Consistency Without Complacency Ft. John O'Nolan  ## Metadata - Author: [[Josh Cirre]] - Full Title: Consistency Without Complacency Ft. John O'Nolan - Category: #articles - Document Note: --- tags: [#AI-coding, #open-source, #Laravel, #JohnONolan, #Ghost, #developer-tools] Author: John O'Nolan (transcript of interview) Source: [Consistency without Complacency ft. John O'Nolan](https://www.youtube.com/watch?v=7OPLPG-7Inw) Reference: Transcript file provided date created: 2025-07-14 19:45 --- # Title: 🚀 Consistency without Complacency — Insights from John O'Nolan ## Table of Contents: - [📝 Summary](#📝-summary) - [📖 Main Content](#📖-main-content) - [🗒️ Notes](#🗒️-notes) - [📖 References](#📖-references) - [🗣️ Paraphrase](#🗣️-paraphrase) - [📰 Newsletter](#📰-newsletter) - [✅ Action Items](#✅-action-items) - [🧪 Hypotheses](#🧪-hypotheses) - [📊 Methodology](#📊-methodology) - [❓ Questions](#❓-questions) - [🔮 Future Research](#🔮-future-research) - [💡 Implications](#💡-implications) - [🧠 Learning Styles Assessment](#🧠-learning-styles-assessment) - [📊 Accessibility Assessment](#📊-accessibility-assessment) - [🌍 DEIB Assessment](#🌍-deib-assessment) - [✅ Logical validity](#✅-logical-validity) - [🔍 DARVO Analysis Prompt](#🔍-darvo-analysis-prompt) - [🔍 SEO Evaluation](#🔍-seo-evaluation) - [🎯 Conclusion](#🎯-conclusion) ## 📝 Summary: John O'Nolan discusses Ghost 6.0’s upcoming launch, lessons from over a decade of open-source leadership, and the transformative power of AI-assisted "vibe coding." He explains how AI tools like Claude Code and Opus 4 allow him to build complex Laravel applications at an unprecedented speed. The conversation touches on balancing speed with control, the importance of storytelling in sports fandom, and using AI to finally solve long-standing SaaS business problems. ## 📖 Main Content: O'Nolan describes the upcoming Ghost 6.0 release, emphasizing architectural evolution rather than full rewrites. He reflects on "cleaning house" — updating docs, switching support systems, and refreshing outdated codebases. He explains his calm approach to big launches: accepting chaos as part of the process rather than fearing it. ### 🗒️ Notes - [**Ghost 6.0 Launch (Timestamp 10:55)**](https://www.youtube.com/watch?v=7OPLPG-7Inw&t=655) - Major new capabilities, no complete rewrites - Architectural evolution paving way for future growth - Updating docs to Mintlify to reduce maintenance load - [**AI-Assisted Vibe Coding (Timestamp 24:12)**](https://www.youtube.com/watch?v=7OPLPG-7Inw&t=1452) - AI tools enable tackling problems previously deemed too complex - Examples: custom CRM for Ghost, ranking algorithms - Speed and productivity increase (90k additions in a month!) - Important balance: task scope too large = AI struggles; too small = underutilized - [**Laravel as a Core Framework (Timestamp 32:00)**](https://www.youtube.com/watch?v=7OPLPG-7Inw&t=1920) - Chosen for stability, strong conventions, mature community - Easy to integrate AI coding due to its opinionated nature - Useful for rapid prototyping and large-scale data insights - [**Philosophy on Launching and Iterating (Timestamp 21:58)**](https://www.youtube.com/watch?v=7OPLPG-7Inw&t=1318) - Over-preparation can prevent shipping - Accepting "controlled chaos" helps momentum - Embrace rough launches as learning opportunities - [**Sports Storytelling Parallel (Timestamp 17:00)**](https://www.youtube.com/watch?v=7OPLPG-7Inw&t=1020) - F1 is compelling because of personal stories, not just the race - Drawing parallels to community-driven open source storytelling ## 📖 References: - Ghost: [Ghost.org](https://ghost.org) - Claude Code: [Anthropic Claude](https://www.anthropic.com/index/claude) - Laravel: [Laravel.com](https://laravel.com) - Jeff Huntley’s post: Mentioned in transcript, recommended for AI coding perspective ## 🗣️ Paraphrase: 1️⃣ Original: "These days I’m pretty chill about just accepting that launches are chaotic. If you’re fully ready a week before, you’re probably launching too late." - **Paraphrased**: Nowadays, I stay calm knowing chaos is inevitable. If you're completely prepared far ahead, it might mean you're delaying the release unnecessarily. 2️⃣ Original: "AI lets me tackle projects I would have needed millions for. It’s like unlocking new superpowers." - **Paraphrased**: With AI, I can now tackle projects that previously required huge budgets. It's like gaining a set of new super abilities. ## 📰 Newsletter: 🔥 **Dev Notes: Ghost, AI, and Laravel — Oh My!** John O'Nolan (Ghost founder) spills the beans on Ghost 6.0, sharing how they balance massive updates without complete rewrites. He champions "vibe coding" — using AI tools like Claude Code and Laravel to build beyond imagination. This approach empowers devs to tackle ideas that felt impossible (hello, custom CRMs and wild ranking systems!). John’s zen approach to chaos is a masterclass in shipping fast and iterating smart. 🚢💨 Check out Ghost’s updates and consider if AI might unlock your next big side project. [Read more →](https://ghost.org) ## ✅ Action Items: - Relate John's approach to my open-source advocacy: - Emphasize merging "vibe coding" into onboarding contributors — showing them what's possible, even without full expertise. - Promote embracing imperfect launches as part of contributor growth and innovation. - Highlight importance of stable frameworks (like Laravel) in education and community projects. - Encourage AI tools to lower barriers for new contributors — tying into sustainability and inclusive growth. - Aligns with OSPO goals: - Foster innovation and transparency in contributor engagement. - Prepare contributors for future AI-integrated workflows. ## 🧪 Hypotheses: - AI tools significantly reduce time-to-market for complex SaaS features in open-source contexts. - Encouraging "vibe coding" mindset in contributors increases project experimentation and long-term retention. ### 📊 Methodology: - Conduct comparative studies measuring contributor output with and without AI assistance. - Collect qualitative feedback from contributors on confidence levels and perceived agency. - Analyze project metrics (issues closed, new features, retention). ### ❓ Questions: - How might "vibe coding" change contributor mentorship models? - What safeguards are needed to ensure AI-driven code quality remains high? - Can AI fully replace traditional peer review in open-source? ### 🔮 Future Research: - Long-term effects of AI usage on open-source project health. - Exploring decentralized data insights like John's CRM approach in other ecosystems. ### 💡 Implications: - Could democratize complex development tasks. - May shift open-source governance and community roles toward orchestration rather than hands-on contribution. ### 🧠 Learning Styles Assessment: - **Readability**: Intermediate-Advanced (developer-centric vocabulary) - **Visual Learners**: Minimal visuals, suggest adding diagrams of Ghost architecture or AI workflow. - **Aural Learners**: Could include podcast/audio version. - **Verbal Learners**: Strong use of analogies (e.g., F1 storytelling), suggest more code commentary examples. - **Physical Learners**: Lacks hands-on tasks; suggest including exercises like "try writing a small Laravel feature with AI." - **Logical Learners**: Strong logical structure in discussing workflows and AI decision logic. - **Social Learners**: Limited; recommend encouraging community code reviews or pairing. - **Solitary Learners**: Supports independent experimentation via AI coding examples. - **Naturalistic Learners**: Not present; could use natural metaphors (e.g., gardening growth cycles). ### 📊 Accessibility Assessment: - **Screen Reader**: Good, conversational transcript; ensure alt text for any visuals. - **Keyboard Navigation**: Transcript-based, accessible. - **Contrast & Readability**: High contrast text needed in slides or code demos. - **Text Resizing**: Not specified, should ensure. ### 🌍 DEIB Assessment: - **Cultural Relevance**: Strong, diverse analogies (e.g., global sports). - **Inclusive Language**: Generally inclusive, uses casual friendly style. - **Neurodiversity Support**: Conversational breaks help processing; suggest adding summaries per section. - **Representation**: Could include more non-tech perspectives. - **Psychological Saf - Summary: This week, Josh and Will are joined by the CEO of Ghost, John O'Nolan. Ghost is an open source publishing platform that John built and has run consistently for over 12 years and yet still finds time to try new things. What's the deal? - URL: https://www.youtube.com/live/7OPLPG-7Inw?si=n8HSAR6Itq8W8rQq Ow.[Music]Hey. Hey.[Music][Music][Music]Down, down, down, down, down, down,down, down, down, down, down.Heat.[Music]Heat.[Music][Music]Hey,hey, hey.[Music]Hey.[Music]Hey. Hey.[Music]Hey[Music]baby.[Music][Music]Heat. Hey. Hey. Hey.[Music][Music]Hey. Hey. Jesus.Ow. [Music]Ow.