Episode of Every's "AI & I" podcast in which Dan Shipper interviews entrepreneur Andrew Wilkinson, cofounder of Tiny — a holding company whose portfolio includes AeroPress and Dribbble — about how he uses Anthropic's Claude Opus 4.5 with Claude Code in his work and life. Wilkinson explains why Opus 4.5 makes vibe coding viable after his earlier Replit-built apps proved buggy, and walks through automations he has built: a custom email client, an AI relationship counselor trained on his relationship, AI meeting notes, and a system that texts him daily outfit recommendations. He also shares Claude Code tips and workflows, prompt-writing practices (including not injecting your own opinions into prompts), and how this hands-on exploration is changing how he evaluates and buys software companies at Tiny. Chapters: 00:00:00 - Start 00:01:07 - Introduction 00:02:48 - Why Opus 4.5 feels like the iPhone moment for vibe coding 00:08:31 - Why designers have a unique advantage with AI 00:14:10 - How Wilkinson built a custom email client with Claude Code 00:18:13 - An AI trained on your relationship that predicts your fights 00:30:40 - Using AI meeting notes to make your life better 00:35:11 - Don't inject your opinion into prompts 00:40:21 - Wilkinson's Claude Code tips and workflows 00:47:59 - Your personal stylist is a prompt away 00:53:17 - How AI is changing the way Wilkinson invests in software Resources referenced in the episode: the book Made to Stick (which Wilkinson uses in prompts when writing copy with AI) and Every's compound engineering plugin (https://github.com/EveryInc/compound-engineering-plugin). # Channel Presentation Observations **Video:** "How Andrew Wilkinson Uses Opus 4.5 in His Work and Life" (ID: 01a0f15a-b165-77c6-ae3b-2f9012e51438) ## Explicit Self-Description Evidence - **Channel name stated in outro:** The host reads a scripted outro referencing the show as **"AI and I"**: > "You absolutely, positively have to smash that like button and subscribe to AI and I." - **Host identified by name:** The outro names the host as **Dan Shipper**: > "It's a journey into the future with Dan Shipper as the captain of the spaceship." - **Claimed topical scope:** The outro explicitly frames the channel around ChatGPT/AI knowledge: > "pure unadulterated knowledge bombs about chat GPT." - **Stated call to action:** Standard subscribe/like solicitation ("hit like, smash subscribe") demonstrates a public-facing subscriber-growth posture. ## Observable Presentation Patterns - **Format:** Interview podcast between host (Dan) and a guest (Andrew Wilkinson), with casual rapport: > "Andrew, welcome to the show. Great to be here, dude." - **Intended audience demonstrated in-address:** The host repeatedly signals an audience of **listeners who are technical, AI-curious, or own software businesses**: > "I think probably a lot of people listening own software businesses or are five or 10 years into software businesses that are not necessarily AI native." > "if you're from Superhuman and you're listening to this and I get this wrong, please correct me in the comments" - **Audience addressed as AI-interested but mixed expertise:** References to "people who are really into tech and AI, that just have not felt it yet" position the show as speaking to a spectrum from beginners to frontier users. - **Host's own credentials shown in-dialogue:** The host mentions teaching a course and an affiliated engineering product: > "I taught this course maybe like three weeks ago... called cloud code for beginners" > "Have you tried our compound engineering plugin?" — indicating the channel is connected to a company building AI engineering tooling. ## Sponsorship Evidence - **Mid-roll ad present:** A **Framer** ad read appears mid-episode ("framer.com/dan for 30% off"), demonstrating commercial sponsorship integrated into the show's presentation. ## Claims vs. Demonstrated Distinction - **Creator claim:** The show is "the epitome of awesomeness" and a source of AI "knowledge bombs" — promotional self-description, not verifiable from this material. - **Demonstrated:** The episode delivers a long-form guest interview on AI tools (Claude Code, Opus 4.5, Lindy, Granola), with live screen-shares and product demos — consistent with an AI-focused interview format. - **No channel-wide conclusions inferred** (e.g., upload cadence, overall guest roster, or channel identity beyond what is stated/shown here). # Delivery Observation: "How Andrew Wilkinson Uses Opus 4.5 in His Work and Life" ## Format Overview Two-speaker conversational podcast (host Dan Shipper interviewing guest Andrew Wilkinson), with a pre-roll teaser of the guest's most quotable moments, a mid-episode ad read for Framer, and a hyperbolic AI-generated-sounding outro. Delivery is casual, fast, enthusiastic, and anecdote-heavy. ## Tone - **Consistently high-energy and superlative.** Both speakers repeatedly use intensifiers: "blown away" (appears at least three times: [0], [195], [208], [242]), "crazy" (recurs at [11], [298], [561], [1142], [2054]), "unreal" [147], "astounding" [883]. - **Playful, buddy-style register.** Frequent "dude" from both speakers ("Great to be here, dude" [69]; "dude, I'm just having so much fun building" [298]; "Dude, just use Google" [21]). - **Self-deprecating humor and profanity** ("my butthole relaxed a little bit" [3614]; "What the fuck?" [2219]) keeps the tone informal. ## Recurring Phrases and Verbal Habits - **"It's like..." analogy framing** is the dominant explanatory device. Examples: - "It's almost like everyone is using the yellow pages and you're like, 'Dude, just use Google.'" [21]/[748] - "like a palm trio, right? Not an iPhone" [170] - The I Love Lucy chocolate factory comparison for email overwhelm [948] - The pizza restaurant analogy for software margins [3322–3377] - "every piece of software is just cloud code in a trench coat" [2682] - **Repeated false starts and self-restarts**: "you know, I I've," "I I basically," "it it's" occur throughout (e.g., [1112], [2107], [2841]). - **"And so" as a connective crutch**, especially in Wilkinson's longer narratives ([219], [254], [970]). - **Filler "like" density is very high** across both speakers, marking a