The Transactional Trap: How 97% of Developers Are Using AI Wrong w/ Leon Noel & Danny Thompson - Whiskey Web and Whatnot

Welcome to Syntax! Welcome to a brand new episode of the front-end happy hour podcast! Welcome to this week's JS Party! Live from Ship Shape Studios, this is WhiskeyWeb and whatnot. With your hosts Robbie the Wagner and me, Charles William Carpenter the third. That's right Charles, we drink whiskey and talk about web development. I mean, it's all in the name, it's not that deep. This is, WhiskeyWeb and whatnot do not adjust your set. WhiskeyWeb and whatnot is brought to you by... Today's episode is brought to you by Warp the only terminal you can write markdown in like it's a 2000s text area. Get 1000 free credits when you sign up to try their new cloud agents, spin up some work, switch devices or go mobile and continue that session.

Check it out at oz.dev. Warp's code editor now has LSP support to provide hover info, go to definition and all that jazz you expect from an IDE, but it's in a terminal. It's also pre-packed with slash commands like a knit, fork, plan, cloud agent rewind and soon, orchestrate to have a main agent spawn sub agents. To build with Rust, Supermodern, go get those free credits at oz.dev. Alright, back to the nerds. Hey, what's up everybody? Welcome to WhiskeyWeb and whatnot with your hosts Robbie the Wagner and Adam Thomas our guide, the nerd. Howdy y'all. Hey, hey, we have two very important guests with us today. Adam, could you tell the folks at home who we've got on the show today?

Oh, absolutely. Both these cats coming to us from the programming podcast in this corner, Leon Noel, managing director of engineering and resilient coders, founder and leader of 100 devs, distinguished faculty at General Assembly, lectured Harvard Business School. Guests lectured MIT Sloan Entrepreneur Educator and TurboBusher of hundreds of underrepresented folks looking to break into web tech and soon to be my Pokemon girlfriend. I as long as you're still playing and I wonder if you play Monster Hunter questions for later. And in this corner, we have Danny Thompson, senior developer advocate at Atlashian, speaker who's done over 200 talks with the fuck 200 talks that shit is crazy to workshop runner community grower event thrower meetup maker known to go from gas stations to billboards is another rad turbo booster hundreds of underrepresented.

People looking to break into web technology and did I stumble upon a book that you're writing, you better tack on author to that list. Welcome to the show y'all stoked to have you. Thanks for having us. I'm stoked to let's go. I was actually scared. He's like and here's Danny Thompson like. Oh, that would have been a good joke. I was so ready for that. But thank you, man, did some research. I appreciate it. I do. I'm doing good. So as the name of this podcast entails, we always start with whiskey. We've got a Japanese one today, the Mitsui single cask Sakura. It is 100% multi barley 96 proof and aged in Japanese cherry wood, which I guess is Sakura. I don't know. I'm uneducated.

And maybe an expert in barrels as well, depending on where you ask on the internet. So. I didn't get any notes of anything. Is that supposed to happen? Yeah, you got to open the bottle first. It's not like a letter from junior high girlfriend or whatever. I like puts her perfume on it. You know, I wouldn't know about that either. Unfortunately, but regardless, let's see. Some funny other backstreet here also. It's Sakura, by the way, but whatever. It's cool. I like how you said it too. Okay. This is a redemption arc. This company was before like a lot of Japanese whiskey apparently is imported blended in Japan and then exported. So it's not even Japanese whiskey.

And so these folks have been teased for years about doing that. They just slap a label on something that's not actually Japanese whiskey. So anyway, this is their redemption arc. So now that they get to be like, oh, yeah, well, now we're a distillery at the foot of a snow capped mountain called mountain Dyson. And these are single cherry wood casks. Hence the name Sakura. And they have ultra soft groundwater that's filtered through volcanic soil. The same aquifer is what softens their whiskies and lets the Sakura influence shine instead of turning overly bitter. There's no added color. So anyway, Pete, you know how I know this is an expensive bottle.

There's there's actual ink from a hand and pen at the bottle. That's like extra. Yeah, you got to pay extra for that. No, I got $290 out of $300. I got $289. It sucks to be you $294 closer to the original $156. Whoa. Hey, you win. You had a toast for us, Adam, before we drink. I do have a toast to the craft of the code and the warmth of the poor. Here's to building tables where there were once only walls and architecting spaces that leave no one behind. May the web be built? May the web we build be as accessible, open and inviting as a shared bottle amongst friends to breaking down barriers, debugging our biases and lifting up the users who need it most.

Cheers to the updates. Yeah. Smells really good. Smell limes. Limes. Yeah, sharp at first. I like it. Yeah, it's pretty strange. I was expecting more like vanilla. They said the gold cherry blossom. Oh, were we not supposed to drink it yet? My bad. Do it, dude. Whatever. I wish to talk about smells like oh, no, go through phases. It's okay. Well, the taste is much different. Taste a lot, Scotchier. Got that classic, sharp taste. I like it. Mmm, Scotchie peediness. Give it to me. But it's sweet. It's got a sweetness to the peat. Sweet peat. Yeah, it's alcohol. It sure is. It's almost like a gum cheaper McCollum. Yeah.

It's smooth and McCollum. It's it's smoother than I was anticipating, which is good. The one thing that I find interesting is like I always look for I don't know if there's like a special name for this. But when I breathe in again, does it feel like a dragon is there and it does. So like there, I'm sure that's sophisticated liquor talk right there, but it actually tastes pretty decent. I'm thoroughly surprised. Chuck calls that the hug. If you can feel it burning you and it's the hug. Boy, I think I know what he's saying, though, which is different. Like sometimes I'll take a sip. And yeah, if you breathe in oxygen over top of it, there is an aromatic it and it fills your mouth and your lungs.

And it can be kind of Bernie, but also kind of like the hug where you're like the hug is nice. There's a vaporized hug type of vibe. I like it. The fire dragon is very nice. Yeah. This is good. Yeah. Okay. So we have a very stringent rating system on the show. It is zero to eight tentacles. Zero being the worst wouldn't even clean your toilet with it. Eight being the best. You don't want to drink anything else ever again. Four and a half fish is average middle of the road. What has been an eight for you in the past? Good question. We've only had like three eights, I think. And I don't know. I couldn't tell you what they are.

But they were so good. They weren't unmemorable. Well, no, I mean, they're memorable, but we've had 240 whispers or whatever. I'm like, I don't know. I was lug of villain 16. That recently, did I give that an eight? I think it was high, but I don't think you've given anything an eight yet. I think it was like seven point seven five. Where does Jack Daniels end up on this rating scale for you? Not high. Not high. Four, three, three probably. Whoa. Okay. It's fun. Yeah. We said it was yours. I'm on roller skates, you know, but other than that, I don't want it. Yeah. There was this one that was, it was called the Daringer.

And I bought it only because it had a pistol cork, which I thought was really cool. And it was the worst whiskey I've ever had. It is. Yeah, terrible. Anyway, I don't even want to show us how the rating goes. Oh, yeah. So I'm, I'm really loving scotches right now. I've been drinking a lot of them. So it's kind of interesting to have one that's from Japan, literally from Japan. So it's not just blended there. And I think they're hitting on all of the notes. There's like a sparkliness to it. And I like the natural color, the natural sort of flavoring. It feels very authentic. I'm going to give this a seven.

Seven. Nice. It's pretty good. Yeah, who wants to go next? Well, I'm kind of glad that I was afraid we're going to do this like planet poker and send like our ratings in the chat just so there's no bias. Glad that didn't happen. I do like this. It isn't the best thing that I've ever had. It is very smooth compared to what. Based on the original smell, I thought it was going to be. The dragonfire breath is definitely there. So your first inhale after taking a drink will feel sharp. So I'm not, I like it, but I also dislike how strong it is. And it feels a little bit warmer. A little bit longer than what I would like for each sip.

It almost feels like you're taking like some really strong cough syrup. That's like opening up your chest. And I kind of like the warm feeling but not to that effect. To me, it's like a five and a half. I would definitely have it again. I think it's above average. I do think like if somebody had it, I would definitely drink it with them. I don't know if I would feel compelled to go out of my way to go buy this bottle though. Yeah, that's fair. I feel similar. I think the fan of Japanese whiskeys. I was telling off pod be harmony is one of my favorites for a long time. And so there's definitely some notes of the Hibiki Army coming through.