[Music]Hey[Music][ \_\_ ][Music][Music][Music]I think you’re actually muted. Mill willuhhappy Friday.Happy Friday to start the show. HappyFriday. What a great way to start.Thanks for joining us for Riff andRefine. I was about to make a joke abouthow I didn’t take a giant bite ofsomething this time before starting, buteven better, I just left myself on muteas I started talking. So, you know, oneof these days I’ll be able to do thisshow like a pro. But let’s go ahead andintroduce our special guest, numero unoto the show, John Onolan, former tattooartist, contestant on a tech-based gameshow, and oh yeah, I guess he’s spentover a decade building Ghost, andshows no sign of stopping. Good to haveyou on, John.Thanks for having me, guys. It’s anIt's an honor to be the inaugural guest. Absolutely. We were like, "Who should wehave first?" And Josh was like, "We haveto have John." And I was like, "Ah, Ijust met John at React Miami. Heseems so cool. I would love that." I waslike, "What? What are youthinking?" And he's like, "Well, hekeeps posting about all this stuff,but not giving any details, so why don'twe just bring him on and then we'll makehim talk about it."Oh, but usually that's kind of how it goes. Yeah. Yeah. Usually, we kick off the showwith a few updates. Since you're ourguest, we'll let you slide. Wewon't make you go first. Josh, why don'tyou give us a little breakdown ofwhat you've been up to? Yeah, this past week has been reallyfun for me. I've beenkind of all into the world of howI can automate a lot of what I'mdoing in my personal life, myprofessional life. I'vebeen just in this world ofrunning Claude code dangerously, likethe article we read.Yes.Yeah. And I've been, you know, justIt's kind of like asking for more help frompeople, too. So like we haveRobo Josh. I don't know what that means,butyou're automating everything. Soyou're, you'reOh, there we go. Exactly. Yeah. I am, Iam completely automating not myselfbut everything else that is like, "Hey,here are all the things that I wish I wasdoing better and faster,"you know. Um, and like just asking forhelp more, too. I think that's like thebiggest thing I've been doing, like thispast week is, in everything I'm like,"Hey, what if I didn't try to do thatmyself?"I wish I could do it; I wish I had that,self-control. I'm still terrible at it, but now it'slike, "Okay, let me, you know, let mefind someone to do thatfor me, or let me just automate it, orjust like, "Hey, what would happen ifthis was automated?" Even if it isn'tperfect, at least I could do it; it freesup my time, kind of thing. So, that'sbeen my week. Just been kind of likefiguring out what can I not do in a lotof ways. What about you, Will? Well, what you just said has gotme thinking about how I need to do thatbecause my week is like everything, aswe mentioned last time; everything is kind of just changing forme right now with the company beingacquired. I am so busy. But the thingsthat bring me joy and the thingsthat make me want to wake upevery day and do the stuff that wedo are usually some sideprojects that I'm working on, where it'slike, okay, I'll get my job done. I'll do what I'm supposed to do therebecause I know I'll have a littleextra time to do this fun thing orthis ambitious fun project.And I'm not having time to do anyof that now, and it's just sodifficult. It's just been really busy weeks. I needto find more ways to automate thingsfrom my life so that I can get some ofthat back in my life. But one thing that is inmy life right now is working on myLaracon talk, which is coming along great,if I do say so myself. I don't knowabout the talk itself, but theslides are really the thing that I'mvery used to designing. Things are going really well. Um,well, funny story, I’m in thisdesigners group onDiscord or whatever, and it’s abunch of people who are roughly Idon’t know, a thousand percent better than I amat design. Uh, and so there’s thisdope work channel in it, and I postedmy opening slide and was like,"Hey, um, this is the slide for thetalk I’m working on at the end of thismonth and this kind of tries tocapture the vibe of what I want therest of the slides to keyoff of. I’m really pumped about it, butI want you to tell me everythingthat’s terrible about it." Um, and oneof the guys who’s likesuch a good designer looked at it and immediately just exposed me. Hesaid, "This looks um," Hesaid, "Um, it looks like you’re tryingto be dangerous,but you’re not. Like, it looks like youwant this to feel edgy, but it’svery safe. Like, you’re doing thisgrid. You’re doing like all theseAnd he totally just nailedit. And I was just like, "Oh, dang."Like, you got me. That is 100% theIt's like I wanted it to feel likethis really cool progressive, like, kindof design. And he's like, "Nah, dude. You didn't nailit. Uh, here's what you needto think about." And he gave me reallygood feedback on what to do. And it got better after that. Um, as it usuallydoes when you actually putyourself up to get critiqued. So that was a fun andenlightening experience this lastweek. But yeah, that's about it,other than being a solo parenttoday. I took the kids to lunch. Um,and let me tell you, me and three kids,six, three, and one, was quite theinteresting lunch. Let's just say weleft a heavy tip for whoever had toclean up the table after we left. Oh man, I feel that.Yeah. John, what have you been up to thisweek,man? All kinds of stuff. We're gearingup for our Ghost 6.0 launch.[Music]Oh no.Oh, I think you just muted yourself.I think you accidentally muted yourself.Haha. I'm not the only one. They put the right space bar.Yeah, on the space bar. Must be.[ \_\_ ]Oh no.Um,can we swear? Is this a family-friendlyshow?Yeah, we'll go for it.It's live. You can do whatever you want.Whoever the [ \_\_ ] we want. Um, so gearingup for launch, Code 6.8 is our first bigmajor release in a couple of years, andso it's kind of crunch time, gettingeveryone working together to getit across the line. Um, we need to get everythingdone so we can announce it to the world,which should be fun. Um, a lot has gone into it and it feelslike there's still a lot to do. So, alot of that is kind of hoppingacross; you know, all of you guys willknow many, many different things you haveto update when you're doing some kind ofbig launch. And then somehow, everysingle time, you find a long tail ofextra things that also need to beupdated, and we should probably do thatnow, right? And the list just keepsgetting longer.Oh man. So, I'm kind of in the thick of it.Yeah, I run into that all the time whereit's like, "Oh, I just got todo this one thing, then it's done." Andthen when it's done, you have thelist of all the things to do when thething's done.It's like, "No,it was dependent on at least, you know,eight other things." Now,and there's always like one more thinglike, "While we're here, let's migratedocs platforms and while we're here,let's switch to a new support inbox helpdesk system." And you know,how much are you absolutely changingwith this? Like, what how much of like acomplete rewrite is a lot of this? Uh, the product is adding a couple ofmajor new capabilities. Um, nothing'sbeing rewritten.Um, so, and it's architecturally, it'sevolved quite heavily, which is kind ofcool and it's paving the way for somemore stuff we're going to be doingin the next months and years. Um, andthen, website-wise,you know, we're trying to kind of, it'syou get to that point where themarketing website is just getting long. In the tooth, you know, and every littlechange is like,yep.Oh, goodness. Like we're like threeTailwind major versions behind andthere's like whatever came beforeTailwind CSS system is also still inthere.Still kind of in there. Yeah. Yeah.Yeah. So like pulling the docs out andthrowing that over to Mintlify is aneffort to one improve the docs, but twojust like make that codebase a littlesmaller so that it's then easier toevolve beyond that. So there's a lot ofthings like that that are getting refreshed. Butjust a nice little cleaning housekind of situation. Yeah, there's a lot of cleaning housecleaning.You're never—it's never the fun stuff,but it's like the stuff thathas to get done for your—like I don'tknow. Like I don't—I don't run abusiness, but I do a lot of the tasks. Iwork at startups usually untilrecently. And that's like that'salways the stuff that surprises youwhere it's like, "Oh, like you're not justA code monkey or a designer, you'renot just doing your one little area. It's like, nope, there's like a bajillionthings all the time, always. TheMintify docs migration has been prettynice though, because um however manyyears it was ago that we last refreshedthem, maybe like five years, I kind ofjust like ripped off Tailwind docs quitea lot, right?Um,which a lot of people did, and Mintifydid too.Right,so when we migrated our docs, uh,it really doesn't look any different. Imean genuinely like it's added