spontaneous, unscripted delivery. ## Teaching / Explanation Habits - **Concrete dollar figures to make points vivid**: "$40 a day" [11]/[561], "$100,000 a month payroll" [208], "$300 bucks an hour" [1729], "5K for a session" [2017], "$10 share price and now it's at 35" [3690]. - **Screen-share demonstration** mid-episode ("Let me screen share... Can you see this?" [1116–1124]) with step-by-step walkthrough of the Deep Personality app — a show-don't-tell teaching move. - **Quoting external authorities**: an Upton Sinclair quote [345–356], Made to Stick by Chip and Dan Heath [1199–1233], references to Jason Fried [628]. Used to legitimize points rather than as structured instruction. - **Personal anecdote as evidence**: nearly every claim is backed with a first-person story (the gaslighting contractor text [2086–2117], the Starbucks gift certificate false-fraud audit [2472–2494], the beard-grooming prompt [100–147]). ## Communication Patterns - **Collaborative interruption and building**: the host interrupts to redirect ("Well, let me just stop you right there" [1165]) and the guest accommodates — pattern typical of a friendly, familiar pair ("we've spent like a good amount of time over the last like six months" [69]). - **Question–reflection loops**: host asks a reflective question ("how has this changed your whole view of the world?" [3197]) and the guest answers in structured "two feelings" format ("I carry two different feelings" [3275]). - **Balanced hedging**: the guest pairs enthusiasm with explicit caveats — "I probably wouldn't make a hire and fire decision entirely on that" [2372]; "you got to be really careful with this stuff" [2496] — a consistent pattern of enthusiasm tempered with caution. ## Uncertainty Notes - Speaker attribution in the transcript is sometimes ambiguous (the `>>` markers are inconsistently placed), so a few quotes above could belong to either speaker; attribution is stated only where context makes it clear. - The outro [3734–3768] is tonally distinct from the rest (likely a scripted/comedic bit); it should not be treated as representative of the conversational delivery. # Observation Record: "How Andrew Wilkinson Uses Opus 4.5 in His Work and Life" **Speakers:** Dan Shipper (host, Every) and Andrew Wilkinson (guest, investor/founder, Tiny). Speaker identification is based on conversational structure and contextual references (e.g., Dan named directly in dialogue, Andrew's companies referenced). --- ## 1. Andrew Wilkinson on Claude Code with Opus 4.5 ### Overall assessment - **Strong enthusiasm:** "Over the last two weeks, I've been playing with Claude Code with 4.5 and I am just blown away." - Claims it feels like **"$100,000 a month payroll of engineers working for me 24/7"** and later, **"30 free employees... paying them like $40 a day."** - Reports previously abandoned "vibe coding" ~1.5–2 years ago after being impressed but frustrated with Replit ("seeing the future but it was using like a Palm Treo, not an iPhone" — impaired output, bug loops), pivoting instead to Lindy and traditional AI agents. ### Thesis on tooling - All tools will converge into **"a single command line"** and eventually **"a single voice"**; Claude Code is used for almost everything because it is extensible, can access the system, configure things, and run MCPs. ### Comparison to competitors - Tried **Gemini 3** to build the first version of his app: "it is incredible, but the tooling is just not as good... it doesn't feel right to use." - Switched to Claude Code, which was **"so much better at understanding my intent and doing long-run tasks"** — claims it could work for four hours on enough instructions. - Concedes (Dan's framing, Andrew agreeing pattern): other frontier models are sometimes useful, but Andrew's stated experience is Claude Code's ergonomics win. --- ## 2. Dan Shipper on Opus 4.5 and Programmers ### The moment - Believes most of the world, even tech-adjacent people, "have not felt it yet" and that a small group "just get it." - Claims: "It feels like we've achieved AGI or ASI in programming and it's just the first bastion of that." ### Two archetypes of programmers 1. Those who **lean in**, accept they no longer program themselves, and become "an architect or an engineer." 2. Those in **denial**, which Dan illustrates with an Upton Sinclair quote: *"It's never expect a man to understand something that his paycheck depends on him not understanding."* ### Prediction - "I think Dario is correct. There will be no programmers, not in the way that we understand them, in two or three years." ### On Anthropic specifically - Claims Anthropic "seems to have figured out how to simultaneously max the engineering brain... and also its human empathy," unlike models that "as they get better at programming they've got worse at other things." - **Qualification preserved:** Dan notes he sometimes uses Gemini 3 or ChatGPT 5.2 "for a really hard bug fix that Opus 4.5 can't get." - Admits compulsive usage: "I have to block Claude because I'm spending more time on Claude than maybe I would on X... one more prompt, bro." --- ## 3. Designers as the Empowered Group (Dan) - Dan's **2026 prediction**: designers are "the unsung group" most empowered by Opus 4.5, because they can now take work end-to-end without depending on engineers; cites his creative director Lucas as an example, and places Andrew in this category. **Andrew's agreement/extension:** With chunking (wireframe → Photoshop mockup → copywriting), even copy is "perfect" because he mentally runs algorithms from books (Jason Fried, *Made to Stick*, "Hooked"); he historically delegated software to avoid coordination problems and one-on-ones, and now realizes he can "build automated systems to do all of these things." --- ## 4. Andrew's Built Tools (Demonstrated/Described) ### "Deep Personality" (relationship/personality app) - Prompted Claude: "If you were a therapist and wanted the best picture of a couple, what would you want?" → it listed ~20 clinically validated tests and inventories; he built a combined interface. - He and his girlfriend each spent **40 minutes**; output predicted "every single fight that we have." - Includes a 45-page analysis, attachment