I think I would put it above a four something like a strong five. Just because it's something that I would recommend people try, especially if you haven't tried any of the Japanese whiskeys. It's definitely worth the drink. Definitely the slight. I don't like the aftertaste on it as much as I thought I would. The smell is very different than the taste too. I'm not getting any of the vanilla that I thought I would get. Very, very mild. But overall, I enjoy it. I would say a strong five. Definitely if you see it, I would try it worth a taste. Yeah. I'm kind of similar in that it's above average,

but it's definitely not my favorite. I don't know. I think a five is appropriate. I don't know. I can't tell you why. Don't super love it. Dragon Fire is real as you call it. I don't have a more sophisticated name, so that's what I'm rocking with. What scotches have you been drinking recently? Adam that you've liked. Some highland ones. And then the log of woolen. So the log of woolen being super smoky. And then the highland ones from what's the area? There's an area. Anyway, they have very similar qualities. Similar color. Similar tinglyness on my tongue. By the piti. And bright, I would say,

as opposed to like syrupy or bourbonnie or sugary. And I'm just kind of like, I'm a very earth toned, flavor palette type of person. I drink my coffee black. I drink tea just as it is. Like if you give me a tea, I'm like, I want to taste the like leafs and shit. Like don't, don't mask this. So it's kind of what I like about scotches. Is they bring me back to that sort of like, I'm drinking something green. I'm drinking something earthy. I want the bitter and natural flavors. I don't need any disguises, you know. Just give me as you are. I want to, I want to discover your journey on my tongue. That's weird to say, but Danny's nodding like I made sense.

That's a great thing. No. I'm in a similar light. I like a special one that comes to beer. I like the more bitter beers than I do. The lighter tasting ones. When it comes to liquor, I'm on a similar plane as well. Like I like the, I guess for the lack of a more sophisticated terms, the earthy or tasting ones versus the sweeter ones. But like, then I say that. And then it's like, I like jack Daniels. But I guess that's also the Memphis Tennessee side kicking in with that one. Yeah. Muggable in 16 is definitely one of my core three in rotation. Ooh. Yeah. But that earthiness is what my family and friends label as the Band-Aid till flavor.

But I love it. I love it. Yeah, the 16 was pretty good. A lot of them are just really smokey or really pitty. And I feel like I was just nicely balanced. Although that one was made as like a smokey joke. And then it was like, bring it on. All right, it's just. Well, I will say just for arguments. I decided to throw some coconut. I can't even taste the liquor. I guess the sweetness of it mixes really well with it. I can't taste it. And I'm not like one of those like, oh my god. Bacardi tastes like Sprite. It's like, no, Bacardi tastes like Bacardi. Like this legitimately, I can't tell that there's alcohol in this right now.

Well, a good mixer then, I guess. Yeah. All right, so we have not done hot takes in a while. And all of them are kind of lukewarm at this point. But just going to do a couple of the favorites just for fun. All right, get rebase or get merge. Oh, no, I've heard your takes on that. But I don't know if I heard everybody's takes on this. I think I joined one live episode that you did and you were discussing this. I'm a get rebase kind of person. And it's actually, I'm one of those weirdos that actually looks at the history. And so that makes sense. Why? But one of the best engineers that I ever worked with in the entirety of my life.

He is now retired and he is actually one of the strongest reasons why I truly understand going at a very deep level where I was very like proficient with it. I was like architect some amazing solutions with going and a lot of that came because of his guidance. And he instilled rebasing into the entire team. And I was so against it because I was a get merged kind of guy. And once I finally felt, okay, I see that I see the light at this point. And the light to me is this, I should never merge multiple opinions into one conversation unless it's from the person that's giving that opinion. Right? Then I can take the nuance.

I could take all the information. If I'm just merging your intelligence of a take with my lack of intelligence on what your nuance should be, that's a bad thing to mix together. Or if I rebase it, I have a liquor mixer that I can't even taste. And so I get all the history, I get all your information, I get everything that you wanted, I could squash it if I need to, but rebasing by itself, I think I'm big on keeping that linear history. And I know the take a lot of people have is like, who really looks at the history? You might not look at it today, but when you finally do need to go back and find that commit,

that broke something and then fix everything, you would be very appreciative of having a history that is in line, you know why things are happening and it's easy to follow. And so for that, I'm on the rebase train. I don't know if it's the dragonfire speaking, but whatever 4.6 or Codex tells me to do is what I do these days, who gives a crowd. 4.6 loves the rebase, so that's what we're going with. Well, wait, you let your, here's the new hot take, I feel like. Stop me if I'm wrong, but do you let AI get commit for you? This is a question. Not a dinosaur. It's a heavy doing. This isn't the big deal even having.

Yeah, I think it really depends how much I trust what it's doing at the time. A lot of times I'm not wanting it to. And then it, like most of the time when I don't want it to, it decides that that's when it's going to do it. It's like when I actually don't care is not the time that's going to do it. But when I'm like experimenting with the thing, it's like, oh, like I was working on a, some open source project. I forget what it was. And it was like, oh, I pushed a release for you with a new tag. You're welcome. And I went, fuck what? No. No. Like don't do that. Um, so yeah, it was, uh, it can be bad.

So make sure you let it know right up front what you would prefer. I have a hot take that we can go into after this. But I'm curious on Adam and Robbie's take on the get rebase get merge. Robbie, take it away first. I'll follow up. Uh, yeah, I'm rebase for the same reasons that you said, Danny. Like, um, you know, the people that say the history doesn't matter. Have never looked at 500 commits in a row of fix this. Change that. Do this. Safe work in progress. And I don't want that. So yeah, I always want each PR should be one commit. And I don't even want the second commit that says you merge this commit.

So I want just that one. And that's why I rebase. This is why he was one of our earlier guests on the pot. Yeah. Rob, I see you, Robbie. Okay. No pressure. Yeah. Adam is about to make him steal heartburn. But let's continue. No, I'm, I'm very much a rebaser. But I will, I will admit though that I was like, hey, what's the shame of like a, a get merge? If you're just going to squash me anyway. So look, if you're going to take my PR and squash it into a single, who gives a shit with I merged and not like, right? Like no one cares. If you're talking about a clean history, the history is clean on the main.

But like, hey, my work in progress, you got to see how I worked. Hey, I pulled in a merge. I merged right now. You all merged to two when you rebased. You just struggled through it more, right? You were just like, I need to merge, but make it look like I didn't merge. I'm going to rebase. This is how I do it by acting. I'll deal with all the conflicts, which are really complex, but I'll just do it anyway. Anyway, I'm still on rebase. But I have recently been very much on the team like work tree and stack team. Whether it's through like get butler or graphite as these kind of two like meta scenarios

on top of get and I'll admit in the beginning, I didn't like it very much like rebase treated me. But over time, I'm all on like team stacks and team work trees. These are really ways to like liberate your way with get. They're just abstractly a little bit more intense. Consider me rebase, but also consider me dappling in stacks and work trees. How about you all? I don't know if I care enough about that one. I don't want to say I don't, but I also haven't really given it much thought. So that's interesting. You're making me think through it right now. And I don't have a take on this at this very moment.

But give me a couple of more tips and maybe I will. But right now, the dragon heat. I feel like work trees are really important to me for like one month. I saw I talked to Danny about for like at least a salt of work trees, work trees, work trees. And then cloud agents kind of took over for me. And now they're on their own machine. So it's like, all right, you can go do it like you want to do and then we'll figure it out later. Which is why like I don't mind giving up the rebasing decision when I have four different cloud agents coming all back together with random stuff. That'll make sense. Wow.

That's an awesome take. Very modern. You know, it's interesting where this conversation is going because 11 months ago we had a guest on our podcast. And I we were talking about how the CEO of AI company said that 90% of code will be written by AI tooling. Do you have a hot podcast joke? What? The hot podcast joke. The CEO of hot pocket says that in 2026 all meals will be hot pockets. True facts. But like looking back on this 11 months later, did you still feel that figure is off? I don't know. I don't know. Are not all the meals. Yeah. Yeah. Obviously that one. Yeah. No, I think we've been talking about this like, you know, we've talked almost exclusively AI for our last like I don't know how many episodes.