a fewfeatures, added like AI search and thatkind of stuff, right?But the layout and the navigationand the design haven't changed atall. So, it kind of, uh, fits in reallyreally easily. It was not very much workat all. So, that was sweet.I like, I love, I feel like that's thelike a lot of, uh,in terms of AI stuff. I think what I getmost excited about, most excited, it's abit extreme, but like I think what ismost interesting for a lot of businesses. Is that all the things that were notworth your time to do to make the quality oflife things better are now a loteasier because you can automate a lot ofthe drudgery that is migratingthings from X to Y. There's a lotof that work that would have had to bemanually done. So you're like, man,that's just too much time and not worthit. On the other hand, that is now justa completely different calculationnow. Yeah,you got to share that one. That one'stoo good.I got two questions for you. Onecomes from Jason. Do youTaco Bell reference?Taco Bell.Taco Bell mentioned.I think I had Taco Bell once in Canada.Okay. It wasn't good.Denver, I will buy you Taco Bell. Jo,you had Taco Bell overthere in London when you were there,right Josh?I did have Taco Bell in London, and itwas funny. I don't know if Taco Bell wasThat was an upsetting photo. Oh goodness.I don't know if Taco Bell was listeningto me, but it was funny because I postedabout how there was a menu itemthat I ate in London that I was like,"This is really good. We don't have thisin the States." And then literally thenext week it was in the States.I'm telling you, we need a Taco Bellsponsor for you, Josh.It just needs to happen. Do you think the UK is like thetest market? You know how likeFacebook rolls out features early inAustralia to see if they stick?I'd be interested to know. I shouldprobably know more talkable lore likethat.You should if you're going to be asponsor, you got to know all the arcaneknowledge. That's a fact.We have a question from Rob. Idon't know if you know what this means,John. But John, who is going to replaceMax at Red Bull?Oh, Rob Hope's in the chat. Dreams docome true.I know of Rob Hope because I…Watched, um, I have takenmultiple of his courses. Um, but Ihave not gotten to meet him yet. So gladto have you, Rob. But I don’t know what this means, John.What does this mean?One of my all-time, uh, greatfriends, uh, Rob, I haven’t seen him fortoo long. Usually we hook up when I’m inCape Town in South Africa, but it’s beena few years since I’ve been back. Uh, Maxis referring to Max Verstappen. Red Bullis referring to the Red Bull Formula 1teamand F1 reference.Where will he go? For anyone who followsFormula 1, there’s been a kind of, uh,news firestorm this week of their teamboss getting sacked. At the end of lastyear, their chief designer of the carleft and went to another team.And rumors have it that Max, theirthree, four-time world champion, isgoing to be on the move now as well. So,where is he going to end up? I thinkMercedes. I know. Here’s what I know about F1.I think this is coming.I know the movie.I know that a lot of people in AmericaI got really into F1 because of adocumentary that came out,Drive to Survive. Correct. And I did not watch it. So, Iam not into F1,but I am curious. Have you been a fan of F1 fromthe way you talked about it before thedocumentary?Yeah, I've been watching.I’d say that’s the case. Yeah. So, how doyou feel?How does it make you feel about this?Like,maybe call them fairweather fans ormaybe trendy fans who are like, “I’m afan now because it got popular.”I think it’s great. I think more peopleshould find enjoyment in something anddiscover,you know, what I love about a sport, thebetter. Um, I think what makes sportinterestingis often not the act of competition, butthe stories behind the people who arecompeting, right? And on the face of it,Formula 1 is just a bunch of peoplesitting in a car driving around incircles for two hours.That’s not the interesting part. That is just the canvas upon which thestory is woven. And the story is theseven-time world champion up against thenewcomer who is vying for his very firstchampionship and who doesn’t care.He’s crash happy, and he’s going to gointo the corner hard, and he’s not goingto break.But the seven-time world champion, welove him. He’s been around for a longtime. And this other guy, he’s just comein; he’s a little crazy. He’s grumpy. Hedoesn’t give interviews. He just tellspeople to [\_\_] off on the radio everytime. And so you get these charactersand throughout the kind of format ofthese races, of them driving around incircles, you get to know thepersonalities and the characters behindthem. And I think what Netflix managed to dowas they managed to capture that. Theymanaged to capture that against thebackdrop of fast cars and glamoroussettings and captured the hearts andminds and imaginations of people whofell in love not just with the sport, butwith the stories and the characters andthe jokes and the inside kind ofBaseball— and that's what makes it reallyinteresting. You watch a couple of racesand you're like, "The race is kind of cool,"but you get to know the people, andyou're like, "Oh, this is great! This islike a character arc that I’minvested in." So then when stuff like this weekhappens, it's like, "Oh, this is huge,because this is like a completenarrative shift that nobody saw coming."This is the Kardashian for dudes,just for dudes. Like, you're following an arc.Oh man. Okay, I have now been drawnin there. I think I might have tostart watching F1 tonight because mywife is out of townfor the evening, and I'm on my own.So that sounds like a perfect thing towatch because I got nothing.Just like Jason said in the chat,like, "I think your single two minuteslike convinced me, hey, I shouldbecome an F1 guy. Like,at least try. I’m definitely trying it.I’ll try it tonight and next weekI’ll let you know how it goes. I did just see the movie a couple ofdays ago as well.Me too.Oh, you did? It was nice.Really good.Man, I cannot tell you the lastActually, I take that back. I can tellyou the last time I was in a movietheater. It wasum, when I watched the Paw Patrol moviewith uh Palmer, who was three at thetime. She's now six.And me, whilemy wife promptly fell asleepbecause she was like perfectly content.It was just her. The other kids were atthe grandparents, and we just were like,"All right, nap time for us. You enjoyyour movie."Oh, nice. Awesome. Okay,I have one more question for you, John,before we jump into a topic. Youlike, you mentioned you're preppingfor V6 for Ghost. I imagine you've hadat least five big multiversions beforethis. What is like one thing that you'velearned that you feel like you're betterat now, 10 years later, when it comesto, "Hey, what should we do next?" kind ofthing? That's a good question. In the contextof these big launches or justlaunching for big launches too. Yeah.Yeah.It all works out in the end.I think that's a great way.I think these big launchesalmost always have some amount of stressaround them. You know, how are we goingto migrate customers? How are we going todeal with backwards compatibility?When's the copy going to be ready forthe website? How is this all going tosync up? Andyou know, it never really comes togetheruntil a day before, ever. And so I thinkthese days, I'm pretty chill aboutjust accepting that that's how it is andnot really getting stressed out aboutit. Um versus I think 10 years ago, it was areal sense of chaos. Whereas now it's just kind of the chaos ispart of it. This is how it goes. There'salways going to be stuff. You can't pullit all together. If you're ready witheverything to launch a week inadvance, you're launching too late. Uh,you know, it should to some extent belike a burning wagon rolling out theThe door with people running after it,desperately trying to put out the lastcouple of fires andput the wheel nut on before it fallsoff. Um, that's just how it goes. And,I think when you kind of make peacewith that,it's a little easier to enjoy it. Okay, I like that. I'm gonna keep thatin mind because I usually overprepareand it's usually, like you mentioned,there's also this point where it'slike you spend some time on things evenif you haven't gotten anything out thedoor yet for like something, a new idea Iwill spend too much time trying tomake itnot feel like a dumpster fire that itthen just never shows up at all.Like I lose my momentum and Ishift to something else before Ieven get across the finish linesometimes.You can also make it worse, right? It ispossibleto overdesign things,especially in design.Yeah,I definitely feel that there'sDon't overthink it. All right. Yeah.I want to hear what youbrought to the show, John. What have youbrought for us?A link, a topic, links orwhatever you want, man. Whatwhich one would you rather talk aboutfirst?Guest choice.Some of them are kind of related, so wecan kind of