05 11 Sayso Okr Transcript

Simon Hauck: Hello. Hello. JP Kalambay: Hello. Simon Hauck: Howdy do. JP Kalambay: Not bad. How are Simon Hauck: All right. JP Kalambay: you? Simon Hauck: Nice chill Monday JP Kalambay: We Simon Hauck: morning. JP Kalambay: are Simon Hauck: Yellow, Ben Boudreau: Hey, what's going Simon Hauck: you have we work. Yeah. How is that? Ben Boudreau: Uh, it's good. It beats being at home. I ran into Alex today. German Alex. Simon Hauck: All right. Ben Boudreau: He was here. Simon Hauck: Yeah, I might need to I might need to do that, too. There's been there was a couple days last week which were kind of a struggle. Just being at Ben Boudreau: Uh yeah, I don't have like a desk. Simon Hauck: home. Ben Boudreau: I'm just like on my laptop. I have like a portable monitor sometimes to set it up. Those that is kind of nice too because sometimes I have like these screens and I'm like I get like too much stuff going on now. Simon Hauck: Yeah. Yeah.

00:01:17

Ben Boudreau: Even though I'm switching tabs I'm like lock down whatever is Sam Gharegozlou: All Simon Hauck: That makes sense. Ben Boudreau: next. Simon Hauck: Hello. Sam Gharegozlou: right. Simon Hauck: Hello. Hello. Jad Haidar: Hey team. Simon Hauck: Uh, G's probably out. JP Kalambay: What? Sam Gharegozlou: Welcome. Ben Boudreau: Yeah, Jad Haidar: Hey Ben Boudreau: I think so. Simon Hauck: Okay. Jad Haidar: guys. Ben Boudreau: Um, okay. So as far as ASO is concerned, JP do you want to delete this and JP Kalambay: Sure. Uh, so Friday and then this morning so far. I just pushed some changes. Uh, I changed the questions in the survey. Um, well really only number three. It used to be like was anything it was essentially was anything kind of off about this session like kind of open-ended. It's still open-ended but I just changed it to what could have made this better. Um, thought that might be a little bit more positive framed. Um and then also since the last session was run by the uh like Paul's team member um there wasn't actually a survey uh because I had it as like a gated feature where only my user can um make like trigger that survey on at the start in like the form to create the session.

00:02:46

JP Kalambay: Uh, so I now just made it open so that like any session that is created that has eight or more participants automatically gets a survey. Um, so I think that should be a bit better. Uh, I haven't actually seen any sessions run on here besides the ones that we've tested yet, but uh, especially after the last one, he did a really good job of like kind of pushing it to users. Again, there weren't very many, but he just kind of really talked it up throughout the stream that I'm thinking there's a, you know, optimistically maybe some of the people who were there like, oh, like maybe I'll just use this myself for, you know, my group of friends or whatever. Um, so having the survey there, hoping that gets us some some feedback. Uh, and then I have like an email alert so that if there is ever a session that qualifies there, I'll get notified. Um, uh, yeah. So, and that's kind of what I just finished right now. Also debugged the join um link.

00:03:43

JP Kalambay: There were a few bugs where if you were a new user and you access the link, it would kind of bug out and you would have to just join with the join code. So, just fix that as well. And then um yeah, now kind of just gonna talk to Socrates a little bit, see about next steps about small improvements. But other than that, the main thing this week is going to be prepping a pitch to the Top Shot guys to try to get that event going that's like not a stream, but just an event that they can kind of promote. Um yeah. Simon Hauck: Nice. Um, I just had a thought that you said that maybe in the survey as well it might there might be a nice opportunity to say like to mention to the users that they can use it themselves like just putting that copy in like you can use this go off and have fun kind of thing or something. That way we can like cuz they might not think to actually use it themselves.