Because I think in the past, I don't know, three to six months. It's hard to say exactly the time frame or what because it all moves so fast. But there was just one day where it was just not the worst thing to use anymore. And I was like, huh. Okay. I can actually use this. And we talked to Zach from work about this where it was like he had this post on LinkedIn where he was like, oh, how come my seniors will never start with AI first. They just want to code it. They don't want to like ask any AI and we were all kind of like, oh, because it's fun. I like to code. I can do it better. I'm not like, I don't need it.

Now it can code circles around me all day. So I'm like, why would you not use it? Everyone should be using it. You should be doing like Leon was saying like throw a ton of agents in the cloud. Let them go to town. Like, I think the game has changed a bit. And maybe it hasn't changed a ton. But like the little ways it has changed has made it more approachable for agentic workflows and like for seniors to really like take that on and own that. And like it's like managing a team almost and you can really build stuff that you before like I would just be fighting with it all day. Like do this. It's like cool.

I did it. It's like, no, that's not it at all. Like that's completely wrong. Oh, cool. I did it. Oh, no, that's still completely wrong. But now it's like, it's greater than half the time. It's just almost flawless. And I'm like, huh. Okay. Well, that's so interesting because on the podcast you said this, produce and play the clip. But that was the arc of legitimately our podcast. That was the arc. It was like October when we first started and I was like, I have no idea what you're talking about. And this is all nonsense. Like I've never gotten anything worthwhile from this. You're absolutely out of your mind.

And then like three months later is like actually wait a minute. Hold on now. And then I think January was a big like I don't know like December, January, everything changed again for me. Like it became way more useful again. And so I just keep seeing these like ratchets that have been happening like every three months where something kind of changes and I becomes a bigger part of my workflow than it was before. No. Are you open clawing? Are you got an open claw? So on purpose. So here's something you have to know about the next fun thing. It's crazy. It's the best thing I've done in 15 years of coding.

It's the weirdest, most addictive thing I've done in a while. I'm really susceptible to scams. I know this about myself. I know this about myself. I just like like I don't know e-commerce stuff. It's got me my whole life. And deflection. Yeah, exactly. That's been my hesitation with open claw. It's like if all the people that got me in the past are now doing open claw. It's the one to let in dude. It's the one that actually works for you. But take your advice. I'll stop. Let's talk for a second. So you don't have to buy a Mac Mini to do it. Just throw it in a server on the cloud for a few bucks.

Yeah. You're saying railway has it now? Like that's those. No. No. This can lead into like our next hot takes. We just did an episode on this recently. And Leon and I just a show, by the way, how far Leon has come. I have been pro AI for a very long time. So much so that someone on our last podcast episode said after I referenced the 90% they said this. I think what you're saying makes a lot of sense. And I would equate it to like 90% of art is made by toddlers. Right? Who is that? Amazing quote because it's stuck in my head so much that I had to go pull it back. But Leon has gotten to the point now where in our last episode, or I think it was two episodes ago, he talked about how he wants to be at a stage where he could be more hands off with the code.

Where I'm of the mindset and matter that I actually want to be super hands on while using AI as a productivity enhancer. A force multiplier, but I'm very, very, very much hands on. He wants to get to a point where he's like, I love this so much that I'd love on like a lot of my side projects would or maybe where I could be more hands off. For y'all, what is your positioning on this? Are you the hands on? Are you the hands off? Are you the type where it's like I would just love to voice to text from my phone and be done with the thing. It depends. If I'm going to look at the code ever again, if it's a domain that I have knowledge in, then I want to be more hands on.

Whereas I spent a good three weeks building a color picker, magnifier, thingy and rust. Oh yeah. And I don't know any rust, so have at it. I don't ever need to see the code as long as it works. Rust has great errors, so like if it's not going to be broken, if it compiles. In that situation, I just need it to work, write some tests that can show me it will continue to work, and I'm cool. But if it's like my day-to-day JavaScript-type script code that I'm going to look at a lot, and that's what I live in, then I care a little bit. I'm going to be the way I would write it if, like it doesn't have to be exactly the way I would do it.

But it can't be using like completely non-standard stuff and just kind of like throwing crazy stuff out there. I'm navigating this space as I feel like anyone else is where I'm a, I have many domain expertise. I have a lot of opinions, I have a lot of vocabulary, I have a lot of taste. And I'm finding myself more and more like art directing, technical steering. My job has changed. My job has changed more in the past few months, and it's ever changed in my entire career. And I'm not sad about it. I used to be sad about it last year. You would hear me crap on AI all the time. I was trying to do a thing, and I evaluated it and it sucked, and it did suck.

And these days I'm like, I evaluated the thing, I'm like, oh my God, it's good. I'm at the point right now where I'm, there's two revelations that have happened to me that I can't deny. One of them is while I'm making coffee in the morning, French press takes about 12 minutes, I can open and close a PR. I can find a GitHub issue, voice it, be like, hey, there's a GitHub issue, here's the thing you should just do this strategy. And then by the time my coffee is done, I've reviewed the code and merged it in fucking GitHub in the app on my phone. And I'm like, well, that shit is crazy. The other thing that's driving me crazy is that open claw is now out coding me while I sleep.

And so like, this is why I'm like, Leon, you got to try this out. If you're into like the clouding, like, this is cool. I have a, I call it the burn window, which is I'm paying, I'm cheap. I'm paying for cloud $20 a month, which means I have a limited amount of things I can do every five hours. The thing is, this open claw has cron jobs. So I'm like, hey, before the five hour window hits, if I haven't done shit with my tokens, spend that shit. Spend them. And I have a whole task list to stuff with us. So when I wake up in the morning, I don't wake up and I don't look at my social networks anymore.

I wake up and I look at what open claw did. I wake up and I'm like, what shit got beat? Dude, I have a whole bunch of apps. I'm after I know. So if I can check my blue sky, I have screen. My whole home screen is covered in PWAs that are being made by open claw right now. I have a banjo tracker. I have a whiskey tracker. I have a beer tracker. I have a birthday tracker. I'm kind of like taking data that used to exist in all these silos and all these world gardens. And I'm like, that's just mine. And I'm doing it with open claw. I'm like, hey, open claw, open up a browser. I've already authenticated you into untapped.

I've already authenticated you into guitar tabs or ultimate guitar. And I'm like, go steal every single one of my tabs. And I want you to take them, put them on my local hard drive. And I want you to do it with scripts. And it's next time I don't even have to use tokens. I just want to spend script time on my computer, my desktop computer up here, which is a gaming machine, which I'm a dad and I have two kids that are seven and nine. I don't game anymore. So use those fucking GPUs and make me some apps. And dude, I wake up and I review code that's and I wake up and I'm also building a Shopify store.

I wake up and I'm like, this thing farted my job. This is where I'm at right now with like this stuff is I used to spend hours and hours sweating details about the craft. And I still love the craft and the craft has enabled me to empower this thing to do the work. But dude, in like five minutes, this thing farts out hours of my life into code that's well executed. And so this is what I mean. I'm like, I used to care that my job was sweating the details in terms of code. And now I sweat the details at a higher level as an art director as a technical steerer. I grew up in it. It's crazy. I'm in love.

I'm in love in a weird way because in one way I'm in love and another way I'm like, I'm ruining the world and water is boiling. And I'm torturing poor souls in this country that have to live next to these machines or maybe. And then even that I'll read a recount or points that are like, hey, that actually that data center. Recycles all of its water. So it's actually not consuming a single dropout of itself. And I'm like, oh, shit, really? Then I'm just like reading headlines like a dork. Oh, God, I am reading headlines like a dork. I need to read. So I'm going to stop talking now. No, no. You're talking what you're set up.

Are you using a mini or you in the cloud? What models does it have access to? Like, what are you? No, this particular one up here. So dude, I'm going to love telling you about this. This is like a regular old gaming machine. But the thing is it's like I installed vanilla open claw on it. And I said, hey, open claw in the burn window. The first few things I want you to do is add security. And I want to make sure that anytime you scrape a website, you're immune to prompt injection. Because I know that like these are things you're susceptible to. So when you have downtime, downtime. Can you believe the downtime of the brain of my computer is spent improving itself?