style, dark-triad stats, ChatGPT prompts for a personal AI, and a "relationship blueprint" for couples. - Headline ("In 25 minutes, you'll understand yourself better than a therapist would after 10 sessions") was generated by prompting with Jason Fried / Chip & Dan Heath (*Made to Stick*) principles. - Claims the site would have cost **$5K–$25K** with a designer; he built it "in two hours... prompting while building other stuff." - Plans to release publicly; unsure of vertical (romance, corporate teamwork, hiring?). - **Tension preserved:** Dan asks for third-party perspective prompts ("not objective, but another source of information"); Andrew declines to show couple results as private. ### Custom email triager - Gets 200–300 emails/day; previously required one or two assistants plus himself. Earlier Lindy automation routed mail, cut load ~50%, sent "choose your own adventure" reply emails — but hit edge cases. - New Claude Code app: finds unarchived response-required emails, ranks by priority, offers multiple-choice or Q&A mode, drafts responses. Usable daily within a week. - Dan's comparison: "Superhuman spent like five years in beta" (with an explicit caveat asking Superhuman to correct him if wrong). ### Personal stylist automation - Daily 7 a.m. job: checks Victoria weather → generates four outfits from a Google Sheet wardrobe CSV (photos processed by Claude) → renders via Nanobanana with reference images of him → texts via Twilio. Previously paid a human stylist "a couple thousand bucks" with poor fit results. ### Kids' school email automation - Ingests school emails, texts actionable reminders (packed lunch, field trip form links), and auto-populates a parenting calendar. ### Lindy "red flag" meeting agent - Records meetings and texts him post-meeting if manipulation/narcissism red flags reach a **"very high bar"** (fires once every 2–3 months). Anecdote: contractor used "gaslighting... reframing"; agent flagged it minutes later. ### Super Whisper "good boss" prompt - Phone action button launches voice-to-text; a prompt converts harsh dictation into "what a good boss would say." --- ## 5. Dan Shipper's Tools and Practices - **Granola** for meeting notes; used its AI assistant to draft an employee handbook from his last 25 meetings ("pretty good... I probably wouldn't just put it into every employee's hand immediately"). - Values it for **pre-game and post-game analysis** of difficult leadership conversations; claims it substitutes partly for executive coaches ("5K for a session"). - **Limitless pendant:** recommends to Andrew; uses it for perfect recall (ADHD-related memory issues; contractor anecdote) and feeds recordings into AI for interpersonal analysis. - Taught a one-day **"Claude Code for beginners"** course (~3 weeks prior): non-technical people built a full expense-tracker app from a prompt; participants customized it (no-spend streak calendars, savings goals). - Describes **agent-native architecture** ("every piece of software is just Claude Code in a trench coat"): every feature is a prompt; user-customized prompts can be promoted into product features; Every's 2026 focus. - Every's **compound engineering plugin**: spawns ~5 subagents to research best practices, the codebase, and open-source examples into a plan with technical requirements/acceptance criteria; overkill early, useful in large codebases. - Uses **ChatGPT's Atlas browser** ("I use it every day" — Andrew agrees) so he "never touch[es] a settings dashboard." --- ## 6. Epistemics and "AI Hygiene" (Shared, with Tension) - **Dan's caution:** AI sometimes "subtly reinforces my interpretation" which later breaks; when challenged, it flips to agree — raising the question of how to hold AI output skillfully with limited information. - **Andrew's safeguards:** - Don't inject your own opinion into the prompt; set a high bar for flags. - Analogy to body language: patterns matter (his dad always crosses his arms — can't read that as closed-off). - Wouldn't make hire/fire decisions entirely on AI analysis; it "usually confirms a feeling" of cognitive dissonance. - **Dan's limitation note:** transcripts lose tone, pauses, facial expressions; mentions **Hume**'s API for real-time emotion detection. - **Andrew's cautionary anecdote:** prompted a company audit "as if I was suspicious"; AI reported "$90,000 of fraud" which was actually employees buying Starbucks coffee. "You gotta be really careful with this stuff." --- ## 7. Therapeutic/Psychological Uses - **Andrew:** "It's not the same thing as therapy, but the best way to get that kind of thing... that has ever been invented." His therapist is open to Claude-derived insights; the relationship GPT helped resolve a fight by articulating a childhood trigger ("the moment I said that she softened"). - **Dan:** acknowledges the "psychosis" discourse — "if you have real mental and emotional problems, ChatGPT should not be encouraging your delusions for sure" — while arguing good therapists are expensive (great ones ~$300/hour) and price doesn't guarantee quality. - Andrew compares the psychosis panic to early iPhone criticism (no App Store) — likely to be resolved. - Both want larger context windows (5–10M tokens) to incorporate a year of conversations; Andrew notes current models "lose the plot too quickly." --- ## 8. Software Business Thesis (Andrew) ### Moats collapsing - "The moat for software used to be... programming is hard to learn... very expensive to hire [programmers] and now it's basically free. And so your moat has to come from something else" — brand, distribution, hardware lock-in — "I just don't see them being very good businesses in the long term." - **Pizza analogy:** if a $10,000 machine makes the world's best pizza, pizza restaurants don't vanish, but **margins compress** (10–15% → 1–2%); consumers benefit, owners struggle. - Calorie-tracker example: a kid vibe-coding a million-dollar/month app "picked up pennies in front of a steamroller" — copied within three months. - **Tiny has "really slowed down on buying technology companies"**; most software businesses are "thin wrappers" on a database or AI call. - Analogy to water-powered factories: convert to electricity or die; software has the advantage of being able to replace human work internally. ### Macro worry (short-term, not long-term) - Not worried about 20 years out; worried about **"another great depression"**: white-collar layoffs (lawyers, programmers, maybe doctors) pushing people into blue-collar work (e.g., HVAC), which then also sees margin collapse from oversupply. - Personal coping: education, preparation, then not worrying — parallel to Pacific Northwest earthquake preparedness (basement supplies, Starlink). - "I think a lot of people should be pretty freaked out." ### Investment positioning - Believes **"owning compute is probably one of the only ways to own a toll bridge to the future."