00:04:46

Simon Hauck: They might think that this is only a space that they're invited into that they can participate in, not host themselves. I don't know. I don't know if other people feel anything about that, but sounds good. Ben Boudreau: I just I think like surveys are hard enough to get people to complete in general and if they they don't figure that out that they can use themselves then that's it's next thing we got to fix in the product you know like so I would just like keep the survey like as concise as possible Simon Hauck: True. Ben Boudreau: to get as much information as we possibly can but I small people who will go through that is um so is Simon Hauck: Okay. Ben Boudreau: The Paul is kind of like running on it own now. JP Kalambay: Yeah. Ben Boudreau: Nice. Um, that's cool. Um, do Okay. So then for the top shot stuff, do we need do you need you want any help like JP Kalambay: Possibly Ben Boudreau: contacting people setting this up or JP Kalambay: I'm gonna kind of give Matt like the day to get back to me when I messaged um like a group chat that I have with uh Guy and I'm not sure if it's Guy or Gee, but I'm going to say Guy.

00:06:03

JP Kalambay: Um, and Matt Matt said that sorry guys said that he was like out for something that day and wouldn't get back to me, but he would pass along the information. Um, so I'm going to give him the day to get back to me on that, but probably like end of day I'll book him again like, "Hey, can we set up a call here?" Um, yeah, they had mentioned that like the only blocker might have been from the NBA side. Um, but yeah, I don't know. I also don't think that they that they know the product well at all. So, I'm going to see if I can kind of get them on a call and like pitch it to them properly so they know what it is and then they can kind of go from there. But, Ben Boudreau: And I wonder if after the call with Naim where he was talking about like the branding like if we the branding is out of JP Kalambay: yeah. Ben Boudreau: it then maybe the NBA doesn't need to be concerned. It's a top shot thing you know not an NBA thing.

00:06:58

Ben Boudreau: Um, okay, cool. It seems like everything's going along well. Um, I don't know what else to say. Do you do you need anything Do you need anything from us? JP Kalambay: Not currently. Uh maybe in that pitch meeting if we can kind of maybe I don't know if that's going to be like kind of like just a casual talk or if it's like a more formal pitch. If that's the case, then maybe you kind of do that as a team. But if it's just a chat, then maybe I can handle it myself. I'm not sure. I'll just kind of wait to hear back from him and then I'll kind of have a better idea. Ben Boudreau: I think either way though it's still kind of be nice to like hone the value probably even if it's casual it's still kind of fish. Yeah. Simon Hauck: Uh, when do we want to do that? If we want to hone that JP Kalambay: I I'm I'm going to I'm pushing for as soon as possible, but at the same time I he just told me on Friday,

00:07:57

Simon Hauck: into JP Kalambay: oh, like I'll I'll pass him along the information and we'll get back to you. So, I don't want to poke too quickly, but I'll probably if I don't get back to him by like let's say 3:30ish, I'll probably just poke him again. But yeah. Simon Hauck: Okay, good Ben Boudreau: Yeah, good job. That's awesome. Um, okay. I don't really have much I don't have anything else to say on say. So, that was quick. Um, do you want to use this time for anything else? Sam Gharegozlou: Um, is there nothing else? JP Kalambay: I don't think say so related though. Sam Gharegozlou: Okay. Um, I want to pivot real quick to the A website. It's in a very, very, very sad state. Um, and I mean, you're seeing the the updates that Dapper might do. Roam's sort of been tweaking around himself uh with design. It's probably looks like it's AIE, but um it it looks quite sick. And then comparing I don't know if you have seen it, but this is the it's a work in progress, but uh this is like the base MVP that's being worked on.

00:09:24

Sam Gharegozlou: Actually, we're going to share Ben Boudreau: It's been terrible. Sam Gharegozlou: Yeah, like it's it's pretty sick. JP Kalambay: Nice. Yeah. Sam Gharegozlou: Much better than what it is now. And you look at that little cool adapter and Axium company at the bottom here. Um, and then we go to axum.co. That's not my guess. JP Kalambay: I a little Sam Gharegozlou: Uh, JP Kalambay: outdated. Sam Gharegozlou: little is an understatement, but Simon Hauck: I like the tiles at the bottom though of the website. Sam Gharegozlou: yeah. Simon Hauck: It's very simple, but it's elegant. Sam Gharegozlou: Yeah, Simon Hauck: But yeah, the rest of the website is very Sam Gharegozlou: it seems off to and our logo isn't showing up here for some reason on the top left. Simon Hauck: white. There's not a lot happening Sam Gharegozlou: There's for sure supposed to be a logo here because you can click it and it takes you back to the home screen. Simon Hauck: here. Sam Gharegozlou: But I don't know where it went. But it's very wide, very blank. I mean, we did it on purpose because we just wanted to retool and and get something up.