weave some together here. Ithink topic-wise what has definitely been onmy mind lately isjust vibe coding. Just AI coding. It'shard for me to think about much else atthe moment becauseI've got a question. Vibe coding. Okay.When you say vibe, do youmean actual vibe coding or do you meanlike AI-assisted coding? Do youmean like I don't look at any of thecode and I just yolo whatever shows upon the screen and then I try to use itand if it doesn't work, then I just askthe magic box to try again? Or is it morelike the— is it more of like the— likeI'm a professional? I'm approaching thiswith a scalpel versus like, uh, it'sA magical genie. It's somewhere in themiddle. It's such a fuzzy line, right?Everyone has a slightly differentdefinition of vibe coding.I never open the file; I just askthe magic box for things and a UI comesout. Extreme end. Um, definitely notthat, but equally not um, I'm codereviewing this with the rigor of umcollaborating with a human being. Forinstance,I haveon my main project right now 12,800 andsomething tests. Just for the audience, what's your main projectright now?Oh yeah, we'll get to that. We'll get tothat.Uh, I haven't even read any of thetests. Uh,let alone. Yeah. If there is a thing thatyou should not have to ever read when AI writes, it shouldbe tests. If you can run them and theypass, I'm good, bro. I'm fine.Yeah. Until every now and then I'll seejust like the diff of a test file fly byand you're like, "Wait, why the [ \_\_ ] areAre you testing that? What's going on? Whatare you testing? You're testing that Ais not equal to B. What's going on? Stopthis now. Um, so it's somewhere in themiddle, and that's it; it's going so quickly.The capabilities of what is possiblenow for me individually, but also morebroadlyare just so insane to me.Um,and it's very, very exciting. I mean, Ithinkyeah. Yeah. There's someone who I won't name whotweeted yesterday, like a GitHub chart,and it was like, "Here's everything ourteam shipped in the last 30 days,"right?And it was like a thousand commits and there weremaybe 30,000 lines of code changed. I've done three times that, just me,in the last month.It's nuts, like the amount that I'mdoing and able to do.Right. Right.Is crazy. So that feels a little insane sometimes.Yeah.That's where I am with it. I've got IBrought a link that kind of speaks tothat. That's by a guy called JeffHuntley. Where do we post these? I meanI sent it to you already, but do youstick these in the chat or something? Yeah, we'll take care of it. You'regood.It's the AI versionof good old Aaron Francis. You canjust do things with an AI flavor.And really the case he makes, and thisis actually a post from February. It'snot like a super recent one. It'sjust one I had bookmarked because Ithought it was particularly good. Thecase he makes isOh, yes. Okay. I actually know exactlywhat you are talking about here. Allright. Pasting it in the chat for y'all.Actually, that's not it. That's I don'tknow what that is. Oh, it's like theTwitter link to the thing. So, I'm goingto get the actual link for y'all becausethat's garbage. Y'all,I'll put it in right now. Perfect. He makes the case that thefuture belongs to idea guys who arewilling to completely upend theirworkflows andsee AI as a completely new way of doingThings. And um, I like the way he's written it.You know, it's not toomaximalist. It's not too minimalist.For me, it strikes the rightbalance of how to think about thesethings, right?Um, and so that's yeah, I don't know. Icould talk your ear off for an hour onthis topic alone, butdo it.That's what we're here for. Becausewhat are you guys doing with it? Here'ssome thoughts that I've had about it, which is justlike I saw somebody postthis whole thing where there's likesome company that made some kind of likescientific research of some sort thatsaid people are actually less productivewith AI.Um, they had people who didn't use AI andthey measured output or something andthey had people with AI measure or likeI don't know what measurements they used,but it just got me thinking about itwhich is just like, uh, kind of goes backRemoving the barrier to entry to things,which is to say, I think that there is a case to be made that you mightactually be slower on the things that you’re using AI forin some cases than you would if you justcoded some. If you were working on somethingthat you didn’t have AI to help you with,from the perspective of, I think,when you don’t have AI, what do youchoose to work on? You choose to work onthe things that you have as thehighest priority, but are also thingsthat you know you can do, that youunderstand,and you, of course, like. Because you havethe domain expertise, you can fly throughsomething as a good programmer when youknow what you’re doing, and you know the fullscope of work to be done. WhatI think AI does to a lot of decision-making,kind of like this article talks about, isjust like it throws that initialdecision into a completely differentpath where it’s like, okay, because Ithink I can tackle something harder or Ican tackle something I don’t fullyunderstand, I’m going to dive intothis topic that I don’t have as deepdomain knowledge on. Take a little longer than the task thatyou have the full domain expertise on,but it's way faster than if you tried totake that on without any assisted help.And so, to me, it's like opening thedoors to, yes, I may do slower worksometimes, but it's because I'm takingon work that I never would have thoughtI could take on before.You know what it makes me think of? Ithink studies like this arevery silly.Uh, borderline ridiculous. I think it'slike if you went back in time whencomputers were first invented. They filleda whole room. There was a server rack andyou had a bunch of punch cards andthere were a bunch of people going, "Wow,these computers are so incredible. They're going to change the world."Right.You're computing. It's like a hugecalculator.And you time those people at basicmathematics tasks.Yeah, in putting a punch card into thisgiant machine and, oh, you think it'spowerful and you think you're so muchfaster, but actually, you know, if youhad just followed what we taught you inSchool with lines and pass carry the oneand add it all up. That’s, in fact, muchquicker. And you might be true on a micro level,but on a macro level, youare not seeing the wood for the trees. You are not seeing the broader point,which is not where the technology isnow, or what tasks are being done with itright now, but what is the trend? Whatis the trajectory? Where is this goingnext? What does the skill of knowingsomething about computers give you 12months from now, 24 months from now,and what will you be able to do with itthen? Yes. Yes. Yes. That concept, likethat idea is just like what I love mostabout this shift that’s happeningright now is just like people beingwilling to try something they didn’tthink they could do before and gettingto see all the cool stuff that comesfrom people who had ideas but neverthought they could make those ideas areality. It’s so exciting. That’s me and Laravel, by theway. Like GPT.What do you mean you and Laravel? Whatare you building in Laravelagain? Yeah. Okay. Okay. I don’t remember.Fine. Fine. Fine.So, I’ve been messing withLaravel since like 2015, but I alwaysgot to the same point with Laravel. So,I’m primarily a designer and front-enddeveloper. Andit's been an on-again, off-again relationship, if youwill. Yeah, and I can reason my way fairly fardown the stack comfortably, particularlyin terms of understanding high-levelconcepts. But then when we get into[ \_\_ ] middleware andrequest routing, you know, the guts ofwhere things start to get ratherabstract and the logic chain startsto get hard to follow. Um, there’s ajust a syntax error you could get stuckon for a month, and you need to readthree books to understand what it is. Uheventually, I would always hit some pointlike that, and then I would just be outof my depth and I wouldn’tbe able to keep going. But around GPT-3.5, I could start solvingthose problems by asking questions. Itwasn’t coding for me. But whenever Igot stuck, I could copy-paste and be like,“Hey, what do I do now?”That unblocked me. And so ever sincethen I've just been kind of going ham onLaravel projects. I've found that AI isvery good at Laravel because I thinkit's been so stable and so welldocumented, and it's such an opinionatedframework that there aren't a millionways of doing things. There are like threeor four, and everyone kind of chooses oneunless situation X, and then you do one of the other ones,right? Um, so I've just been going nuts. I'vebeen doing like side projects, smallthings. I think that's kind of where Istarted. And then more recently, I'vebeen solving a problem that I've wantedto solve for 12 years — a ghost that wasout of reach and impossible. Prior tonow,I think would have cost me millions ofdollars.Yeah.Um, and