, but this MCP tool is not using all of, it's not get all of information from database. I prepare, a moment, I can show you. I prepare tools, get profile by ID, get profile by number, profile status history, and I can store it in GitLab. And yeah, and here you can get,

16:59

S… Speaker 1 (blake1to1)

This MCP tool works with guardrails and gets information only for this user. Because... That's interesting. Yeah. I can send the link about it. It's Google. Yeah. Let me see what that looks like. That's cool.

17:23

S… Speaker 2 (blake1to1)

that's cool because we are not it's not new for a lot of companies a lot of companies have a similar problem yeah i mean because there here's the other thing so so take the roles take take the rules out of just like and that like what we were just talking about is that's basic software development we're trying to figure out has no new ai component to it that's just basic software development that we're trying to figure out what to do because

17:52

S… Speaker 2 (blake1to1)

Because the MCentrix platform, the HCM right now, it's old. It's very legacy. And I don't want to just keep adding to it. So I'm still thinking about that. Honestly, my current stance is I'm trying to figure out how long it would take for AI to migrate to Vue.js.

18:23

S… Speaker 2 (blake1to1)

keep the backend, create some wrappers. So like just new view components, but then when it needs an API, it creates an API that just invokes the existing code in the project. Does that make sense? And so the second layer to this is the AI piece. So obviously where the chat comes in is if I ask for

18:51

S… Speaker 2 (blake1to1)

What's the status of Priya Sharma? I'm going to go to SourceDirect for any background. I'm going to go to HCM for people management. I might go to SafeSource if there's continuous monitoring. So it's like the AI piece has to communicate with all of those projects. But I don't want, again, this is not...

19:16

S… Speaker 2 (blake1to1)

Technically, this is not part of your all's roadmap. This is a completely separate thing. So I'm trying to think through how that works. And that to me is where this back end for front end is efficient because it can orchestrate and create the different queries.

19:45

S… Speaker 2 (blake1to1)

that the model can then go to those databases for, those MCP servers for. So if we have a source direct MCP and we create a query based on the RBAC.

20:00

S… Speaker 1 (blake1to1)

for that user, you know, it can go directly to the table and query for it. At least that's the thought. Yes, and MCP have a similar approach. One moment, I'm trying to show you something, how it works in real life, I don't know. One moment. Okay, ready to start.

20:59

S… Speaker 1 (blake1to1)

Sorry. I tried to show you my solution. I have already shown it a lot of time ago. It's a DFIA. Do you hear about it earlier? Which one? DFIA. It's a workflow process. For example, here is my personal assistant.

21:27

S… Speaker 1 (blake1to1)

On the start, I have a trigger about... It's just a webhook. I have a special address and send information from my iPhone. After that, I have a transcription. And the main interesting here is a classifier. A classifier trying to understand...

21:58

S… Speaker 1 (blake1to1)

your prompt. Maybe it's a task, reminder, report or a different idea. And next he goes to the different approach, different way.

22:15

S… Speaker 1 (blake1to1)

Yeah, and in your case, it's similar. Here we create Gord Rails, or sorry, you request, send to classifier, and after that, yeah, and after classifier, you create a Gord Rails from a WS or a custom solution.

22:42

S… Speaker 1 (blake1to1)

And in addition to your SQL of different requests, WordRail for your query. It's normal, it's simple. And I really like this approach because you can see how it works. You can check and test.

23:07

S… Speaker 1 (blake1to1)

And they create a lot of similar... Two weeks ago AWS created a similar solution for... Really? Yeah, I can send you the link later. Yeah, let me know. I think it will help to create workflow for business, I think.

23:36

S… Speaker 2 (blake1to1)

yeah that'd be cool let me let me yeah if you can send that over that'd be great all right but the last five minutes we have let's take a quick second to talk about the team and the development velocity so yeah i mean look we i can't hide or i can't dodge the fact that there are hundreds of companies that are pushing the fact that they're building three times four times faster i have not committed us to three to five times faster i've said look

24:06

S… Speaker 1 (blake1to1)

bare minimum, we should be able to at least double our velocity. And so, yeah, it looks like you've got a couple of things here. Do you want to review those? Yes. Take a look. This is our development process. Yes, I have documented all the processes currently underway and well on their current status from my perspective.

24:34

S… Speaker 1 (blake1to1)

Yes, it's our process. Stages, idea, analysis, print planning, architecture, coding, and others. Yes, and I market blue processes in the Jira, as I understand. And I have already written you about my Jira agent.