00:10:37

Sam Gharegozlou: But even if we want a basic thing, I think we can do a better better job at it. We can remove hammer and tusk because that's pretty much dead. Yeah, I don't really want to link back to relative, but we can. It's pretty cool. Ben Boudreau: What's the deal with that? Sam Gharegozlou: were there other Oh yeah. Ben Boudreau: Is that still going for Sam Gharegozlou: Yeah, they're still there. Um, yeah, we can do I think we can do a a cooler thing like highlighting our companies but also like some of the work we've done and then we can like you know write some more blurbs of like this. Oh, you know, blah blah blah by X million people and blah blah blah million transactions that we may be able to throw some of the older stuff in there, Simon Hauck: Mhm. Sam Gharegozlou: too. But if we can like take a first stab at doing something basic, but not too basic, but more better than this right now, I think would be great. whatever we can do in a few days, a week.

00:11:55

Sam Gharegozlou: Um, there's going to be a lot of traffic going to the new sites in the next probably a couple days. And I doubt people are going to scroll all the way through them, but if they see this, I'd like them to go to something that's like, you know, somewhat cool. Uh, Jad Haidar: Yeah, Sam Gharegozlou: yeah. Jad Haidar: just a quick question is uh the the a website is on Webflow, correct? Sam Gharegozlou: I don't Jad Haidar: I believe I believe it's on Webflow. So, it's kind of like a drag and drop editor. Sam Gharegozlou: know. Jad Haidar: Uh, we'll have to look into it and see see what So, as far as I know, Web Flow doesn't have like an AI like a like a VIP coding thing. So, we either have to build it manually. Uh, Simon, maybe you and I can look into look into it. Uh, I'm pretty sure it's on Webflow. Simon Hauck: Yep. Jad Haidar: I just haven't I haven't been there in a long time, so I'm not 100% sure. But sure, we'll figure it out.

00:13:01

Sam Gharegozlou: Yeah, Simon Hauck: Yep. Sam Gharegozlou: I don't know. Dappers was lip flow. I don't know what they're doing now, Jad Haidar: Yeah, Sam Gharegozlou: but Jad Haidar: looks like it's on Vers now. So, I'm guessing it's an XJS app. Sam Gharegozlou: yeah. Jad Haidar: Let me see what's going on. Sam Gharegozlou: need to update that as well. Yeah, let's see what we can do in a short period. Uh but like also like this services section too like JP if we start doing um start talking to folks about you know consulting you things like that well they're going to go to our website so I want them to at least be some mention of our expertise at the very least and we can throw in um AI and agents and all that fun stuff in there. Uh again it's pure marketing so most of it we can we can massage the wording and things like that but it' be good to get this updated too because like yeah we've leveled up like crazy on blockchain decentralized web but we're also leveling up on the AI field and things like that we have a lot of SAS experience too etc etc so would be good to highlight

00:14:16

Simon Hauck: Okay. Yeah, I can jump on that. Uh, do you want that top of the list for things to Sam Gharegozlou: from the basic thing. Yes. Simon Hauck: do? Sam Gharegozlou: Um like something better than what there is now and we can regroup on a fuller wider scale redesign Simon Hauck: Okay. Sam Gharegozlou: later. Simon Hauck: So, kind of keep it simple. Just give it a little brush up, a little snazzier. Sam Gharegozlou: Yeah, Simon Hauck: fix the worries and get rid of the dead links kind of thing. Okay, Sam Gharegozlou: I think so. JP Kalambay: I think it'd be a pretty big improvement even just to port it to Nex.js that alone I think would would Simon Hauck: cool. JP Kalambay: be a you know more modern update. Sam Gharegozlou: Oh, and and this this f****** thing is we're getting way too many this spam emails. Um I don't know if there's another better way to fix that or or what. Um like Simon Hauck: What kind of spam? Like people or Sam Gharegozlou: Uh, Simon Hauck: bots?