is currentlycolor me intrigued. That is a fantasticway to introduce this. So if you are building a SaaS business,usually, say you're using Laravel, doesn'treally matter. Whatever you're building—a SaaS app, people sign up; you've got a bunchOf customers, maybe there are tens ofthousands of customers. You got a bunchof people using your thing. What are you going to do next? Who isyour top user? How much are they using the product?What's going on with it? Is thisfeature getting adopted? It's pretty easy to find out, right? You justhave a look in the database. You write aquery, and it spits out a report, sorting usersby number of likes, number of posts,whatever it is. But when you makedecentralized, distributed software, which is not one codebase with onedatabase, but every single customer hasa new instance of the app with a newdatabase in its own container, completelyisolated. You really can't get a sense of whatanyone's doing, who the customer is.Yeah.Who's the most active user? Howmany monthly active users are there?Not a clue. We havethat's interesting.We had 30,000 plus customers for manyyears. Many more tens of thousands more thanSites are more than that. Within those sites arehundreds of thousands of users whoare logging in and using the product, andwithin their sites, they have tens ofmillions, hundreds of millions, tens ofmillions at least, uh, subscribers. Yeah,and members. What's going on with all that?Well, we've got 30,000 MySQL databases,and doing a join across 30,000 MySQLdatabases is not really feasible. So,we'll never know, I guess, was theanswer,right?So,um,I feel like we need a drum rollsound effect at this moment. The first thing we did a couple of yearsago to try and solve this for thesites that we host, where we have access toall the databases, is that we justdon't have any tools to find outwhat's in them or to understand what'sgoing on with our customers. Uh, we builta stats service that essentially goesand fetches some data, dumps it all inone big table, and well, a few big tables,and then it lives there. So at leastnow you've got data extracted. Lives somewhere, and then we kind of tried hookingup Metabase to it so you could sort oftake a look at it. But if anyone's evertouched Metabase before, you will wishthat you hadn't.Not fun, not fun; been there, done that.Nothing fun about it. And then allthe data is like this raw; it's juststuff that's come straight out of my SQLdatabase. It's not in any usable format.You can't browse it; anything you clicktakes an hour and a half. Long story.I built a Laravel custom CRM with anAPI that ingestsdata from our private databases aboutour customers, goes out to the worldwideweb, and hydrates that data with whateverit can find. Say like Alexa rank, AHF'srank, open rank,Twitter data, Facebook data, socialdata. It pulls in as much as we canfigure out about a website from itspublicly available metrics, combined withprivate metrics. And now I have a way ofseeing who are our top sites, which onesare getting the most subscribers,which ones are getting the most growth,which are the ones that we should bePrioritizing which users used to loginto the product a lot, but now haven'tlogged in at all and might be slippingaway, and maybe support should reach outto them. Um, there are like a thousand usecases that connect to it, but it'sessentially this box of an app that I can plug anything into, either datasources, use cases, or admindashboards, API endpoints,and find out how I should be thinkingabout the business as a whole. How can Iunderstand what the customers are doingum within the product? How do we usethis data to help support teams in doingcustomer support or product teams inplanning what to build next or whichcustomers to go and talk to aboutgetting beta feedback or marketing teamswhen they need to get a testimonial thatthey want to put on the homepage. How can we connect all these differentpieces together? Now, the app withAIis quite easy to build.Yeah, but it's the complexity of thenumber of different things it does. Ithink would have, prior to maybe a yearago, taken a team of three to five, sixto 12 months. That's a million bucks. Like,um, two months with Claude Max,anything is possible. Uh, so that'slike my baby project at themoment, my main one, uh, that I'm spendinga ton of time with. And I'm justcontinually blown away by how mucheasier Laravel makes all of this becauseof how many of the conventions are justsolved out of the box. This has beendone before. You're not the first personto try and do, you know, scheduled jobs,or emails, or batch processing, like anyof this stuff. Not to mention incrediblepeople like Aaron Francis, who alsodoes a ton of weird automated stuffwith Laravel that you canpoke on, and Josh, who does great videos. I mean, Laravel has one of thebest communities, not only because it’s so old,mature—maybe is a better word, becauseit’s so mature. AI is so great. Um,yes,AI is just fantastic for itsability tounderstand what you're saying. Like,there are three or four good waysto do this, and it has amuch better and easier job of figuringout a good path forward. It's not great; there's thisincredible community withLariccasts and all the smart, friendlyindividuals in the community who want tohelp.Yeah.It's also the stability.Yeah, because if you go back three yearsin JavaScript, whatever it was we usedthen, we hate it now.Yeah, it's garbage. It's absolute trash.Can't believe you're still using that.Yeah.Yeah.Yeah, it's cool that you say thatyou started with GPT-3.5 withLarbell because that was around the timethat I also started building myfirst applications with Larbell.I was like, "Hey, I don't know, um, like, Iliterally don't know what a job is. Idon't know what batching is. I don'tknow, you know, what queues are becausethat doesn't really exist too much inthe JavaScript world. Like you have toreally dig if you want to do all of thatin Node kind of thing.And so I would copy all my code to ChatGPT, copyit back and forth. That was thefirst real app that I built within. Laravel, like my first SAS project,kind of thing. Um,you know what this idea makes me thinkabout?What’s that? You know that, per, you know,Professor X, he goes into his littleroom with a little bridge, and it'sjust this giant room that makes him feel likehe's extremely sorry. Like, that just like—you’re making, I think it—I don'tremember, maybe Cerebro? I think it’scalled Cerebro or something, like you’remaking Cerebro. You’re making this, like,giant brain or whatever. Sorry, Josh. All right, your project. No,no. I was just thinking of howthat kind of kickstarted it, but atthe same time, kind of like what yousaid, Jonas, there’s not too muchthat’s changed. Even currentmodern—I think the only majorthings that it gets wrong are some issues likewhen it comes to the scheduler; itdoesn’t handle the Laravel 11 app service, but even that isslowly improving with things like MCP servers andstuff like that. It's usually able toimprove upon that when Laravel 10 cameout—or Laravel 11, I should say. Um, butwith that, what tools do youCouldn't have built this a year ago, yousaid,but so that leads to, so like, what isyour preferred process? Like, what toolsdoes EMTT say? What tools are you using toLaravel vibe code? How did you writedouble? What you said, like soaround like 6000? Yeah, some crazy amount. Uh, it was 90,000 additions and 20,000deletions in the last month. Um, that’s inone repository; I have more thanone.Not a big deal—just one repo.That’s just—that’s the main one, so it’smore than that. So the answer is, itchanges constantly. Um, so the answertoday will probably be different if youask me a month from now, and it wasdifferent a month ago. But today it’s, uh,Claude Code exclusively, Opus 4. Um,usually embedded within Cursor, likewithin the Cursor terminal. Doesn’treally matter where it runs. Um, and then usually two or three of themrun in parallel. Um, and then one one thingthat I think is a bit of a hottake: I think anyone who’s unconvincedby AI coding hasn’t tried Claude Codeand Opus 4. Um, because I've spent a bunch of timeeven in the last week, I was like Ishould try all the tools again in caseone of them's like overtaken and gottenbetter. And so I went deep into thelatest codecs, the latest Gemini. Ididn't try the new Grock, um, all the 03models, uh, through cursor, through clawcode, through the new Gemini CLI,through the new codec CLI, through theGitHub Actions web interface—all of it. Nothing touches Claude Code and Opus 4.It's just no competition. And if I doany of the other ones, it's a world offrustration. It goes around in circles.It deletes all my roots and thenit can't run the tests, and then itdoesn't understand why it can't run thetests. Um, but Claude Opus,my goodness, it's unbelievable. It's ifyou hit the right balance of how todescribea feature. Um, and the right balancemeans how