25:00

S… Speaker 1 (blake1to1)

I prepare some screens, how it works. It's not screens, it's because I need to start to check it. For example, here.

25:18

S… Speaker 1 (blake1to1)

It's a new feature. I think a new user. And, for example, Karen can ask my agent, can you estimate this task? Yes. And my agent get information from Confluence, get information from our code, from, I don't know, different tickets, and trying to understand the estimate effort and what is missing to proper handoff for developer.

25:47

S… Speaker 1 (blake1to1)

Yeah, and after that, here is a different screen. The system trying to create effort, how many days we need, additional information, and have a button create task for this. And the last one.

26:15

S… Speaker 1 (blake1to1)

If my agent doesn't understand something, for example, missing profile, it's for a help desk. If a system doesn't understand something, for example, missed profile, the environment, specific scenarios, he will ask, please reply with.

26:37

S… Speaker 1 (blake1to1)

And after that, my system creates some test cases, tickets for a developer, another. Right now, I have already created these use cases. Analyze future, analyze bug, generate test cases, create test tickets, split into stories. How does this get deployed? How are you running this agent? Is it local?

27:03

S… Speaker 1 (blake1to1)

No, right now I'm working in my test Jira environment because I told you that I need to create AWS tunnel with Jira and my MCP server.

27:26

S… Speaker 2 (blake1to1)

Okay. So I like this. I think this is great. But I think what I like more about what you were showing me is your gap analysis and what – so with the end goal of increasing our throughput, I'm curious to know, are we tackling – Tackling? Are we –

27:58

S… Speaker 2 (blake1to1)

Is our approach the proper approach? Is it correct? Meaning, like, what I've been asking the team to do is, like, look, let's use these AI agents. Let's use Claude for getting through our feature work. And then once done, we need to bolster our...

28:21

S… Speaker 2 (blake1to1)

automated tests so that we can trust and be able to very quickly test the new stuff that we are putting in with AI quicker. But I'm also wondering, are there other areas of our process that we can fix or optimize for AI use to be able to make this better faster?

28:45

S… Speaker 2 (blake1to1)

If developers are still copying in the old requirements, that might not help AI build them. But if we can change the way we create requirements, that would be optimal, right? So can you go back to that other tab real quick? Yeah. So ideas, business requests, no AI analysis of incoming requests. So like the very first row, I would be like, okay, well,

29:19

S… Speaker 2 (blake1to1)

Could we set up some kind of your JIRA agent maybe to listen to any new client feature request tickets that come in and analyze it for any missing information? So if they submit a request and it's missing details that we always want, it can automatically reply and ask for those details would be amazing. Yeah.

29:46

S… Speaker 1 (blake1to1)

Of course, I can create this solution, but I'm going to do it in the second step, because my first step is to start by hand to...

30:00

S… Speaker 1 (blake1to1)

Check and test it. But later, it's not a problem to start managing automatically in workflow. Totally understood. And then when you say requirements analysis, no auto specs, acceptance criteria, risk detection, do you have a...

30:17

S… Speaker 1 (blake1to1)

ideal state for spec driven development like what what we should see in a requirement i would love to give that to gaurav he's currently working on that with his team the product team so that we can streamline this so do what is i mean do you have an example of like what a really good spec looks like

30:41

S… Speaker 1 (blake1to1)

I need some time to think about it. Make a note, if you don't mind, to circle back with me on that, because I would love to see from a first-person point of view what a good spec looks like, because that's what I want to get us to. I want to get us to spec-driven development, where we can just take what is in the requirement, load it up into a dev agent, and then be able to run it. Because ultimately, here's where I want to get.

31:09

S… Speaker 1 (blake1to1)

I want to get to the point where we have fully autonomous agents that can start to develop code for us. And the only way we get there is if we have good requirements, clear specs, right? So, all right, let me get your take. What, so, because I got to run into the call. What is your...

31:36

S… Speaker 1 (blake1to1)

Where do you feel like we're at when it comes to utilizing tools? I mean, it's clear we have a lot of gaps. It's clear we have a lot of gaps. But do you feel like the direction that I'm pushing the team is in the right direction? Or is there something else we should be focusing on first to get us to that 2x, 3x output reliably, of course? Yes. Right now, I believe our current bottleneck, it's in the...

32:05

S… Speaker 2 (blake1to1)

how tasks are assigned. Right now, developers focus on creating skills, education how to code works, and now we have an opportunity to create different... to work on a different...

32:29

S… Speaker 2 (blake1to1)

departments how to use AI, because I have already shown you the link, sent you a link about... It's a solution from JetBrains. I have no assets there, but maybe you will have assets. It's a solution from JetBrains to work with... It's a monitoring for all of JetBrains.