00:15:18

Sam Gharegozlou: I think it's a combination of bots and people, but it's a lot. Simon Hauck: What are they? Is it just like I want a job or is it like Sam Gharegozlou: Okay, I'll show Simon Hauck: Okay. Sam Gharegozlou: you. like 39 emails and it's all just crap. Come here then. But I'm not clicking on any of the links cuz Simon Hauck: Yeah. Sam Gharegozlou: F like maybe we do a reset to remove that and just create a basic form and that might stop the spam for now. That's crazy. It's It's just all like this. Simon Hauck: Yeah, people just cold calling and then sending random other bits of Sam Gharegozlou: It's very few cold calling. Simon Hauck: information. Sam Gharegozlou: It's mainly just useless. It's just spam or useless stuff. Like this is cold calling obviously, but yeah. Simon Hauck: It's like a spammy cold call. They messed the billion people. Sam Gharegozlou: Yeah. This stuff is just like pure spam. Some fishing. Simon Hauck: Yeah. Sam Gharegozlou: Yeah.

00:17:11

Simon Hauck: I wonder if there's something in the form that we could vet or filter out stuff before it gets sent. I don't Okay, Sam Gharegozlou: Maybe. JP Kalambay: Yeah. Simon Hauck: cool. Uh yeah, that should be done. Can be done. Sam Gharegozlou: All right. Simon Hauck: Uh Jad, do you want to do a call on that later? Jad Haidar: Yeah, yeah, yeah. Let's let's talk about this uh right after this call with Simon Hauck: Okay. Jad Haidar: Okay, Sam Gharegozlou: Um, do you want to do a daily and then move on? Ben Boudreau: Sure. Jad Haidar: let's do it. JP Kalambay: Yes. Ben Boudreau: Um I I've been working on um kind of amounts to like a big skill, a claude skill that is kind of the the missing piece in that diagram like the agentic diagram we made before. And uh I'm about to try and integrate it, but I'll just kind of a diagram. They share my screen. JP Kalambay: Yep. Ben Boudreau: Uh the basic idea is that it kind of the goal is to kind of keep everything in the repo as possible.

00:18:53

Ben Boudreau: And then there are There's like one agent who kind of will read the data and the product research and he'll spin up these other agents that run on a prawn to to kind of make sure that the bets that we create are going through and progressing. So on the runners the idea is like we'll we'll fire up a runner for some sort of work they want to do. Like one of the suggestions I made was that we get a lot of support emails about people not understanding that um when you get a subscription it's like per or So one of the things suggested was like hey uh put up a dialogue when you subscribe that says by the way this is on a per organization basis something like that right that one's pretty simple example the idea is like you start like make a make a markdown file that says hey this is one of the bats we're working on and then this runner will like he'll wake up like once every couple hours or something and check and make sure it's it's going right.

00:19:54

Ben Boudreau: He'll make he'll kind of have the context of the bet that's running and he'll he'll fire up these I don't know if it's really well documented here, but he'll he'll fire up like coding agents to go do their work and he'll he'll be the one who checks it and says like does this look correct? Then once it's in the codebase, he will he will like f follow up like a week later or longer and check the metrics and make sure that like hey this is working. And the idea is like there'll be a there'll be like the main code branch and there'll be like a separate branch for this guy and he'll like he'll write all his information in there. So everything he learns, everything he finds out will just be separated in a different branch and that will eventually kind of get merged into main once it's like completed and we have learnings that we find out. So that's kind of the overall structure. I have no idea if it's going to work, but I thought it came out to kind of a cool place.

00:20:50

Ben Boudreau: So I'm going to try and integrate it into Toby now and see if I can get it all set up correctly and and firing. And then I think the idea is like a we could plug in the thing that Chad's working on as like the coding piece and now kind of give it power-ups and say, "Oh, now the coding works really well end to end and maybe it'll work well with what GE is working on. Maybe we'll fiddle with it a little bit so that all the agents that fire run through AIOS. I not sure but it's kind of more like the structure that I'm working with here." And yeah, I'm going to see if that works. And kind of a cool thing too is I was I worked on a lot over the weekend. I was using uh the clawed uh phone connection thing. So I was like out on the weekend. I was like driving around. I got to the store. I was like let me quick check like send it a prompt on my phone.