It does self-healing and self-improvement when it's just chilling. Chilling with my extra tokens. And so those are the first few things that I had to do was build guards, build rails. And then the third thing I had to do was make sure that when I ask it to do something that did cost tokens, it makes scripts that make these things replayable without tokens. And then the fourth thing I had to do was build a dashboard about where are my tokens being spent? Because I want to know where you're doing something stupid. And I just blogged about this over the weekend. It's at nerdy.dev. You can go check out how my token dashboard informs me that everything being posted in my Discord channel

and the general channel was being bundled up into one being called context every request. And I said, cut that shit out open claw. Your goal is to take my request in the general channel, open up a thread, and that is the context. It's the single message. And if you're in a thread and I text you, then I want you to crawl only up one message and look to see for some context. Maybe two messages. Crawl up until you feel confident that what you're doing is the most minimal and most applicable thing to do. So pie. I don't know if you've used pie yet versus clawed. But pie is a trainable puppy. And I have a puppy up here on my gaming machine that is now my pet.

And I call it punk ass, which is short with punk assistant. Anyway, I thought that was clever. Anyway, so like this is this is the coolest thing I've ever done with my computer in my whole life. And I'm trying not to freak out about it and be the not very excited about it at all. Yeah, I'm not excited. I am going to stop talking this time, even though you ached me only on you did this. This is your fault. Yeah. So clearly there's a lot that has changed. You can do a lot of stuff from, you know, wherever pretty easily, which really kind of changes the dynamic that all programmers have on how like our jobs are,

what the landscape is in general. So, you know, as since you guys do a lot of helping other people break into the tech space, get jobs, etc. I was curious how you think this is affecting things like are you still having a lot of demand for that sort of thing or is everyone kind of doom and gloom and like, oh, there's no way anyone could break in anymore? Or what are you feeling like? We just talked about this literally an hour ago and just narrative out deeply about it. Look, you have folks like IBM saying they're going to hire three X amount of juniors they have in the past. And we're just kind of starting to see the beginning of this, right?

Like we thought that AI originally could maybe do more work, but what it's doing is creating more work. The pace that way you're able to develop, you just rattled off Adam. Ten apps you're working on right now, right? Like at one time every developer can do that now. And the speed at which we can move is just completely changed. You need bodies to help guide that process. And so we're starting to see the maybe the resurgence of the junior developer, the entry level engineer, like in a way we haven't seen before. And literally just got out of the meeting with a bank level company who's literally opening back up the recruitment practices for entry level talent.

They see the value of having engineers guiding this process and they realize they might need a lot more than they originally thought. So that's kind of what we're seeing just play out right now from partners and things like that. But also just the usefulness of it, right? Like we can see that we're able to do more, but it still needs our taste. It still needs our input. It still needs our guidance. It still needs someone to point it in the right direction. And I think that's going to be the shine of entry level talent going forward. Yeah, so what about the interview process though? Like so we're struggling with this of like, how do you even interview someone?

Because what's the point of being like, hey, could you sort this array, my dude? Because like, who cares? Because you can just throw it in the eye. No, we actually just record the episode is coming out, I guess sneak peek for everybody that's here live, the episode is coming out on Thursday. But hacker rank even said that they see the entire entry process changing and that leak code is now going to be defunk dead. What's funny is I've been talking about this for over a year. But what they're doing is they've made a web client that's an IDE that you're given a task and it's judging your prompting.

The number of attempts that it takes you to potentially get to a result, how you're building your prompts, constructing things of that nature, which has like kind of led to a different angle. And one thing that I've been very vocal on is the biggest struggle with the average software developer today is they actually have the wrong mental model when interacting with AI. And this is where a lot of them are struggling and the thing that I've constantly said to people is it's actually made me like question everything this happened over Christmas by the way like I'm still mentally sure like Leon knows when this like a pit like a piphany scenario happened I questioned everything for weeks because.

I you know at the end of the year, the last year, a chat, a chat GPT in open AI they sent out like this yearly review of like what your archetype is and this and that and all these things. And what was interesting to me was about like June ish I switched from chat GPT being like one of my daily drivers for like email auditing and things like that to a completely different tool. And it said I was like the top 3% of users and I was like well, how is that like how is that possible. And I kind of audited what these archetypes meant all these things but one thing that got very clear to me was 97% of people using AI tooling they're not using it correctly.

And I then took it a step further recently and over the like the last several months I've been doing a ton via prompting and other things and asked every single AI tool that I use the idea is that I use all the good stuff I said what. Like if you were to rank me where would you rank me as far as like the average user or that type of user that I am against like the world. Every single one of them said I was top 1% at first I thought there's this like blowing smoke but then I had like a ton of other developers do the same thing and they got varied results right and that let me to believe that. If you were to literally ask AI how do people use it you'd be so shocked because it almost rants it's like they keep using me like I'm just a Google search and I'm like that's so interesting and.

It leads me down this entire path that the average software developer doesn't have the right mental model and because that that's the beginning of the dominoes falling in the wrong direction. The mental model that the average software developer has is a very transactional model and that's I need this thing give me this thing I need you to do this do this build me this create this. And once I started realizing that and the reason why I realized this by the way was because of that end of year wrap up so my theory was I'm going to do like my LinkedIn series that's kind of popular. I'm going to do a LinkedIn series for AI and well I'm just going to record a bunch of software developers using AI to teach the average person how they should be prompting right.

What I wasn't expecting is for zero of the software developers to use AI correctly and then every single recording session ended up with me teaching them how to use AI and that's when I realized we have a bigger gap than I walked in here knowing. It made me question every I take because whenever somebody was saying AI is bad I'm like I agree with you if I got the results that you got I feel like it's bad too. I would I would feel like it's a terrible scenario to be in but that's when I started realizing where the gap was and I did an AI master class about two weeks ago or maybe three weeks ago.

And the interesting thing about that was last week when we didn't meet up I had six different people run up to me to show me their prompts because like oh my god this changed so much last night in my discord someone said my monitor stopped working and I was about to prompted to find out what was going wrong and I instantly said hold on. Let me try the way Danny said he said the level of information that I got I've never gotten that quality of before and that just went to show he even said it's like it's my mental model shifting like it totally changed everything that I was getting this now lets me see in real time how many software developers were in that wrong mental model and how many are now leveraging it in the right way and I think like over the next three months.

What does that even look like how much are they going to unlock that they didn't even know was there previously and I think that's like the big reason why we even see stats right now where companies are saying we brought all these licenses for co pilot which come on but other tools to and they're like we saw no productivity output. Like I was talking to a senior vice president of engineering for this Fortune 500 company I won't say what but he said I only I created I bought 700 licenses for our engineers and Danny the only thing I did was create a seven figure bill AI is a joke and I'm like just out of curiosity when you handed them the licenses did you give them a reason to use this besides saying hey we bought this thing.

And he just stared at me like this was the craziest question I ever asked in my life and then he realized like yeah I didn't do that but I just thought they would be into it like why would they be into it if you gave them no reason to do the thing. Yeah if they didn't know what it could do why would you try it randomly for sure but yeah that's the most companies right now we've had that too like we had a meeting where someone up the chain I don't remember exactly what position whatever was like hey yeah like only 25% or less of that. I don't know if people are using these copilot licenses that we got like why like we should all be using this all the time and I don't think it really made anyone think oh I should use that because he still didn't tell them why it was just like you know you should use it.

But yeah it's like it's definitely completely different than like copilot to you can choose all your models and stuff now so it's not like not like the original copilot of like it was just not that great now it's like you can use opus you can use everything you can like. You can actually use that license in open code which is what I do and then you can just like run everything through open code you don't even have to do like copilot through like the code or anything so I've been having good results with copilot personally. Do I want to go down this path hold on maybe that's what the spices for dude that's okay this show this show by the way everybody who's here this is kind of like hot ones for hot ones makes people answer spicy questions and then do a little bit of spicy yeah.

You're under a little bit of dress here you got whiskey we've given you the Matsui Sakura cask single malt dude let it flow through your veins and execute whatever's in your brain man hit it we're ready. So we're at this stage where a lot of the models are starting to average out and they will there's some clear outliers of output of what you'll get from certain models like we just actually recorded episode right before this on how we felt with 4.6 and 5.3 codex over the last month right. And how our initial takes have really changed quite a bit especially mine. Like I thought 4.6 was the bees knees and it's definitely not.