** - Cites a January investment in **Iris Energy ("Iron")** — a Bitcoin miner pivoting to AI data centers — bought at ~$10/share, now ~$35, previously $70. Caveats that retail investors must distinguish "which ones are real and which ones are kind of made up." - (Also mentions earlier investments in frontier models and data centers.) **Dan's response to pizza analogy:** appreciates it, adds that "what constitutes great pizza is up for debate" — endless complexity in the economy means pizza restaurants survive, but the landscape changes. --- ## 9. Current Gaps and Future of Software - **Andrew's wish:** Anthropic launches its own database/hosting so everything runs in one sandbox (currently configuring Supabase and Vercel manually); Dan predicts "they probably will." - **Dan on the current gap:** doesn't think the industry knows the gap yet; the bottleneck is **reviewing/verifying generated code** — "the bar was can it get to the human without any errors and that's basically solved... the next bar is does it actually work well? And that's still an open question." - Dan's workflow: uses the **Claude mobile app's code section** during breakfast to spin up ~15 parallel fixes/features via pull requests; Andrew says he hasn't tried that but has cron-job sync across his Macs so Claude Code resumes state on any machine. - **Andrew's open question:** "Why do we need software at all eventually?... I really wonder what software continues to exist versus just gets built on the fly for you." Dan echoes: what should be software vs. "just something ChatGPT does" is "another really interesting open question." - Andrew on **skills**: building a copywriting skill encoding his preferred books/tone; on **MCPs**: "still early and some of the integrations suck and they're too slow" but will become a "tentacle monster that touches all of your software." --- ## 10. Personal Life Automation (Andrew, opening segment) - Used ChatGPT Pro to "looksmax" his face, which told him to grow a beard; he now tunes mustache-vs-beard length, and it suggested tinted gel for a patch that doesn't grow. - Dan parallels this with AI-suggested clothing colors (greens/browns for his hair) but hadn't tried looksmaxing. - Andrew: "It's pretty crazy how much of my life is now AI automated." --- ## Preserved Contradictions & Qualifications | Topic | Tension | |---|---| | Opus 4.5 superiority | Both praise it, yet both admit using Gemini 3 / ChatGPT 5.2 for certain hard problems | | AI for judgment | Both value it for interpersonal analysis while flagging sycophancy, prompt-induced conclusions, and the Starbucks "fraud" failure | | Psychology AI | Enthusiasm vs. Dan's explicit caveat about delusions and "psychosis" discourse | | Software businesses | Andrew the builder is energized; Andrew the buyer (Tiny) has slowed software acquisitions | | Emotion | Andrew admits he was "pretty freaked out" a year ago and "a lot of people should be pretty freaked out," while currently enjoying building — "my butthole relaxed a little bit" | **Note:** The transcript contains a mid-roll ad read for Framer and a closing promotional outro; these are sponsor/promotional content, not substantive positions of the speakers. # Speaker Attribution & Provenance Record **Video:** How Andrew Wilkinson Uses Opus 4.5 in His Work and Life **Video ID:** `01a0f15a-b165-77c6-ae3b-2f9012e51438` ## Speakers & Roles - **Dan Shipper** — **Host / Podcast creator**. Identified in closing promotional segment ("Dan Shipper as the captain of the spaceship"); hosts the show "AI and I," interviews the guest, shares his own work (Every company, "Cloud Code for beginners" course, "compound engineering plugin," Granola usage). - **Andrew Wilkinson** — **Guest**. Named at welcome ("Andrew, welcome to the show"); discusses his work, Tiny (software business holdings), personal automations, and his app "Deep Personality." Note: dialogue markers (`>>`) distinguish the two voices; role assignments are explicit from greeting context and closing. ## Promotional Segments (Flagged) - **Framer sponsor read** (mid-episode): website builder promotion; discount URL `framer.com/dan` for 30% off Framer Pro annual plan; "Rules and restrictions may apply." - **Channel subscription promotion** (closing): "smash that like button and subscribe to AI and I" — attributed to an announcer voice, ends "Dan, I'm absolutely hopelessly in love with you." ## Referenced Sources & Third-Party Claims - **Products/models discussed:** Claude Code / Opus 4.5, Gemini 3, ChatGPT 5.2, Replit, Lindy, Superhuman, Framer, Granola, Otter, Super Whisper, Supabase, Vercel, Hume (emotion-detection API), Twilio, Nanobanana, ChatGPT's Atlas browser, Limitless pendant, Starlink. - **Books cited by guest:** *Made to Stick* (Chip Heath & Dan Heath); writing style of Jason Fried; "Optin Sinclair quote" (as transcribed) — "never expect a man to understand something his paycheck depends on him not understanding." - **Third-party claims (uncorroborated in-source):** - Guest claims Superhuman "spent like five years in beta," explicitly inviting correction "in the comments." - Guest attributes prediction "there will be no programmers... in two or three years" to "Daario" (presumably Dario Amodei; name as transcribed). - Guest claims a tweet: "AI enables me to move at the speed of my ADHD" (unattributed, "a guy"). - Guest's investment claim: Iron (Bitcoin miner pivoted to AI data centers); invested at ~$10/share, now $35, recently $70. - **Guest's own products/projects:** "Deep Personality" app (planned public