00:21:47

Ben Boudreau: It's kind of fun. Anyway, Jad Haidar: Thank Ben Boudreau: that's the main thing we've been doing. I'm still working on um I kind of have some time today to pruning some of the Google Cloud uh bills we got Jad Haidar: you. Ben Boudreau: going on and also Jad Haidar: All right. Exciting. Uh, let me also share my screen. I'll show you guys where I'm at. All right. Are you guys able to see my screen? Okay, nice. Uh, so yeah, making good progress on the sand castle pipeline. Uh, I have uh one of the pipelines now. It's uh it's running. I'll walk you I walk you guys through it uh real quick. Uh just um as a quick recap um I've had uh some issues with the the pipeline being deterministic. So I had a bunch of uh like instructions that were baked in at the prompt level. So I'll also show you real quick here what what it looks like at the the code.

00:23:19

Jad Haidar: So the main two pieces are or like three piec pieces are the this mains file which is uh the the harness itself that runs everything. Uh the prompt is what gets fed to the main orchestrator. So there's one agent, an orchestrator that then um runs the the plan through like a multi-step uh pipeline. I think it's 10 10 steps where it it checks u for uh the the provided plan. So the way I I'm doing it right now, I just provide the plan for it, but later on it's just going to pick up the plan automatically. Um but now I just provided the plan. Uh this is the um I I created the plan for the MCP server for uh for for Toby. Uh I figured like that would be a really good uh good test to to get it to run on. Um I backtracked on having a pipeline for the front end separate from the back end. I found that that that wasn't that wasn't working as well as I was hoping uh I was hoping and then I just um backtracked on that.

00:24:34

Jad Haidar: Now it's it's working a lot better and the MCP server is going to span back end and front end uh work. So basically I just created the the spec first um ran it through like 14 different review cycles then from that created the plan on top of it and this is what is being uh fed to the uh the pipeline itself and the first the first step. So here it says uh so it starts the the sandbox okay and then it starts the agent and it looks for the first slice that is not um completed yet and um basically it will just look for the environment to see if there's any pending work uh work that was like so so it has resume uh functionality. So if uh the pipeline for whatever reason crashed halfway through or like we ran off credit ran out of credits for the like the 5 hour window, we can run it again. It will pick up where it left off. Uh that was another thing I was also like working on. Uh and so here it's it's just looking uh it's saying uh let me see.

00:25:49

Jad Haidar: Yeah, it's saying fresh, fresh start, no prior commits, unttracked files are planning artifacts only, which is the the plans, the uh the plan and the spec. And it just starts doing a bit of uh investigation. it scopes out the first uh slice and it starts executing on it and then um it it reviews the each slice multiple times until there are there's no more any actionable feedback. So it just runs automatically. Uh it it implements it completely end to end. And then um one issue I came across was claude versus claude uh like during the review cycles. So just using flood to review cloud review itself. So another uh thing I'm I'm planning to work on as well is add u support for for codeex as another uh run review loop uh that is both baked in at the implementation cycle and at the end as a as a kind of like a like a gate by which it needs to pass. But yeah like you can see here it's it's working. It's still doing its thing.

00:27:07

Jad Haidar: uh the changes here. So it pushed uh four uh work in progress or sorry this is a bit more than four. Yes, six uh work in progress commits and yeah it's just um in every commit it will add a log of the decisions that it made and any open questions so that at the review time it will check the decisions and it will try to poke holes in the decisions and whenever it finds anything it's going to rerun the whole the whole uh implementation cycle again um up until it basically gives the the green light that there isn't anything that it's able to catch. And the next thing I'm going to work on is adding codecs and spin three different reviewers at the same time that will attack it from different angles. One will focus at security. Another one will will focus on like spec compliance and a third one is going to focus on on bug fixes. Um, the reason why I'm being so um like meticulous with the reviews is because in previous cycles I found it to be so like it not being deterministic.

00:28:21

Jad Haidar: It often it it rationalizes itself into doing the wrong things. It's very easy for it to do that. and having these kind of uh adversarial reviews and now adding a codeex is going to hopefully make it a lot better. Especially adding like a different model with different weights and different strength and weaknesses. It's going to catch things that Claude uh missed and and vice versa. And yeah, but the hope is by the end of this is that it's going to ship APR that is we can just try trust blindly. But there's a fair bit more work to do. Uh testing each round takes a fair bit of time, right? The pipeline is working end to end. I got I was able to to get like a like a full full slice implemented correctly. I had to do another separate review cycle on my own just to sanitize it and I was still able to find issues. So, I'll only call this a success once I'm once it finishes the entire thing and then once I do another review cycle, it doesn't find any like major issues, right?