much fidelity do you describeit with? If you describe it with toomuch detail,the agent won't be ambitious enough. Itwill just kind of go and do exactly whatYou're not seeing it because you'regiving it like a little cookie skill. But if you ask it to do too much, it goes nuts, right?And it gets stuck and it ties itself in knots.So, you kind of have to learn what theright size of task is.And then when you notice it—either if it’s getting it inone shot, you’ve usually done somethingtoo small. You’ve given it way too smallof a task if it gets it right on the firstgo.And if it immediately ties itself inknots, you’ve gone too big.And so you kind of just have to constantly adjustto find the right spot. But I would say to anyone who’s playedwith these things a little bit, ask for bigger things than you wouldexpect because often it will surpriseyou with an implementation that—um—justblows your mind.And like my best example of this is so this customerdatabase I’ve been working on, right?Um, I wanted to write a ranking algorithmfor it.And the idea is that I want to have abunch of data about customers. I want toaggregate all this data and then, uh, giveit a Google—effectively like a GoogleRank, so I can have an index, an order, sothat when, let's say, a support ticketcomes in, we go, "It's from customer namethat you've never heard of. That'scustomer number 23." And then you go, "Oh,that's our number 23 most importantcustomer." So now when that ticket getsescalated, the whole team just talksabout, "Oh, this is an importantticket. It's number 23." Versus if aquestion comes in of like, "Oh, it'scustomer number 23,482. It can maybe... it can wait till Monday.Like, we'll get to it. It's chill. Um,so that was the idea. So my prompt,because I didn't really know where tostart. Like I said, I want to have, Iwant to have this algorithmand I want it to rank customers. I wantit to take into account, you know,like how popular their domain is, whichis sort of like an estimate of how muchtraffic they get, as well as howrecently they've used the product, um,how many of their features they'veused. So if they leave the defaulttext or the starter welcome post, that'sa bad thing, but if they've writtenmore than 10 posts, that's a really goodthing. Like that. And Claude goes, "Cool, I canhelp you with that. First, let me writea CLI tool that will test how thealgorithm works and allow us to evaluatechanges." It then went and wrote aLaravel command that would have a dryrun feature where you could run thisranking algorithm and see wherecustomers rank now and based on a changemade which sites moved up and whichmoved down and by how many places. Thenit started writing the algorithm and ateach step of a change it made it wouldrun the CLI tool that it just wroteitself to evaluate its own work anddecide whether or not the thing that itwrote was good or not. Now, had I givenit a prompt of I want you to write aranking algorithm that works exactlylike this in this order with theseweights,I never would have had that, like,say to do that. Yeah.Yeah. So, you have to every now and thenyou got to let it go offand see what happens.Thatis so interesting. That's reallyinteresting to me because I think inSome ways, you know, you kind ofmentioned at the start, you're like, "Hey,the reason why I never clicked with Laravel wasthat I was more of like—I knew the bigpicture stuff, this is what needed tohappen, but I was more of like afront-end dev, like a designer. You'rethinking of what is the actualoutput, the actual outcome, that kind of thing. And I think so many people, especiallymaybe Laravel developers who have beenbuilding Laravel for so long, know like, "Hey,this is exactly what I need. I needthis service provider. I need thiscommand. I need this particular batchoutput." And so most of the time we'retelling AI, "Hey, do this this way."And I even fall into this too, wherelike, "Hey, I need a commandto do this."Yeah.And it just writes a command. Butsometimes, maybe it’s not a command Ineed. Maybe it’s like a scheduled joband I’m just not thinking that highlevel in a lot of ways. That’s reallyinteresting.It’s a lot of fun.That does sound like a lot of fun. Oh,man, now I want to do more projects. This is what I'm talking about. I needmore time. I need to schedule.I'll tell you a good cheatcode related to this that I just learnedtoday, which is technically I broughtthis as a link, but okay,it's got a twist. There is no linkbecause Claude apparently doesn'tpublish their change log. So, yousort of have to rely on undercoverinformation that people leak on Twitter. They have just made it possible for theGitHub action Claude thing to use your Maxsubscription. So if you run slashinstall.Okay. I was going to ask about that. I waslike have some incantation thatyou sent us.Yeah. So you go into your repo, you runClaude. It pops up the Claude UI. Youtype slash for shortcut and it'llautocomplete. It's install GitHub orinstall GitHub repository. You run that.It will then create a PR into your GitHubrepo to add a GitHub action which willrun Claude in a GitHub action. Then onany GitHub issue, you can say, "Fix this issue" and it will literallystart responding to your GitHubissue, open a pull request, and then youMerge it. But what's cool is you can also justattach, like, a GitHub action, you know.Install PHP, composer, install npm, installnpm dev, and have it run the whole testsuite after it has fixed everythingor written new tests or done everything.Now, before you could do all that, butit would cost you extra because you hadto put in a Claude API key. Now, you canjust run that install command, and evenif you've already installed it, it willgo and update the existing one, and itwill use your existing Claude Maxsubscription. So, it's included forfree. And then there's an extra cheatcode in here, which is, you know, there'sthis five-hour window of your limitswith Claude code. So when you send amessage,okay, so you send a message on Claudecode, you have five hours of API limitsand then they reset. So if you go opus, opus, opus, opus, opus,after three hours, and then you're like,poof, no more AI for you. You'retwiddling your thumbs for twohours.Go to our lesser models. Good luck.So the time window starts from when youSend your first message in a 5-hourwindow. So if you send your firstmessage at 10:00 a.m., it runs to theend of the working day where you're just kind ofscrewed. So, if you set up anotherGitHub action that opens an issue andtags Claude at 8 a.m. or 7 a.m.,3 hours before you wake up,you have this extra time; you're preloading acontext, a limit window that wasn’tavailable to you before because you wereasleepand doing no work for a coupleof hours while potential tokens accumulate. So thenyou can wake up, use some tokens ashard as you can, window resets, now youget to go through until 5:00 p.m. and, um, soI've got a GitHub action that now runsevery morning and fires that off. Oh man.Wow.Which is likethat optimization makes no sense unlessyou're going hard.Yeah. Unless you're like a heavy, heavyuser, which a lot of people are. Let mejust sayhaving guests was our best idea. Josh,this is so good.Well, you are. This is so good. Oh, IPity every guest who has to come afteryou and live up to this. This isfantastic. Okay. See, I did not know about this, so Iknew that there was this time limit andevery now and then it would tellme, "Hey, this resets," because I haven’tgotten it to work well in Sonnet, forexample. Once I’m done with Opus, I’mlike, "Ah,we’re done." Like, I’ll just code by hand now or just copyand paste into,you know, something else.Um, but I did not know it was fivehours. So, that’s reallyhelpful. And like recentlyyou send your first message. Yeah.And you can’t track it anywhere. So, youhave to just know when it is.No. Yeah.Interesting.You have to vibe. You have to vibe. No,you’repretty much vibe limits.Vibe.I like how you saidthey're really taking this vibe thing alittle too far. Like, no, we do need afew things. Uh, could you help us out here, Claude?It'd be nice to have a few metricsto track.I like how you said you need to utilizeevery single token that you are payingfor.Oh, it’s literallyfive hours, five hours, five hours.It’s not even every token I’mpaying for; it’s every token that Icould have, right? I’m greedy with thesetokens. I want them all because ifClaude Code costs $2,000 a month, Iwould pay it without blinking.It wouldn’t even be like,“Where’s my credit card?” That would bepreloaded. If it costs $20,000 a month,I would seriously consider it as abusiness expense.Yeah. I was saying when you’ve gotI think there are definitely differentlevels to it, but it’s like once you’vebought in and you’re a business, this iswhat they talk about with like, whenyou’re building SaaS businesses, likethat’s really not that much to a big andif you’re big, like you’ve been around forover a decade. Like