release), email triager, personal stylist automation, Lindy red-flag meeting agent, "good boss" Super Whisper prompt. ## Availability & Uncertainty - **No curated metadata supplied**; roles derived from transcript only. - Speaker turns occasionally unmarked in transcript; attribution inferred from conversational flow where markers absent — flagged as inference, not explicit. - **Comment availability:** implied open (host invites Superhuman to "correct me in the comments"); no explicit metadata on comment status. Over the last two weeks, I've been playing with Cloud Code with 4.5 and I am just blown away. I literally feels like I have like 30 free employees and they're just working 24/7 and I'm paying them like $40 a day. It's crazy. It's almost like everyone is using the yellow pages and you're like, "Dude, just use Google." I built this tool that tells you the problems in your relationship and how to solve them. And it was scary. I'm realizing like I have a software company here that I just dreamed up out of nowhere. There will be no programmers, not in the way that we understand them in 2 or 3 years. The moat for software used to be it's very expensive to hire them and now it's basically free. And so your moat has to come from something else because I just don't see them being very good businesses in the long term. \>\> \[music\] \>\> Andrew, welcome to the show. \>\> Great to be here, dude. Big fan. \>\> Good to good to see you again. Um I we've spent like a good amount of time over the last like six months or so, which has been really nice. you've been uh kind enough to invite me to a couple different things that are um were really just really fun events. So, I'm really glad that we get to take this relationship to the pod. Um I love the facial hair and the haircut. You look great. \>\> Thanks, man. Thank you. I've been growing it out. You know why I grew out facial hair? \>\> Why? I had chat GBT do an analysis. I I put my face into Pro and I said, "I want you to look looks smack me." And it told me to grow a beard. And now I'm refining the mustache length versus the beard length and dialing it all in. Wow. \>\> But uh yeah, it's it's uh it's pretty crazy how much of my life is now AI automated. \>\> That's really interesting. I got to I got to do that. I've done that for um coloring. So like, you know, I'm wearing these greens and these browns because apparently that's like good for my hair color which has a little bit of red and brown whatever in it. Um but I have not asked it to look looks max me like that. So uh I'm down, \>\> dude. I mean, for beard grooming alone, like I have a little blank spot where I can't grow. And so it was like, "Oh, take a tiny amount of tinted gel and just put it over there." And I was like, "Oh my god, I don't have the spot anymore." And then it was like, "Oh, you want to cut your beard in this particular way." And it's it's unreal. I'm so I'm so excited to tell you about all the weird little automations I've built because most of them have been in my personal life and then there's a bunch of other ones that are uh that are work. But man, it's so fun. \>\> What's the next one on your mind? like what are you what do you want to tell me about? What are you working on? What are you thinking about? \>\> Um so I would say about a year and a half ago, two years ago, I got obsessed with Replet and it was like seeing the future but it was using like a palm trio, right? Not an iPhone. And I found that it was I mean it was so exciting to just be able to say what you want and manifest it and have something come out that worked but it was always impaired in some way or it didn't work or it' be in a bug loop. And so I kind of put vibe coding aside and mostly focused on Lindy and more kind of traditional AI agents and stuff over the last year and a half. But over the last two weeks, I've been playing with cloud code with 4.5 and I am just blown away. I feel like I have a $100,000 a month payroll of engineers working for me 247. And it's really confirming like I have this whole thesis that basically um all the tools are going to go away over time and it'll increasingly just be a single command line and then eventually it' be a single voice that you interact with and cloud code because it's so extensible and can access your system and configure things and run MCPs. I'm just finding I'm using it for almost everything now uh work-wise and and I'm just blown away. And so the thing I've been hacking on is um me and my girlfriend were talking about how nice it would be to have like a a GPT that's trained on our relationship. And so I asked I said, "Look, if you were a therapist and you wanted the best picture of a couple, what would you want?" And so it listed this huge like it was like 20 different um tests, like multiple choice tests and inventories. And so I just went to Claude and I built this tool that basically I called it deep personality and it does a deep analysis on your personality and your relationship and then it tells you the problems in your relationship and how to solve them. And it was scary. Like we sat down and we read it out together and we were laughing because it predicted every single fight that we have in our relationship. So I I've just been loving that. I'm actually releasing that to the public uh really soon. Um, so I'm excited to do that. But dude, I'm just having so much fun building. \>\> That's awesome. Look, okay, so I want to talk about this thing that you built, but let's back up to just the Opus 45 thing. It is such a moment and there are some people like you that just get it and just know that we've just reached this new level. And then I think think there's like most of the world, even people who are really into tech and AI, that just have not felt it yet. And it's so interesting how long it takes for people to catch up. It feels like um we've we've achieved AGI or ASI in programming and it's just the first bastion of that. And so I think you get two different archetypes of programmers. Those that lean into it and love it and realize this that they've given up on the fact that they need to program themselves and now they're just kind of an architect or an engineer and they're