00:29:36

Jad Haidar: It might find like nitpicks, which is, you know, something we can live with, but major blocking issues or or bugs are are not something we can we can feel confident about. Um, and yeah, it just takes a while, you know, having going through all of the review cycles and then there's a bunch of gates as well. U you know, it it it runs a llinter, it runs the all the test suites, it tries to build as well, make sure that nothing is breaking. Um, another step later on as well to to further strengthen and for for us to have like more confidence in the pipeline is to add a like an end to end testing pipeline where it spins up a u a a Chrome browser inside the sandbox headless Chrome. So meaning there's there isn't any there aren't any buttons but um the the AI agent can interact with it. it can click on buttons internally and that's kind of that that will be like the final uh the final gate where runtime errors are not things that we can catch deterministically through just unit tests.

00:30:45

Jad Haidar: We need we need like an full end to end uh flow but that is still a a fair bit far off. We're going to have to do manual cing in the meantime now until I get that done. There's a lot of, you know, issues that I I have to be mindful about when implementing it, but um making solid progress. It's it's actually doing pretty well. I'm quite happy with the direction it's it's taking, but it yeah, it's just still needs a fair bit more more work before I I I feel confident about shipping it and saying, "Yeah, like can just rely on this thing. It will do the thing. It will do the work for us. We can just trust it blindly." Um, but we're getting there. So, that was kind of really it. Um, when you guys end up uh interacting with it later on, we're just going to host it on a on a web server so that it it does its thing on its own. We don't have to like trigger it ourselves.

00:31:43

Jad Haidar: Uh, but if you guys want to interact with it, you have to first build the the Docker image. uh provide the API key for for cloud and later codecs and um I think that should be it in terms of setup. You just build the the the image and then you just run the the pipeline against the given um plan that is generated by the uh agent skills uh plugin which is another thing we'll also have to like wire up to this so that it does it it does it on its own. So yeah, do you guys have any any questions? JP Kalambay: Just excited to try it Simon Hauck: Yeah, I'm very curious about that review process. JP Kalambay: out. Simon Hauck: Sorry, Ben, you're about to talk. Ben Boudreau: There you Simon Hauck: No, it was just it was mostly that it when I was I jumped back in last week to do like the homepage because I try to take a crack at it every now and then and it's definitely that process of it like thinking it's done a good job and I have to manually go in and say,

00:32:51

Jad Haidar: Yeah. Simon Hauck: "No, you didn't do a good job." And so that's the moment I'm really curious to see how that's being Jad Haidar: Yeah. Yeah. Simon Hauck: tackled here for sure. Jad Haidar: Um, Simon Hauck: excited. Jad Haidar: one thing I found is that it really struggles with like pure UI like front end stuff when it's trying to like build something up to spec. If if it if you're just giving it the the freedom to do what whatever it wants, it can generate like a okay enough result. But if you want it if you want to coersse it to build something that you know fits a very specific vision on the on the UI side of things specifically is where it really struggles and adding that uh feedback loop where it's able to run run the the app in Simon Hauck: Mhm. Jad Haidar: the browser take screenshots compare it to whatever it's trying to build itself and iterating over that hopefully should should bring us a lot closer. But even even then, yeah, even then it's going to be a bit iffy. I'm I'm trying to find ways to make it more deterministic.

00:34:01

Jad Haidar: There's a fair bit of work that needs to happen at the code level for it to be able to do that. Um, having a like strong design system will go a long way, you know, like these this is like a very rigid structure. They use these things all the time. not just like components, but you know, layout guidelines, color guidelines, spacing guidelines, all of these things, things that can just tap into instead of um figuring out figuring out on its own. And then we can add review gates that specifically check for the compliance of these um guidelines. That should really help us a lot. That should definitely help us a lot. Simon Hauck: Yeah. Jad Haidar: And if we have like multiple uh adversar adversarial reviews that are from different attacking it from different angles using different models that should also like bridge the gap. Um but yeah yeah it's it's still it's it's still it definitely struggles on the on the front end part. Simon Hauck: Yeah. Jad Haidar: uh backend stuff is a lot easier for it because it's a lot more deterministic and it can just write uh tests way more effectively for it because it's mostly like it's not a visual thing.