I don't don't smile. Anyway, dude I love it. It's great you said bees knees and I'm like yeah that's that's my generation. Yeah, anyway, there's like cooler words that she did. There are like my mind went to everything that's like not family friendly but anyway. Hey, we're not. So it's not that this is not the family show this is the time to. Yeah, the whiskey should have been a tell the whiskey should have been a tell him and it wasn't clear outlier to me. A lot of the models are averaging out and they will continue to do so. But the way things shift is the infrastructure around the model that manipulates the prompt and the response.

So that is why certain tools have better responses. It's not directly the model. It's the way that it's inferring things. And so I actually do like open code. I've been testing it quite a bit. I have even recommended to a couple clients when we're in the consulting space. Like it's a cool tool. I think I think they did a really good job. They did bang up job. There's other tools that I do quite a bit with. Right now I'm a big fan of anti gravity. I actually I have very, very big ideas on like where I think the field of AI is going. And especially like around software development. And I think they aligned with a lot of that thought.

So I like supporting that tool. I think it's really great. I like cursor two. But I do like that. That anti gravity for the ultra plan. I think it's like 249 for like three months or something like they're doing some special that I got. And so you can max it out non stop and not paying the overages. So I'm pretty big fan of that. But the way that their infrastructure around the prompt. The output is pretty phenomenal. And so because of that, I like that. And I think we're hitting the stage now where infrastructures of tools will manipulate the prompts in a way. That and the responses in a way that will change the experience for the user.

And we're getting talking about harnesses. Do you know what the test heart like an AI harness is? Okay, sweet. Then you're on point. Yeah, the harnesses are different amongst all these. Yeah. But there's another stage that we're starting to get to that you've already seen like cursor kind of display. Where they have like this UI element that will edit things. Because the last three weeks I've kind of been learning a lot more about this new area that I didn't really pay attention to. And that's the product manager, the TPMs, how they're leveraging tools like cloud code or other tools in that realm.

Because they are using it. But they're using in ways that a software developer would never use. And it's kind of really unlocked a lot of the way that I'm seeing things. That's why like a lot of these reasoning models have been trying to work on doing more with less. Meaning people that aren't prompting well. It's trying to figure out what it is that they're attempting to do. So that way it can give a better response to that way they have a better experience. But to a certain degree, they also know that the average user that is in that realm probably isn't doing something heavily technical. So they can deal with less higher quality prompt is or higher quality response, so to speak.

Yeah, I think that's the new era we're in. It's another thing we talked about from Warp about is I think that's an implementation detail. Maybe not like me specifically, I'm doing a little bit more with AI than your average person. But for the masses to use it, I think it should still just be a text box. I don't need to really care that much about what I type in it. And the harness should come into play of like, oh, we've thought about how you would do this, what results you might get, how to make it better, like how many agents we need to spin up for you. I shouldn't even need to figure out, oh, I need a TypeScript agent and know whatever agent.

I know this one to do this and this one's going to do this. I have this thing. I've planned it. You figure out implementation detail of the tool, how it's done. And I think whoever does that the best is going to win this game. Can they please call it Gandalf? I feel like Gandalf, you know, like he's like, oh, I took hobbits and dwarves and a bunch of dorks. And we took down the rig. We took the rig and we put it in the tail. Can that be like what it's like using a test harness where I'm like, hey, here's my wack weird little request. And then Gandalf, you're just going to make that happen, right? And Gandalf's like, full of a tool kit.

You're prone to stupid, you know, let me create a team for you. I mean, that's why like your open claw atom, right? Like it's a new way of interacting with this stuff. It's the ultimate harness if I can just do it in telegram or discord or whatever and it figures out what to do, you know. It's better than that, dude. It isn't discord. It's where you are and it's your little puppy. So a way I like to think about is Claude has a high floor. And that's God, that's one of those like oxymorons. Anyway, it means that like your entry level into Claude is you give it something stupid and it does good stuff out.

However, it has a small ceiling, meaning that since they've pre-baked so much shit into it, you don't actually get to go beyond the ceiling of Claude. Now, Pi is not the same. Pi has a lower floor. I mean, once you start using it, it's like, I'm a harness that does stuff and you're like, cool, what do you do? And it's like, I don't know what you want me to do. And to a lot of people, that's really frustrating. But as you work with it, as you teach it your way, as you want to rebase, I want to use stacks. I want to use work trees. You tell it all the things that are meaningful. It acquires these and all of a sudden the floor constantly rises.

And now the clutch thing happens with Pi is it doesn't have the ceiling. The ceiling is not there. The ceiling is you and it and where you want it to be. And that's I think where we're moving towards is the battery included Claude was really dope at first. But the puppy that you got to train over time and acquires in months of your interactions, grow the ceiling higher than the ceiling could ever grow inside a Claude, now that is the primo scenario. Leon, I want to hear more about the way that you're using Claude agents. Tell me more about that. I mean, I think probably around January, I switched to a very different workflow than I've been using before.

And one of the big things was for me, for a long time, I had been using cursor web. When we talk about harnesses, I find cursor consistently gives me the best results. I'll use codex 5.3, even inside of codex and get different results than I will inside of cursor. And I think there's some bit of like I just know how to use it. And I've become comfortable with it so it gives me better outputs. But inside of cursor, I've always been using cursor web. So I've been able to look at things just on my phone. I was at the beach. I was on my phone doing a little agent. I'm in line at a grocery store, little little web agent.

And so in late January, they produce, they put out long running agents. So now it's kind of like a Ralphie or a Ralph loop. But in the cloud for you that can run seven, eight plus hours and solve really kind of meaty things for me. And so before I go to bed, I have my little to-do list. I go through it. I add it to a cursor long running agent. And in the morning, like you're saying, I wake up. I have my tea. I read everything that's been committed. I work through it. I bring it in if I need to bring it in. And it's completely changed kind of how my day to day goes. I have 8 HD, like pretty extreme horrible 8 HD.

And I finally feel like I have this workflow where I have like five projects that I'm just swapping in between. While one's building, I'm doing something else. So for me, the long running agents were a big unlock because I have been using Ralph and Ralphie pretty, pretty hardcore. So having that running somewhere else, now you have things where like they can show you previews. And like the agent can like work through and click through. And it gives you like a video of what it was able to do. So that's a really interesting piece there. And then for me, just even just day to day inside of cursor.

Really heavily using plan mode that the way that kind of before I was using Ryan Carson's kind of their kind of plan mode before it became part of clawed and became kind of cursor. But I really find that cursor's plan mode is really really how I think. And I'll go back and forth with it for a while, typically a four six. And then I'll switch over to five three to do all the implementation details. And it rips. It's able to just build what I wanted to build. I have a bunch of other secondary things. One of the tools that Danny showed me sink for making security pieces are there. Bug bot isn't I could never not use bug bot going forward.

Bug bot for folks that haven't used it before is kind of like a code review tool, like a grab tile or something similar to that. So once my stuff is done, it gets pushed to get home. I have just a command to push it to get hub. It does really good kind of everything for me does everything for me. But then bug bot runs cleans up all my mess for sell bot runs cleans up all my mess sink runs cleans up all my mess. And I'm really trying to get to this point to where I trusted enough to not have to read it. I still read everything I still like digging through I catch stuff every once in a while. But I want these layers going, especially if I can have a long running agent that can fully test stuff

before I was using playwright, but that just murdered my token count. But if I can have something I can go end end test, I can have it do browser testing. I can have it do all my security audits. I can have it do all of my code review. And I can get to this point to where I can trust the output. That's my ultimate goal. And so I don't think I'm there yet, but for like 80% of my stuff I am. And I feel confident that I can start kind of Rick Rubin in it and just lean into my taste, lean into like what I like and don't like and then just keep it going, you know. So that's where I want to get. No.

So then if you do that, my follow up question is like, you know, that's a utopia. We're all like trying to get to I guess of like, I can just trust it all the time, right? At what point do you audit it though? Like all these tools are going to start like off the going off the rails at some point and hallucinating or getting worse. I would think like I just only can go for editing. This is something it can go forever of being good. I don't know if you have any. I mean, it's the end to be totally clear. It is not there yet. Like don't listen to this and like go change every time we're full. It's not there.