excited. And then the other archetype I meet is um there's a good Optin Sinclair quote uh that hits on this. It's never expect a man to understand something that is paycheck depends on him not understanding. And I feel like a lot of people are just kind of in denial. But I think um Daario is correct. There will be no programmers, you know, not in the way that we understand them in two or three years. It's so interesting how fast it happened. And also I think even the there's there's this interesting thing happening with um Gemini 3 is out. Um you know Chadbt 5.2 is out and those models are good and they're really smart and sometimes I actually use them for you know like a really hard bug fix that um that Opus45 can't get. But it's it's wild how uh anthropic seemed seems to have figured out how to simultaneously max the like engineering brain of this thing and also its human empathy sort of like common sensess like I think a lot of the models as they get better at programming they've got worse at other things and that has held them back from being this sort of general purpose engineer and anthropic just figured it out in a way that it open is just this sort of ergonomic humanlike thing that also so doesn't get lost in bugs and just builds and builds and builds and builds. And I've been noticing for myself, I'm curious if you feel this, I have to like I have to block Claude because I'm spending more time on Claude than maybe I would on X or like whatever. Like I'm I'm uh I'm not doing my other stuff because I can't stop being like one one more prompt, bro. One more prompt. \>\> 100%. Dude, it's funny you mentioned Gemini 3 because, you know, there's this hilarious cycle that you're very familiar with. Every quarter, uh, you look on X and suddenly all the all the Frontier models start releasing the new version and it's like mind-b blown, you know, um, emoji, they cooked with this one, blah blah blah. And, you know, I've seen a lot of them and and don't get me wrong, they've been getting better. Um, but I tried Gemini 3 and I used I used it to build the first version of this app and it is incredible, but the tooling is just not as good. It doesn't sound right. It doesn't feel right to use. And I ultimately just switched to Cloud Code. And Claude Code was so much better at understanding my intent and doing long run tasks. Like I feel like it can do if I gave it enough instructions, it could work for four hours and keep going. And dude, I'm like I'm with you. I I've never been this excited about work. Um my when I think about like originally I don't know if people know my background but originally I was a web designer and my happy place was headphones on Photoshop and you know CSS front end but I would always get so frustrated because from the moment that I produce the um stylesheet and the HTML I can't do the follow through. I can't execute the web app. And now I feel like I can really move at the speed of thought. Like some there was a great tweet by a guy. He said, um, I finally feel like AI enables me to move at the speed of my ADHD. And I feel that. So, I've actually been I'm I'm someone who sleeps nine hours a night and I wake up to pee at like 4 in the morning and I just get out of bed because I'm so excited to work on this and and it just feels like I'm It literally feels like I have like 30 free employees and they're just working 24/7 and I'm paying them like $40 a day. It's crazy. \>\> It's crazy. This is actually one of my predictions for 2026 that I think you're an interesting case for is I think the the unsung group of people that are going to be most affected and empowered by this era of new software with Opus 45 are designers. Um and there's this group of designers. So I think I think you're in this category. There's someone that works for me, our creative lead, our creative director, Lucas. um where they're like I've been able to make beautiful things for forever but I've always as soon as I got to coding like it wasn't something I could do and so I had to depend on all these other people to do it and especially in a world where code is cheap and anyone can make a vi vibe coded app designers who know about like how to make a great experience that makes you feel something now have these superpowers where they can just like take it end to end and I think they're like some of them are just going wild and they're so excited about it and I think you're you're one of those people. Well, even even if you think about chunking, like so, you know, I would do a wireframe and then I would do a Photoshop mockup and maybe in the Photoshop mockup I have filler content and then I got to spend four hours doing the copywriting and I'm really just doing the same principles like I've read all the books and let's say that I'm writing something I'm going to be like how would Jason Freed put this or you know what are the principles in made to stick or near hook I'm just going to run that algorithm them in my brain and then I'm going to have to spend a bunch of time doing it. I'm finding the copy even is perfect. And so it just feels like um I you know I'm somebody that doesn't like coordination problems. I don't think anyone does, but I get particularly frustrated and I usually throw my hands up and don't uh follow through as a result. And the way that's played out in my business life is that I love building software, but I've ultimately decided that that's my hobby because I get too frustrated doing like one-on- ones and product management meetings and all that stuff. And so I've delegated that and now I'm going like, "Wow, I don't need to delegate it. I can build automated systems to do all of these things." I'm curious. Do you have do you have other examples of creative people that you're hanging around with that are using AI in interesting ways? \>\> Let me think about that. Um, most people I know are still afraid or they don't know how easy it is. Like I think, you know, when you talk to someone, even for me, I was like, "Oh, cloud code that's intimidating. I don't know how to set up like an environment or dev server. I don't understand any of that." And I think even like um I did a workshop um for fun with a bunch of my friends