00:35:23

Jad Haidar: It it it re reasons with it a lot more effectively. Um, but once you need to deploy the back end, that's where it starts becoming a bit less effective. But even then, it should be able to use CLIs to just deploy, monitor the deployment, get feedback. It's all deterministic stuff. The the front end is where it really struggles, I found in in my experience. But we're getting there. Simon Hauck: I think v visual language is a lot more abstract than we give a bit credit for and it's a lot more detailed than code is. Jad Haidar: Yeah. Simon Hauck: Even though we like to think it's the opposite. Jad Haidar: Yeah. Simon Hauck: But Jad Haidar: Yeah. Yeah. It It's still It still struggles with it a fair bit, Simon Hauck: yeah, Jad Haidar: but yeah. Trying to figure out ways to to um bridge that gap for sure. We'll see where. Simon Hauck: exciting. Jad Haidar: But yeah, now you guys can see here um it finished implementing everything and now it's doing the it's at step eight.

00:36:31

Jad Haidar: it invoked the review finding skill which I shipped here as a as a as a vendor uh skill. I have another skill here called commit message. We can talk about that another time. But yeah, it's now running the first review cycle. You can see here the adver adversar review of phase zero and starting to do it investigation here. So let's see where it lands. That's pretty much it for me. Sorry, this was a bit of a lengthy one. Oh, what's up? What's up, Ben Boudreau: What's so what what's stopping you from shipping it right now and us using Jad Haidar: Ben? Ben Boudreau: it? Like if if you were to get a piece of code and say how to make this change right now, would you use this? Jad Haidar: Not without oversight, but I would I could Yeah, Ben Boudreau: Yeah. Jad Haidar: not without oversight, but I can Yeah, I I can definitely ship this over and you guys can start experimenting with it. But it still needs manual review.

00:37:33

Jad Haidar: It still needs, you know, make keeping an eye eye out on it. Um, it's I'm not confident with it. Ben Boudreau: I I think that's I think that's fine though. But I guess I'm saying that like if I was to start a new piece of work, Jad Haidar: Yeah. Ben Boudreau: would I just kick this off and then I would come back in like 30 minutes and then I'd be like, "Okay, now I'll review from here rather than like the 10 iteration Rust committing quad to get to that Jad Haidar: Yeah. Yeah. Yeah. Ben Boudreau: point." You know what I mean? Jad Haidar: Yeah. Yeah. 100%. Yeah. Yeah. Ben Boudreau: Yeah. Jad Haidar: It's it's it's ready for that. Yeah. Ben Boudreau: Yeah, I think you should put it up there. Jad Haidar: Yeah. Yeah. I'll I'll wrap this up. I'll push the code. Um and I'll hand it over to you guys. I'll I'll also generate some some instructions for you guys to like run this locally

00:38:04

JP Kalambay: Oops. Jad Haidar: and yeah try it out. Try it out. We can port this over as well to other other repos. It's now going to be a onetoone port depending on the context. as we get closer to this uh you know the the dream of having the like a template repo that you know uses the same tools and we we start like when when we start working on new projects it it should get a lot easier. We can just bundle this in into that repo as well and it can just get the work done automatically for it. Um, I might even consider externalizing this as a like as an npm package that we can install so that it just becomes easier for us to share it across different projects. But yeah, I'll um package all of this up and send it to to you guys uh JP Kalambay: Cool. Jad Haidar: for you guys to try it out. Ben Boudreau: Sit down. JP Kalambay: Um, for me, I kind of already got into it, but yeah, just kind of try to get some small wins on ZSA while setting up the shot event.

00:39:24

JP Kalambay: It's more or less Jad Haidar: Nice. Nice. Uh, you should be able to copy a bunch of things that that are already implemented here. JP Kalambay: a Jad Haidar: You can probably just copy the the prompt and the mains file and ask your agent to um adapt it to say so and it should it should do a pretty good job at it. Uh I would also suggest invoking the uh what is it called the review uh finding skill. Once it it finishes it finishes the implementation just keep doing that until you reach a point where like you feel okay like this is this the the port is good to go now you know because it always misses things uh from the first from the first time you really need to like keep poking holes at it until it it uh it gets there. But yeah, these are the two main files that you should care about. There's this scan secrets thing here that I added later. I was just being paranoid because there's a good chance it might, you know, do things that we don't want it to do.