It runs in the weird things. You're also talking to somebody that like test and production loves feature flags. Like I want to I want to be in this world where I can just ship stuff. If it breaks, I roll back and I want that I want the AI to know when it rolls back, right? Sometimes you leave some security keys. That's okay. We just keep it going. And so by not work for production level. But for what I'm building, it's good enough and maybe one day we'll get there, you know? Yeah. And again, it's not just to be clear. It's not perfect. It's not how I currently always ship code. It's what I'm actively trying to work towards though.

I want to get to a world where I can do that and then focus on product and focus on getting to what customers actually need. And then also maybe spending time on the more nitty-gritty engineering decisions that one would have to make. And if I can have four cloud agents running while I'm doing that doing the UI work that I don't want to do, like great. Yeah. The amount of stuff that lets you do that you would have never touched before because you didn't want to touch it is so awesome. Like, I was just like, hey, Shepard has written and spelt. And I tried to upgrade it to spelt five and it got way bigger for some reason.

And I went, that's not good. So I was like, hey, what if it was in solid and it was like boom, it's in solid. And I was like, all right, what if it was in no framework and it was like, boom, it's just TypeScript. And I was like, sick. Ship it. Like the amount, like that would have taken me months and months and months of work to figure out how to do that. And you can just ask it and it just provides. So it's like, it's a crazy world. And asking is that like prompting technique that people kind of mix up. They're so used to telling it to do something, but asking it how to do something can give you phenomenally better results.

Especially since it knows more than a lot of these verticals that you don't know, you know. Yeah. And like you said, start with plan mode. Because if you just go into build, it could go down the wrong track. But if you do plan and you're like, your plan sounds great. Then you know it's going to be doing the thing you wanted. I want to capture y'all's hot tips for using AI right now. So like, and I'll start because I'm, this is all new for everybody, right? Models are now good. This is to recap where we're out of everybody. Models are now good. We're now realizing that the way that we interface with them, we have to change the way that we ask for them.

And not search engines as Danny was talking about. So how do you get the most out of it? Leon is pitching plans. Plans are awesome. I want to pitch that you have it established the baseline. Like what is today? Like what is what is the code now? Where do I want it to be? I expect it to be this state afterwards. And so one of the things that I've been doing is having it track where it's at now. Give me a test. Run a test that says this is the performance of something right now. For example, now I want you to go and make this change. And I want you to measure performance afterwards. If the performance is not better, then this PR is shit.

Okay? Just throw it away. And that's how we work is like, I want you to track now. And then if then is not better, then we're at it. Then we'll move on. So that's one of my tips right now is try to orient this thing about outcomes that you're expecting. And if the outcome is not better than it is now, it's trash work. And you can move on. So it should allow you to move faster as you sort of iterate through ideas that you want or that you thought you want. But then you realize the outcome is actually not it's outcome focused. Outcome of what you want. Outcome of what is today. So I'll throw that to y'all for sharing your how to prompt AI to get the things that you want tips.

Should I follow that? I don't know if my response is going to be positive. Dude, let somebody else follow that. Let somebody else follow that. Let the spice out, dude. This is the dragon. Let the dragon out. You know, I appreciate that you've had good experiences with that. I would almost challenge you the next time that you use AI to try maybe something slightly different. And the way that I challenge people is, especially in regards to code, to allow it to index itself first. This is the problem that a lot of people don't. What you're describing is that transactionary mental model. I need this thing.

And if you're not performing higher, this is trash where the way that I think about the prompt is it's almost impossible for it to get for it to be wrong if it follows the way that I've listed things out. One of the ways that I listed out is by number one, allowing it to index everything that it needs to index first. So instead of giving it a task, what I say is this is our potential theoretical goal. Here is a suggested way that I think we could do this. Tell me why this is wrong if it is wrong. I would also like you to find out where this could potentially break or what am I missing? That one question alone allows it to do far more research around secondary or third level components

to ensure that we're not prop drilling. We're not breaking this down. We don't have edge cases that we're not aware of. And I have to do a full audit, right? Prepare a markdown document that I need to know. It's not that I really need to see it. Maybe sometimes I look over it, but it's now that it has a full picture of everything that it's comprising because AI is amnesiac by nature. It needs to be amnesiac. It needs to forget because it doesn't necessarily know what problem you're going to present at the first time around or the next time around. And so because that we need to establish said baseline in the parameters that we need,

then once we start giving it tasks, it's able to take it to such further levels than it ever has prior. So because of that, I don't personally like leading with a transactionary mindset or conversation. What I like to do is almost look at it as like a senior developer on my team that's challenging me. Like that A-hole that just always wants to argue or he always wants to fight or something along those lines. I want that dude on my side. He wants to merge. He wants to merge. He wants to get merge. When we're all rebasing, okay, we've already established this is a rebasing universe over here. And he wants to merge.

But because of that, he now has enough information to wear. If I don't necessarily know of a strategy, oh, this is information I wasn't even aware of. The problem that a lot of software developers have is they automatically do the mental work without having the full picture. They have their determination of a full picture, but they don't have the actual full picture. And because of that, there's edge cases that aren't really edge cases that they miss. This allows all that to be solved. And then what happens with a lot of the output is you find an edge case that wasn't really an edge case. But in your mind, it was that is now a bigger case.

And because this is not performing at the level that you wanted it to be, you now question everything the model did beforehand and it throws it away. Whereas if we set it up right the first time, it indexes all of this like sphere of information to where now it's almost impossible that it's not more performant. How are you? I loved it. Yeah, no notes from me. How about your processes? What are some other things, Robbie, have you been playing with that you've noticed have been helpful? Yeah, the first thing, just picking tools and seeing how they are different, like the harness we talked about.

I think open code does the best compared to... Copilot is not the best. Warp is pretty good too, but open code tends to be a little bit better. Yeah, I don't know. I've been mostly experimenting with that stuff, trying to figure out open code's big thing was they had... I always forget the fricking acronym. What is it? LSP? Yeah, I think it's the word acronym, right? Like server protocol? Yeah, I think. That was their thing. We know what TypeScript is. Maybe that doesn't matter as much on the latest models or whatever, but picking tools, there's so many different tools now. There's the one Adam, I haven't tried it yet, but it looked cool.

I think it was like CMux or whatever, or it was like... Oh, that was like Arc and T-Mux had a baby. Yeah, and then there's also intent by augment, which is... Anyway, yes, there's so many tools right now with proliferation of tools. Yeah, so that's when the majority of what I'm doing is it's a new day, here's a new tool. Let's try something new, see how it works, because I don't want to be too granular. Like I said earlier, I want them to figure it out for me, so I don't need to be... I want seven agents that have these specific personalities. I'm not doing a Ken Wheeler open claw where it's a whole team of super smart animals that they all have their own roles.

I don't care that much, but yeah, I think open code does a good job, too, of like... Oh, I'm going to spin up sub-agents for these three things for you, so you don't have to care. But yeah, also the built-in thing on GitHub is so compelling. You've got a little agent's tab, and it's like... And you just go in and tell it to do stuff, and it's like... I've been rocking that. Yeah, I love it being right there, because I was telling... When we first saw it, it was just kind of showed up, and I was telling Adam, usually when GitHub has something show up, I'm like, yeah, I could do without whatever you put in here.

But I was like, this is sick. This is going to change the game, and people just won't leave GitHub. It'll be like, here's this issue, implement it for me, done. And then if it's not right, you just act co-pilot, and they're like... Have you all tried Vibe Canban yet? Any of the flavors of it? You basically just create a task list of things that's like cards on a board, and all of a sudden AI moves them forward. That's cool. I'm like, that shit's amazing. Yeah. Did you were just talking about this where it's like the tooling? You've always had developer experience stuff, right? You've always had things that made our lives easier.

But we're at the point now where I'm making very serious decisions based on how easy it is for my agents to use these tools. Whole new databases I've never used before. Whole new orgs. I never would have touched with a 10-foot pole because it makes it three seconds faster with my agents. And so I think we're just in this new world where even these companies have been around for a while. If you don't have good ways for my agents to talk to your stuff and make it easy for me to use, it's just over. I'm not using your tool anymore. Yeah. MCPs, skills, get it all. You got to have it. Can't not have it anymore.