Ghost is nothing tosneeze at at all. Deal. So it's like, yeah, businesses likeyours, if it brings these kind ofoptimizations,well, you gotta think humansare expensive, right? Like you look at anengineer's salary for a month.Would you have hired insteadto get the same amount of work?Oh, at least.And they definitely didn't cost $20,000.No, and for the scale of what I'vegot going,at least three, maybe five, and I think itwould have taken much longer, and that'sno comment on the ability of humans todo good work. It is just the reality ofwhat'spossible with human fingers on akeyboard. Well, I think it just goes toI, there's something that I've likestarted kind of like understandinglately, which is just likeum, you more people does not mean youmove and you get more done. Like it doesin some sense. Like there are somethings that you need a team of likereally smart people to work together,more than one person to solve some types. Of problems, there's also a whole host of problems where havingmore people just makes everything moredifficult. Uh, like we, we, a small team ofvery highly skilled individuals, backed by tooling like this, can move wayfaster because you don't have theoverhead of security and all theenterprise workflows that go into bigbusinesses. Like, once you get enoughpeople, it's a people problem at that point where youhave to build in tooling, processes, and workflows that slow you down,that you don't need when you have asmall number of people, even if you mayhave a huge capacity for work. So, it'sjust like, man, this is just such, I think, there's a lot of people whoyou are on this end of this, you can'tsee it, but like way off thescreen over here, this is you, you're onthis end of the spectrum of havinggotten familiar with the tool,spent time doing it, whereas I thinkthe literal majority of everyone is likeeven over here where they've just triedit one or two times and thought,"Like, cool, like cute. Oh, that's cute."Like, that was neat how it did that. The thing is, they haven't actually tried to leverage it for a bigbusiness problem and see what all it could do.And also, to what you mentioned — one thing youmentioned is, I have problemswith this because it might just bethat you're in a position where you can do this a little better as anemployee.There's a feeling of urgencythat I can't get over the hump sometimesof being like, "Okay, if I put a bunchof effort into figuring out how to makethis work, I know it will make me fasterat some point." But until I really getto that point of figuring out theworkflow, the way to phrase things,and the way to communicate with it,it's like I'm under the pressure ofhaving to get things done. Whereas, maybe you'rethe CEO founder who can say, "Oh, it's okay if I take alittle while to get this done"because it was never going to get doneany other way. Otherwise, I wasn't goingto prioritize this. But like, I can spend some time knockingmy head against the wall with this tool.It's interesting. I think that says a lot to people. Like, itsaid a lot to me, which is just like, getout of the grind mindset, like theemployee mindset,because there's so much more that Icould potentially doif I put the effort in. Istill use it a lot. Iprobably use AI-assisted coding a lotmore than most average developers atthis point, but I still don’t comeanywhere close to the way you havetalked about it, for sure. Yeah, I think it took me maybe twodays of that kind of process toget used to clawed code in the firstplace. And I ditched it after one daybecause I was like, this is nonsense.This makes no sense. And it’s because Iwas using it in a terminal outsideof the code editor. I hadn’t run theslash IDE thing to connect to theeditor. I hadn’t got shift-enter workingbecause you have to do a littleconfig thing for that. I didn’t knowanything about slash commands that you can sort of save these snippets ofPrompts that you use again and again. And I was like, this is bad.This is just cursor but with no UI,and it's slow.That's the way I think.And then I just read one really good post.I can't remember whose it was, but there are a million of these on YouTube and on the internet nowof someone saying, "No, no, no, no, no."Load it up in the terminal inside VS Code or inside cursor.Run slash IDE to connect it so now it can actuallytalk back and forth to your files. Use set up a couple of slash commands,which means you’re not typing the sameprompt over and over again. You’re justusing shortcuts.Um, and then you realize,and this is just conspiracies, but cursoreveryone says it binds a lot ofmodels together, right? So you selectopus 4, but what you're getting is thecursor pipeline to opus 4, which mayinclude sub-models to handle your high level question, and theneventually, at some point, you get opus 4,but it’s filtered through manylayers of models. So if you want to raw dog opus,the only way to do it really is cloud code.You’re not gettingAny man in the middle, you're notgetting any likejiggory pokery cost saving that curs istrying to do to try and keep your buildup and their build down. You're justgetting the real thing. And then youstart to notice the difference in theoutput. Like, oh actually it's not just aUI difference. It's more capable. It'sroaming my codebase and findingsomething immediately and understandingeverything seemingly with very littleprocessing time. It's not going aroundin loops and creating the same bug andlike fixing it by creating another bugand going back and forth. It's notdoing that anymore. Yeah.And then the unlock is like, oh [ \_\_ ]this is a new thing.Yeah.So I, yeah, it's worth it.Okay,that leads me to, like,you've got— go ahead, Josh.Yeah, that question then, kind of like atwofold question. What would you say islike a good— like for people who areeither like listening or like, hey, Idon't know where to start with Claude? Code or two: how have you improved the things thatyou've improved, like, kind of like thoseslash commands?Stuff like that when you start a projector when you start a feature, whatever.What are like the thingsthat you're thinking of so that youdon't dive into, "Hey, these are thespecifics," but you also aren't toogeneral? You kind of had the there's thesweet spot. What do you think through?What's your process in order to findthat sweet spot? Yeah, good question. Um, so it'stwofold, I think, on the technologyside. So, um, any AI model, when itstarts to work on anything, thefirst thing it does is it goes and ittries to understand the context of whereit is, especially cloud code,right? Which is basically what directoryam I running in? What currently exists in this directory? So if you are starting a new project andyou open, you know, a new directory, you dumpClaude in there and you go, "Hey, I want tomake a Laravel app that blah blah blah,blah blah."It's going to go, "What's in here? Nothing."What do I know about Laravel? Something. All right, let's go. It throws a bunchof stuff in, and off you go.Maybe you get lucky, but in my experience,it's hard to get lucky when yourstarting point is a completely blankcanvas. Um, so what I like to do from atechnology standpoint is I like to giveit a set of, if you will, lines on thecanvas to color within. So I'll do myLaravel new, uh, you know, with Work OS,with Inertia, with React. Um, I'll set upall my kind of preferred packagesalready there. I'll probablystub out a README or a CLAUSE.md withsome kind of project overview that I'vethought about, or I've already kind ofchatted about with an 03 model thatsort of lays out what my intentions are. And they'll be high level and vague, likewe're making an app for pet sitters toschedule whatever it is. Once I'vekind of got the hello world version ofit going, which is not much work, right?We're talking half an hour. Um,especially with an incredible frameworklike Laravel, where it's like,yeah, exactly.One command and you're basically 80% ofthe way there.Then I knowThis is the bones of a starting point. If I was going to work on it myself,this would be a good starting point. No.Then I go to Claude and I say, "Readthe read me. This is what we'reworking on. Have a look at thecodebase. You'll see that it's a Laravelapp. Here's the first thing we want todo: maybe describe a feature." And thatlittle setup alone, I found, putsyou 10 steps ahead of most people whomight start in an empty directory andjust ask for something and then get avery nondeterministic output, whichsimply means not repeatable. Every timeyou do it, it'll be different.That is not what they wanted, butthey didn't say what they wanted, really.So, if you can kind of give it someindication of what you want, I thinkyour results are a lot more reliableand less deterministic, or sorry, moredeterministic. They're moreconsistent if you were to do them againand again.So, that helps a lot,man. Okay. Well, I literally have aproject for Laron that I've been