where I invited uh the team from Lindy to come to Victoria and we got a group of about 50 people together and people were blown away by how easy it was to build automations and started building all sorts of cool stuff. But what's crazy is like cloud code can build things that are hundred times more powerful and it's easier in my opinion. It's even easier than building a Lindy automation. Um, and I I find that that's the the gap right now. It's almost like everyone is using the yellow pages and you're like, "Dude, just use Google." And they're like, "Oh, that's I don't know. That's like for hackers." \>\> Yeah. Yeah. Yeah. Yeah. You're totally right. I uh so I taught this course maybe like three weeks ago for um every people called cloud code for beginners and it was just a one-day course and it it was so cool people because it was really for people who are not technical intimidated by the terminal from all walks of life and seeing people for the first time be like uh the demo app was make an expense tracker for me using my credit card data um to help me like budget and whatever. The first time people just like typed that in and like waited 10 minutes and then had a full featured app that was theirs. It was just like this mind I'm getting like chills thinking about it because it was just like such a beautiful amazing moment and by the end of the day they had all pushed it in all these different directions that were for them. like one person, she built um this uh calendar view that tracked um her streaks of no spending days. And then other people had, you know, it was like uh I'm saving for the for a couple of big purchases. Like how close am I? Like there's all these different ways that people make software their own once they have these powers that are just not possible today. And it's so cool to see them be empowered in that way. I kept I kept making the mistake when I started vibe coding of being like, "Hey, um you know, I always joke like when you start anything, you don't want to be the guy who walks into the gym and tries to deadlift 300 lb on day \[laughter\] one." And I've been amazed like I tried I think in my early vibe coding, I would just be too ambitious and then it would be a disaster. And now I'm realizing that it can meet my ambition. So for a long time I've struggled with email triage. I have a very very particular way of doing it and I use superhuman extensively but it doesn't meet my needs and so I've tried to build stuff in Lindy. I've tried custom triage systems with my assistant. Finally the other day I just said um to claude code here's my Gmail credentials. I want you to build a email email triager. Here's how I want it to work. And it basically spit out a pretty simple version web- based version of superhuman that met my exact workflow integrated with Google. It's not perfect, but it but in like a week, it's gotten to the point where I can use it every single day. And that is astounding. Like anyone who's technical knows how astounding that is and how frustrating it is to build an email client. Uh it blows my freaking mind. Yeah, I think for a comparison, Superhuman spent and you know, if you're from Superhuman, you're listening to this and I get this wrong, please correct me in the comments, but I think Superhum spent like five years in beta because like just getting to the point of Okay, we've we've gotten to we've gotten the basics done of a good email client is so hard. Um, so I think the thing I'm curious about this email client is uh uh yeah, what is the workflow that you came to that is better for you in this new world that you when where you're able to invent your own email client? How does it work? \>\> So what I um I get a lot of email I get 200 to 300 emails a day. And so it previously was a full-time job for one assistant, sometimes two, and then it was a full-time job for me. And I would wake up every day feeling like um I don't know if you ever watched I Love Lucy, but there's that scene where she's in the chocolate factory and she's trying to keep up with the conveyor belt and she just gets so overwhelmed. I felt like that at all times. And so what I did that helped me a lot is I built a Lindy automation where every single email that comes in gets processed and routed. So depending on what it's related to, it gets routed. if it's a um anything that's sales or or marketing or anything like that, it automatically gets archived. Um and then the routing probably reduced my email load by about 50% because most things I was just sending on to other people. And then what it did for a long time was it would come up with a choose your own adventure. So let's say you email me and you say, "Hey Andrew, um I was thinking about coming and visiting you in Victoria or you could come to New York or we could do a Zoom. What do you want to do?" it would just send me an email and say, "Here's who Dan is. Here's what he wants. You can choose one, two, or three." And I would just respond to the email with one, two, or three, and the agent would send a really nice, friendly email to you. Um, that was good, but the problem I ran into is often there's edge cases or I want to tweak what it's going to say or whatever. And then for a more complex email, so let's say you sent me an email and you said um I want to come to Victoria, but I'm thinking about one of these five uh hotels and do you want to do a workshop or do you want me to maybe do some public speaking or should we do a meetup or whatever? Like let's say there's like 10 questions embedded in it. There's a So what what I did is I basically created an interface where it finds all the emails that are in my email that are not archived that require a response. It then ranks them based on priority and importance based on time and kind of um who the person is and what the request is. And then it gives me an interface where I can either multiplechoice respond or I can go into Q&A mode and it can say here's all the questions Dan asked and I can say like yes, no, or put in my own thing and then it creates the draft and does it. And so it's just has it's something I wasn't able to create using Lindy but now I can do in my own web app. That's so cool. Oh