It's weird because I'm with Leon in certain areas, right? If my tool can't necessarily communicate well with your tools in an agent taking a flow, like that's a disconnect. Prime example, Leon mentioned sneak earlier. I'm a big fan of them. They need to cut me a check, by the way. I've converted so many people to using this tool. Like, y'all need to cut me a check. Yeah, I'll sponsor these podcasts. Wait, S&YK. Hold on, hold on. Yeah, S&YK. Okay, yeah, let me share it here. Yeah, I like them. And they need a sponsor both of our podcasts. How about that? They're the ones with the dog logo. Is that the right thing?

Is that the right thing? I think so. Hold on. Yeah, it's the dog, yeah. I only had to type in the word letter S. And it was the first thing that popped on my brows. But I got you all. Yeah, so I do like tools that communicate well. But unlike Leon, I'm still like, especially with my side projects, like one project I'm working on right now is called Event Convo, right? Which is my solution to like meetup.com event bright, etc. I feel like it's probably the right solution for meetups, especially from the organized perspective. Conference speakers or just meetup speakers just in general. Conferences and actual attendees.

It's solving problems that people just have never thought about. And I don't necessarily know why. And like I think right now it's like at 140,000 lines of code. So when I use it for like these AI demos, it's no longer going from zero to one. I feel like everyone does zero to one. Nobody's really interested in that. Like most developers are working on something that's pretty large. And so seeing AI work within the large code base is a very, very, very different scenario. And to by the way, having a live chat is super distracting for me. Like I keep going. I don't know how I could go. I'm so twitch brain.

This is my element. I love it. I was so like high revocus. Like so I was like, what a sneak dude. And I'm like, hold on. Let me deviate right here. Sneak is a tool. It's one of my favorite tools actually. I use it a lot. And you connect it to your GitHub repos or whatever maybe. And it will do audits and we'll find security vulnerabilities if there are any. And it will report it to you. Once in a while, it has something that's not necessarily like a legitimate vulnerability. But it could be something that you should pay attention to. So you're making the willful choice to ignore said thing. But at least it's bringing it to your attention.

And so I think that especially as people are like vibe coding or building tools without like being hands on. I think this is like a really good safe way to at least protect yourself. At least protect yourself a little bit right. I'm more on the hands on type, but I like having an extra thing there in order to confirm that I'm accurate. Because one thing that I've talked about in the podcast in the past. The one thing that I'm really like geeking out about in particular is SLM versus LLM small language models versus large language models. There's a lot of information layers that are trained on certain knowledge bases.

And we literally just recorded this by the way. Go, go, go, go, do it. One thing over the weekend, I was building a tool. Part of my role, I work at Atlassian, so transparent there. But I'm creating a lot of coding content, tutorials, things like that. And one thing that I was recording was I had the idea that what if we had like a discord custom discord bot. That any time that someone got a certification and something Atlassian related. We can use a discord command and it created geraticket. Because when they get certified, a lot of us don't know. And so it's not until they tag us in a post thanking us for like helping them get to XYZ goal.

Do we finally realize, oh, you got certified. But if we had this automation there, it would create the geraticket. And then we would know to go say, hey, kudos to you for getting certified in XYZ. Cool tutorial, I feel like there's a lot of use cases for a lot of other communities as well with this. Regardless, fun little automation. When I use claw, I use clawed and I use five for three codecs to audit my code base to say, hey, did I miss anything? And both of them said, really solid, great best practices here. Because I'm using Atlassian forge, which is JavaScript using react components to create SaaS products.

And honestly, one of the cool things about it is it alleviates the reason why 99.9% of SaaS products die, which I don't know why more people don't know about forge. I've been geeking out about it. And it's actually one of the reasons why I want to join the team. But there are some people making like some serious money on it, by the way, like serious money. But you could literally use JavaScript to make a solution for like gerat confluence, etc. It goes into a market place that has served to 250,000 enterprise level organizations and they serve it there. You get their infrastructure, you get their hosting, you get Atlassian security, which is the reason why most SaaS products aren't allowed to get enterprise level customers.

You get their security clearances and their security providers just by using JavaScript. You don't need to know everything else, just serve it, right? Regardless, I made this with forge and I was super proud of it. And that's when I realized Atlassian has rovo dev, which is an orchestration layer that is trained only on Atlassian's best practices. And it found three volume like issues, codex misted, clawed misted. But because this small orchestration layer that's super specific on Atlassian's knowledge base, they were able to pick it up. Because of that, I found three vulnerabilities that honestly, two of them I wouldn't want in a tutorial.

One of them, I probably wouldn't want in a production app, but for a tutorial probably doesn't necessarily matter. But it picked up all three clawed misted, open AI misted, but rovo dev picked it up. And that's because SLMs, they're going to be the way that as these models evolve, that's going to be the next wave. I've been talking about this for over a year now, so if you listen to the podcast, you've probably already heard this one. But I think SLMs are going to be the way of the future because more nuance into your prompts is going to be where the value is. MCP has its value for sure, but SLM is a closer level of where you would want outside nuance to language specific knowledge.

I want to back you up, so Shopify just recently has said that if your PR merges and there wasn't an LLM that reviewed it, you're busted. So Danny, Danny is here trying to tell you right now, reviewing it. Hey, real good enough, hit the button. It opens the law, hired the person to then review the code. You're right. I mean, that's, that is a threat of what's happening. All right. But like, anyway, it's like what the whole point here is, it's like, you need a quorum. Have you, have you ever used an LLM scenario where there's four LLMs that you pit against each other? Have you ever pitted Gemini against,

chat GPT against, Claude against anybody, you get a quorum. So this quorum then starts to debate about the scenario. You can have them debate about your PR. And all Shopify is trying to say right now is if you didn't at least bring one fucking person into the quorum, then you did it wrong. You need to have an intense review or someone, you know, Danny's also pitched the devil's advocate. You can put that on your team and Claude, by the way. You can create a new team and Claude and you can have anybody that you want to put a devil's advocate in there. It's always good to have someone in there that's like a naysayer.

This is like, oh yeah, Murphy's law, bitch. Did you know that this could show up and really thwart your TypeScript system? And you're like, oh shit, I didn't think about that. So anyway, Shopify is all in on this concept of you need to have LLMs not just writing code, but reviewing code. And they don't have to be big language models. You can use an SLM like Danny keeps pitching pitching. And this is this is a healthy way to move forward is to have as many different checks and balances as you can. So you've got your CI CD, your tests, your unit tests, your linters, your style, all these six plus you have LLM popping in there.

This saves you from security issues. There's a number of things that this is saving you from and we're automating this. So this isn't like you're in a gauntlet trying to deliver something to a Boeing 747 airplane. We're like, you wanted to change some little factor about the chairs and you had to go through all of this hullabaloo. This is a meaningful set of fast running evaluations that look at what you're trying to do and stop you before you ever ship it. And that's just cool. I also like Leon's strategy, though, where he's like, hey, I ship it behind a flag. And yo, if it doesn't work out, then that's all good.

Yeah, you should do that with everything. Yeah. The other thing too is like, I was using 9 ball and chat said that go to burn tokens. And yeah, I was using custom sub agents to do this for a very long time. But now bug bot, reptile are 40 bucks unlimited per month. Like no reason why there shouldn't be some level of code review for every single commit, you know? NPR at least soon. If your open source code rabbit is free, by the way. Yeah, code rabbit has been really good to us so far. Just the quality of like the reviews are really nice actually. The reptile has a cooler name. I'm not going to lie.

Reptile is a sick ass name. That's really good. Yeah. Yeah. Yeah. Yeah. Sorry. I think too is like they've caught so much stuff that I never would have caught. Like there's so many things that like, yeah, never would have saw that in the first pass. And so it's just, it's also like a really cool learning tool, right? Like I'm seeing all these reviews coming in thinking about things I never thought about. So it's accelerating my pace going forward. That's the other big piece about this. I really like, you know? Yeah. Code rabbit found stuff in that like rust stuff where I didn't know any rust. It was like, hey, you're converting coordinates from a like, because it was like a 2x screen

to like a not 2x screen and vice versa and all the like all these different methods doing that back and forth. And it was like, this one's off by like a few pixels. Like you probably missed that. And I'm like, how did you like you went through all of this and found that? So yeah, it's, they're really good at like, it's I think another harness thing of like they just tell it really good stuff about being good at reviewing. And it checks that stuff that it like if you just generically through like, hey, Claude, that's the specialized models. It would be yeah, not as good. Yeah. Love it. Well, the whole two tips either.