32:59

S… Speaker 2 (blake1to1)

It's a monitoring for all of agents. It's a Kubernetes for AI agents. I don't know how it works. It's just an article. It's just news. But I think... We've moved your organization management. So here's our account. I'm logged in. Oh.

33:31

S… Speaker 2 (blake1to1)

As I understand, you can send all of your agent monitoring information there and see how it works for all of this development process. How to Jira...

33:55

S… Speaker 2 (blake1to1)

So you connect it to your repositories, ticketing, this becomes our automated AI developer. Yeah, and my main goal to create something similar for our company, because I understand that JetBrains have a team city, we don't use a lot of JetBrains solution, but... Well, I would prefer to stick to JetBrains if we can.

34:23

S… Speaker 1 (blake1to1)

But we don't have to. JetBrains Air is the agentic development environment. Oh, shit. So this is their new code editor, huh? Yes, but it's a strange solution. It's a junior in a new... In a box. It's Claude code. I mean, this is essentially the Claude app. Yeah. That's all it is. Interesting.

34:53

S… Speaker 1 (blake1to1)

Run multiple agent tasks in parallel. Yeah. Duh. I can do that with Claude, though. Why would I use this?

35:00

S… Speaker 2 (blake1to1)

You have it cloud. Is it free? Yes, it's free, but you need to connect your agent. It's just frontend. I checked it one month ago, and it's not working with our cloud code without Anthropic account, but maybe they have already fixed it.

35:34

S… Speaker 1 (blake1to1)

Or use all agents via JetBrains. Okay. So you have to have a separate. So, okay. So it is still paid. I mean, we use JetBrains AI. So technically, I'll mess around with this. That's interesting, though. So interesting. Okay. I wonder, I mean, why would we use this? I'm just, I'm asking questions out loud. Like, why use this over?

36:32

S… Speaker 1 (blake1to1)

Claude within the IDE. Maybe you don't have to pay for the IDE. I think developers don't like this idea. I know. It scares me a little bit. Honestly, I use Claude to do a lot of stuff, but I still open up my IDE so I can look at the code there. I don't particularly enjoy the view of looking.

37:03

S… Speaker 1 (blake1to1)

But what's interesting about this is I could get rid of, I could potentially get rid of the team's Claude account as long as when we use this. I don't understand why this is breaking. What's going on here? It works only with subscription for, it works with the ChatsZPT codec, ClaudePagent, if you have Anthropic.

37:36

S… Speaker 1 (blake1to1)

Because I don't have a license. It thinks that I'm logged in here. Let me play around with this. Good start. It's better. They made it. They released it maybe one month ago. Oh, shit. It has the files in here. Yeah. This is the project. Here, let me show you while we're in here.

38:57

S… Speaker 1 (blake1to1)

the question though is can i point it to bedrock no right now it's not working yeah and i understand i don't understand that's what we need yeah i i will follow it and maybe later they create this solution yeah all right i'll look into that i mean this is cool honestly this could be the future because see here's the branches look at this um

39:33

S… Speaker 1 (blake1to1)

I mean, this could be the future. All right. Thanks for letting me know. I will take a look at this and see. I've got to get this pointed to bedrock. We'll see. But anyway, so do me a favor. Take a little bit, because what I want to get to in the long run is I want to get, obviously, bare minimum, we have to 2x our velocity. And that needs to happen.

40:00

S… Speaker 1 (blake1to1)

happened starting in July. Like legitimately the roadmap, we have to bring the product roadmap up. So that's my push for the team. And what I would love to hear from you is your opinion, what the industry is doing on how to get there. And like, I understand that.

40:21

S… Speaker 1 (blake1to1)

Yes, we can have these AI agents and they can write the code for us. And that should make things faster. But we can't, in my opinion, we can't take what we've been doing.

40:34

S… Speaker 1 (blake1to1)

and inject one little thing and expect an exponential growth. We have to make fundamental changes to requirements and testing and deployments and all of that too. And that's the part that I'm not thinking about yet, but I need to. So if you've got ideas, I would love to hear it from you. Okay. Cool. Thank you, Anton. Appreciate the conversation. Hopefully it was productive for you and you feel like...

40:59

S… Speaker 2 (blake1to1)

you kind of see a little bit more into the web of what I'm dealing with. Cool. Awesome. I appreciate it. Have a great rest of your day and have a great weekend. Thanks.

සාරාංශය

මෙම පරිවර්තනය ගැන AI අහන්න