00:40:28

Jad Haidar: And this one was just make sure that it doesn't um ship any secrets to the to the repo. It's just like a mechanical gate deterministic. It makes sure that there isn't any any API keys that are are are leaking out of the the repo. And there's a few smaller things inside the mains file that that it does as well. Getting there. Simon Hauck: Good job. Uh, as for my things that I've been working on, uh, just updated all of the graphics for the Toby store, uh, Chrome store for the regular and then the mini Toby mini. Um, so that's out. Put the description out and then or at least the Toby mini ones are they're submitted, right, Jad? They're just pending review. Jad Haidar: The uh Toby ones are submitted. They're reviewed and they're live. The Toby mini ones I I have to get that done. I'll get that done after the Simon Hauck: No worries. Perfect. Jad Haidar: call. Simon Hauck: Um, so that uh I spent the morning doing a lot of tickets that piled up over the weekend.

00:41:43

Simon Hauck: Um, and then also using that as an opportunity to get people to review or set leave reviews in the Chrome store. I really wanted to kind of I just want to get that review number up. just a long-term goal. Let's get it to like a 4.5, maybe a six. That would be cool. Um, and then as well as I've been working on the day zero redesign. Uh, have most of it outlined. I just need to convert it into a visual for Jad to see. Uh, which shouldn't be too much effort. It's mostly just getting the copy and then uh, yeah, it should be pretty straightforward. So, it shouldn't take too long. Uh, but I'll pause on that and then probably pivot to the website stuff that you've asked for today, Sam. Uh, Jet and I can work on that. Yeah, that's where I'm at. Ben Boudreau: Oh yeah, also um a reminder I'm off Wednesday. So I'm here today. So need let me know today

00:43:48

JP Kalambay: Um, Ben Boudreau: hopefully. JP Kalambay: in terms of reminders, not next week, but the the week after I booked off as well. So, I they're I mean, timewise, we'll see how the meeting go this week in terms of say so set up. Um, I'm just kind of doing like a little bit of a stationish so I can still have my computer if need be. But, um, my roommate just has a cabin um, at uh, like Watcom, so we're just going to spend the week over there. But I'll bring my computer just in case, but it is booked off. Sam Gharegozlou: Did you put it in Bamboo JP? I just went in there. I didn't see it. JP Kalambay: I It should be in there. I saw an email like a couple weeks ago that you approved it, Sam Gharegozlou: Oh, JP Kalambay: but I can double check. Sam Gharegozlou: okay. I already did. No, if I already did it then. Yeah. JP Kalambay: Yeah. Jad Haidar: To that note as well, uh there's a 4 day holiday at the end of the month here in uh in Dubai as well. I'll I'll submit that over time as well. Sam Gharegozlou: When is there not a holiday there, Chad? Jad Haidar: I mean, Sam Gharegozlou: It's Jad Haidar: to be fair, we only get two big ones like in here. We don't really get the Christmas but Sam Gharegozlou: I'm messing with you. I know. Jad Haidar: but yeah, it's it's like the after Ramadan. Those are like the the the main biggest ones. But and they came this year they came uh after each other pretty Sam Gharegozlou: Yeah, but you guys may be lining up more and more with the western holidays and stuff, Jad Haidar: quickly. Sam Gharegozlou: especially their pivot away from, you know, the the Islamic calendar and all that stuff. Jad Haidar: Yeah. Yeah. Yeah. That that would be that would make things simpler for sure. Sam Gharegozlou: Yeah. Jad Haidar: Yeah. Sam Gharegozlou: Okay. Jad Haidar: All right. Sam Gharegozlou: Thank you everyone. Jad Haidar: Crazy guys. JP Kalambay: Yes. Sam Gharegozlou: Bye. Simon Hauck: All

Transcription ended after 00:46:01

This editable transcript was computer generated and might contain errors. People can also change the text after it was created.