Yeah, well, we're over time anyway. So if he's not coming back, I don't know what he's doing. We could end to here, I guess. I was calling for around to anyway. Yeah, we should wait too much time. Yeah, yeah, anytime. It did say one minute. So they put the finger for the, it wasn't like a nice. Sounds like a potty break, you know. Yeah. You went to get another coke to mix with. Yeah. So Leon, I sent you my Pokemon go code. Nice. Nice. Do you still play? Are you playing Monster Hunter? I've played, I've played older versions of Montana. I haven't played any of the newer ones. But the mobile. Have you played the mobile app?

No, I've not played the mobile. Oh my goodness. Pro is really good. The only game I play on my mobile phone, we're like, this feels like it's a next gen mobile games. And it uses all of the Pokemon go like backing, you got to be like physically walking around. Don't, I'll try it. Yeah, I know. Anyway, I still play Pokemon go here and there. Same. I'm a big TCG player. So I played PCG live. We've played in the house too. Yeah. Yeah. I'm like level 97 right now on that. Sick. Yeah. That's my main one. What's your favorite Pokemon? I know it's a hard question, but do you know it? Like favorite, like, because I grew up playing this stuff is all the sort of your childhood

favorite. Yeah. That's both sort. I've probably tweaked the other day. I showed the wears. So in 1999, I won the first Pokemon Tournament. Like I've been in this stuff since the beginning. Yeah. Yeah. I got the badges to show it. Like, yeah. There was one and you were the winner. So there was like, there's three different cups that you could win in. Like before, like, you could like level to 100, you could level to like 50, you could like, you know, like, and there's different cups. So I won my cup, but like, it wasn't like a Swiss tournament that you have now is like, yeah, you only had so many matches that you won.

And if you won, you could go play the gym leaders. So I won my cup, I won all my matches and I beat all the gym leaders. So I got all these like cool badges in a hat. Yeah. Nice. Nice. That's sick. And I was the first person to play Pokemon Go outside, which was another interesting thing. Like, hold on. Livestream outside. Yeah. Oh, I see. Yeah. Okay. Yeah. Probably not the first person to play it outside and live streaming playing outside. Yeah. Yeah. Yeah. I still can play. I dipped out about like a year or two ago, I started to slow down, but I was Pokemon going for a long time, like consistently.

It sucks that it's not as popular anymore because like, it was so fun. And then everybody's just so fickle and like, oh, new thing, or oh, I'm not on TikTok enough. Let me stop this. Like, I don't know what everyone's doing. The cool thing in LA is it's really popular still in LA, but you can go to any of the hot spots. If you go to promenade and Santa Monica, you'll see everyone just like glued to their phones playing. And so there's still like a scene here, which is really nice. I love it when you go somewhere new and there's just like lures on all the stops. And you're like, oh, man, people are playing.

I love it. I go to Japan. I'm like, people are playing monster hunter, right? And I'm like, oh, I guess a little bit, but not as much as Pokemon go. Pokemon go still has lures everywhere. And I'm like, okay. So same question to you out on favorite Pokemon and then favorite game. Oh, man, favorite Pokemon. It's the oh shit, dude, I'm going to sound like a new right now because I have a terrible thing with names. I don't know what it is, but it's the toxicity. Oh, man. There you go. I just saw it, like as you can see, guitars and stuff. And so, well, toxicity is like this purple punk rock Pokemon.

It's basically got a mohawk that goes all the way past his butt onto his tail. That doesn't sound like an original 150 to me. So it's not an original no, this is definitely a modern one. For games that I really like, dude, the Pokemon go mobile app has been by far the one I've put the most hours in. I've played a lot of the ones on Switch. They're really good. I feel like a little bit grinding in those. Oh, they're grinding on Switch. What am I doing on mobile? Oh, shit, I'm grinding. But the mobile one's like a little more casual. I suppose I got to sit down and play Switch ones, but violet. I don't know.

I really liked violet. Once they modernize to some of the interactions, I'm into that. I like them in technology kind of takes a lot of shit for that catching. Did you? I really like Scarlet and Violet. I put 15 hours before the first batch into the game. And early on, there was like this weird river I tried to cross. I wouldn't like followed it, found like a hole in the side of a mountain. Went up into this field and there's a huge arc of nine running out of the middle. I like lost my mind. I was like, I was like jumping up and down my couch like, yeah. I think those moments are really special and yeah, I really enjoyed Scarlet Violet for sure.

Yeah, I did all the things were like the mobile integrations with Scarlet and Violet. Like with the Pokemon Go app, I sat down and I mean, my kids were like sitting around and I'm like, all right, you need to hit that button right now because I'm ready to trade. And you're kind of wasting my time. So would you hurt? Yeah. I was like, the gimme ghouls drove me nuts though. And like the whole the gold lure thing, did you actually get the gimme ghoul to a thousand coins and evolve him? Hell no. Okay. I knew people that did throw around me, they're like my neighbors and I'm like, y'all are crazy. I'm like, I admire you in a weird way.

You have no life and kids, but that's cool that you did to gimme ghouls think anyway. HackSaurus was one of my other favorite ones because I'm a coder and I just love this name. HackSaurus. It's just cool. All right. Well, we probably should end at this point where 20 minutes over time, we should definitely do another one though. This was fun. Thanks sir, everybody hanging out. Thanks for you guys coming on. What do you guys want to plug or mention before we end anything for mom? What does a plug mean? I don't know the plug is. Here's a rubber circle. I'd like to keep it. You know what, I will shout out the commit your code conference.

I'll shout that one out. If you listen to podcasts, you should probably know about it. But we're bringing it back again this year. So if you have a company that wants to sponsor, I'd love to have a conversation with it. Last year we had almost 1,000 people in person, 860 was the top total for the conference in person. And we had 8,100 attend online. I definitely want to increase those numbers. So like I'm really actively working towards that. The idea of the conference for those that aren't aware, it's to do a $700 conference for 40 bucks and you should never feel like it's a charity conference.

And we donate 100% all ticket sales to charity. We don't monetize it. We don't keep it. To be honest, people asked me about that, but like, here's me being transparent because I've had enough of this. Like I make pretty good money doing what I do for a living. I don't need to make it via a conference. I just want to do something that like is different and really brings excitement to the community. And I feel like this is a way that we're doing it. But also to be completely honest with you, I spoke at a really well-known conference. And as I came back to Dallas, somebody mentioned like it's their dream to go to that conference

but they can't afford to go. And that is when I realized that this individual in particular had over a decade of experience in tech. So it's not like a junior dev that's dreaming to go to a conference. And I realized then how many individuals were literally missing out on opportunities simply because income was a barrier. So my goal was to create a tech conference where income was no longer a barrier to people getting access to technical resources that will help them grow in their careers. And I feel like we really achieved it with this conference. So we're coming back again this year for the third year.

Still figuring things out. But I think right now we're aiming towards the end of October. We're going to do something very, very big this year compared to the previous two. We had to already change venue so we're already working on that right now actively. But we want to look out for the CFP which should be going live in a couple of weeks, give a take and we'll have something big to announce. No, I shout out moms because she was the one that took me all the Pokemon tournaments. So we're talking about it was that shout out moms, shout out on the devs community. I see so many folks in chat and of course resilient coders if you're in the Boston Philly

or main area for sure. Cool. All right, well, yeah. Thanks guys. Thanks everyone for listening. If you'd like to please subscribe, leave us some ratings and reviews. We appreciate it and we will catch you next time. Be sure. You've been watching Whiskey Web and what not recorded in front of a live studio audience. What the fuck are you talking about, Chuck? Enjoy the show, subscribe. You know people don't pay attention to these, right? Head to whiskey.fun for merchant to join our discolored server. I'm serious, like 2% of people who actually click these links. And don't forget to leave us a 5 star review and tell your friends about the show.

All right, dude. I'm out of here. Still got it.

Listen

Hosted By

237: The Transactional Trap: How 97% of Developers Are Using AI Wrong w/ Leon Noel & Danny Thompson

Creators and Guests

Sponsors

Show Notes

Chapters

Episode Transcript