working on. I've shared a bit of it here on aprevious episode, but I've got like twoor three features left in it, and I'mgoing to try this playbook. I'm going todive into cloud code. I've already got aLaravel, you know, React inertia app. I'm goingto add a little cloud MD to give it somecolor, which I think I could doeven easier now that there's more codeto kind of reference when talking through it.And I'm going to report back. I'll give you one more thing to try. Itweeted this a couple of days ago, butmaybe some people didn't see it. Theone other hack that works really well isyou ask Claude to plan something.You say, "I want to build thisfeature. I want you to plan it out.Write the plan in a markdown file."Claude's got a built-in planning tool.Don't worry about that. Just tellit to write the plan in a markdown file. It's going to write you a plan likehere's how I'm going to implementthis feature. It's going to have jobs.It's going to have a queue. Stuff. You say, "Thanks, Claude," andyou kill it. Just close that terminal.Claude's gone. Save the markdown file.Open a new Claude and ask that Claude toreview the previous Claude's plan. AndClaude number two, almost without fail,especially if you ask it to call outany signs of overengineering, willdramatically simplify that plan. It willtake the bones of it. It will understandwhat the point is, but it will scope itway downand turn it into something that isactually going to be a good result. And then you kill that Claude. ThatClaude's gone and Claude number three getsto work on it. Claude number threelooks at the plan withoutthe prior context of the bad planbecause you don't really want that inits memory.Right. Right. And then you start going, um, and peopleyou can get different results by havingdifferent AI models review the sameplan, that kind of stuff. But you don'thave to. A lot of people spend a bunch oftime going back and forth on planningwith Claude and trying to give itFeedback. You're not that smart. You know who'sreally smart? Another AI model. Getanother AI model to give feedback.See how far they can go before you getinvolved.That's really fun. That's royal you, by the way. You,the human, not you.Right. Right. Right. We love, we love theroyal we. The royal you. There'sthat cracked me up when I saw thatstatus. The take them behind the shed.Yeah.Imagery just cracked me up. Oh mygoodness.That's awesome.Okay, that's awesome. All right. Well, Iwill report backbased on trying these facts in the nextshow probably because I've got to getthese last few features done in thisapp. So, my Laravel will talk. Are you gonna be you're going to bedemoing this on stage?Yeah. Live demo.I got to see it.You are going to be there or you're not?Yeah. Yeah. Yeah. There we go.It's going to be fun. I've got to say, it requires audience participationand I'll share what that is going to be when I want it tobe a bit of a surprise. So,this is giving me flashbacks to React Miami already.Yeah. Yeah. I'm a little bit nervous.Awesome. I think we're almost out oftime, so why don't we wrap this up? Um, how aboutJohn, you start us off. What is it? So,now that you've told us everything thatyou're working on, what is the thing that you're looking forward tolike next week? Either something you'reworking on or just like something thatyou're thinking, "Hey, I need to try this." Maybe it's a tool,or a project idea.There's a guy, sorry guys, this is all just AI non-stop, but that's kind ofI love AI.Bring on guests because we care aboutwhat they're into. And if that's whatyou're into, we want to hear about it.I found a guy on Reddit. This isUnverified information. This is fromReddit. Oh, okay.He claims to be running fourinstances of claw code in fourterminals.Okay,in VS Code. He assigns each one of thema role. So, one is an architect, one isa planner, one is an implementer, and one is QA, ifI remember correctly. And that's notthat novel. People have done thatbefore. And you kind of, you know, onedoes one thing, one does the other. Butno, this dude's done somethinginteresting. He gets all four of themwriting to one shared work file tocommunicate with each other. So theyleave notes in the same file, noting wholast updated it. Then the other ones arereading that file and following theinstructions that it gave. So all fourare working on the same problem at thesame time in parallel andintercommunicating with each otherthrough the shared medium of a markdownfile. And, um, Reddit bro 759, not hisreal name, claims that this is workingincredibly well. So I'm excited to, I'mreally excited to give that a try next. Week and see if it's true.Twitter post results on Twitter. Whetherit works or doesn't work, I want toYeah, that gives me real, like both old school and futuristic vibes atthe same time where it's like, "Hey,let's have like two telemarketers talkto each other at the same time on aphone,but also very futuristic too, whereit's like, hey, why don't I just have like all my AIs that arerunning everything just talk to eachother. Make suresometimes there's nothing better thanjust mixing a bit of new technology withlike an old format. Some thingshave worked for a while, and there's areason. I'm gonna try that with my projects. I'll jump in next. Like II've been working on learning, um, likeExpo. So like as a front-endOh yeah.Expo app with a Laravel API. And rightnow it's kind of been like I have oneClaude in a mono repo that has the APIin the back end. But now that yousaid that, I want to try like running atjust start with two instances. OneThat's in the API, one that's in the Expo app, and they're communicating inthe same file. That would beinteresting to try. I also do want tomake a video, a little shout-out for myself, Iguess. My editor AJ said this betweenthis and Josh's upcoming video about AItools that I'm releasing tomorrow. I'm convinced to try Cloud Code. So, Ihave a video that's coming out abouthow I use Cloud Code andeverything else, but now I want to makea video about your flow, John, and justsay how far deep can we go?Oh, it's like this is even betterbecause now you have an originalvideo and you can have abetter video right after. It'slike I made Claude kill itself eighttimes kind of thing. It's like one of my favorite memes, theObama meme where it's like Obama'sgiving himself athumbs-up. That was Josh just now. Just likeimagine just like a thumbnail ofClaude, you know, shooting itselfbehind the shed, you know, butit's like eight versions of that justLike recursiveturtles all the way down. How far? What is the sweet spot? Five,six, seven times just like killing theworld needs to know. Josh, find out forus. Oh yeah. What about you, Will? Um, I have got to wrap up this talk, athis Laracon talk. I had a great chat withChris Sev and Hank this last week andthey did a really good job clarifyingsome high-level ideas, likethis talk is great, buthere are some ways that you could kind oftighten it for the audience that's goingto be there. Um, and I was like, this isfantastic feedback. Uh, and then I'mgoing to meet with Chris again next weekto just kind of go back over thosethings that we talked about andreally tighten it in. Laracon is going to be so good this yearbecause Chris Sev is doing thiswith every single speaker and JohnRudolph Drexler, which I can't help butsay is like an entire name. But John, hedid Laracon EU and his talk onproduct management was so good,and I was like, man, his talks are so good.He's just a really good speaker. He said that he posted about histalk withthat it was like, "Oh, he just madeit so much better. My talk’s so muchbetter." And I'm just like, if even likeone of the better speakers in thegroup's talk is getting better becauseof this process, and every speaker isgetting to go through this, it’sjust going to be such a solid conference. So,yeah, I love that. All right, John. Where can people find you to learn more aboutwhat you’re doing, or just followyou for more Claude code kind ofstuff? Do you have a Claude code likemaybe like an affiliate link thatyou can plug? You might make a bunch ofmoney right now.I wish. Don't give me money. Just giveme tokens. Umthat’s what they need to do. Lower your refresh limit. Likeit was soit got bumped down to four hours nowbecause you recommend so many.That could be both a Black Mirrorepisode and a great affiliate program. Um,I'm John O'Nolan. I'm probably in all the places, whereveryou are. I probablyuse Twitter, whatever it's calledthese days, particularly forClaude stuff and posting. Um, butyou can find me almost anywhere ifyou're curious or want to say hi. Perfect.Awesome.Well, this was an awesome show. Thankyou for coming on, John.Yeah, thanks. Thanks somuch for saying yes.You've all been great as always.Loved having you all here.I'll see you both soon.See you next time., Consistency Without Complacency Ft. John O'Nolan, Josh Cirre, 01k04zavzg4xq1m3829ycyp9ze.