man, I want to I'm I'm excited. \>\> I'll I'll demo. Yeah, I'll send you a demo. It's really cool. \>\> Yeah. Yeah. Yeah. Um Okay, awesome. Well, let's go back. I want to go back to the relationship thing. So So just to like make sure I I understand. So basically you um you created a I guess relationship coach with cloud code that uh you can load in all the context of your relationship and then it did sort of a diagnostic about you know here are the probably the the main conflict areas that you're able to that sort of like predicted them without you necessarily telling it that you're able to use in your relationship or or give me the \>\> yeah let me help me see what it is. \>\> Let me screen share. \>\> Okay. Can you see this? \>\> It's loading. I can see it. \>\> Great. So, um, all this copy is written by Claude. So, this started out as a barebones multiplechoice quiz and it would produce a JSON file. So, where it started was I was like, man, there's all these services out there where I can do a bunch of these tests, but they cost a bunch of money or they're only available to therapists. So, I just said, um, Claude, I want you to run me through the following tests, build a simple interface and combine them all and make sure that it's clinically validated and then spit out a JSON. So, that's where we started. And me and my girlfriend both spent 40 minutes doing it, and then it's it's spat out the analysis. So, where I've gotten to today is I'm gonna sign in. \>\> Well, let me just let me just stop you right there. So, let me go back. I want to go back to the the homepage real quick or just open up the homepage for people. Sure. You can. Yeah. Cool. So, first of all, for people listening, it's a you're on an app. It's called Deep Personality. Um, the first thing that jumps out to me is in is the headline. It says, "In 25 minutes, you'll understand yourself better than a therapist would after 10 sessions." That is a great headline. How did you get it to generate that? So I um I created a um a or I just used a prompt. I basically said I really admire the writing of Jason Frerieded and Chip Heath. Chip Heath and Dan Heath wrote a book called Made to Stick. And the idea is that um you have one shot to capture someone's attention. And uh the book is incredible for anyone who hasn't read it made to stick. I still remember the first line of the book. The first line of the book is, "I woke up in a bathtub full of ice and realized that some somebody had taken my kidney." Right? So, it's it's the most striking opening line ever in any book. And then it talks about the principle that you really only have a second to grab somebody. Um, and so I I used that as a prompt. Uh, and I said I wanted to emotionally resonate. And so, you can see in the background it's added all these little, you know, why do I what is it? What work would energize me? what do I actually want? What makes me unique? Why do I self-sabotage? So, uh, immediately we're trying to grab someone and go, "Oh, I relate to this problem." And then a lot of people also are a intimidated to go to a therapist, but also they feel like their therapist maybe doesn't get them. So, I felt this would grab me if I saw it. Um, from there it kind of breaks down, you know, reassurement, reassuring people data is encrypted, it only takes 25 minutes, no payment, and then here's all the things you can discover. So, uh, why you love the way you do, why work drain, some work drains you and others, other doesn't, uh, the gap between how you feel and how things look. And then it kind of shows a demo um, profile sample of what you get. So you get this crazy like 45page AI generated analysis \>\> um that breaks down all the gifts and challenges of your personality, your romantic partner, your ideal jobs, all that kind of stuff. \>\> And uh and anyway, I mean this is a this is a website that I probably would have previously paid 20 20 $25,000 for, you know, uh if I had a designer work on it with me, you know, maybe maybe less, maybe five grand if I was being really cheap. But I mean, I made this website in two hours, I think, just prompting while I was building other stuff. So, let's go ahead and log in. So, basically, you do a big assessment. It's a multiple choice quiz, and you can just see what it ends up giving you. So, I've already got mine, so I'm going to go to my results. So basically, you get this huge document that analyzes you in crazy levels of depth. Wow. And then below it, you can see, let me show you. You can also uh train your personal AI. So it gives you prompts for chat GPT. \>\> Whoa. and it basically gives you a prompt that breaks down all of your scores and stuff that'll help um create like a great virtual therapist. And then it gives you these cards that kind of explain, you know, the best way to work with you. You could give your boss, you could give your partner about what you need, your attachment style, and then here's just a bunch of stats \>\> about me. So, you know, I'm uh agreeable, very low conscientiousness, which means I don't like details, very extroverted, secure attachment, blah blah blah. And it goes it goes crazy deep like even like dark triad, narcissism, uh all sorts of all sorts of stuff. So, it's just like it's like the most personal therapy session you can possibly do privately with yourself in 40 minutes. \>\> That's really cool. One thing that I I I kind of want is I want it to in addition to my self assessment, give me some prompts I can give to Claude or ChateBT to say like based on what we've talked about, how what do you think of Dan? So I can get a little bit of like an objective, not objective, but like another source of information in there that's not just me. \>\> Yeah. I mean, so here's here's my girlfriend Zoe. So I added that and then you can say analyze us as romantic partners or for everything. So, romantic partners and then it'll go off and do this. And this was the scary I won't I won't show this just because it's private, but but um this was just so wild. I mean, we were um we were reading it out together and just laughing our heads off because it was like all the things we fight about perfectly laid out. And then when we put it into our GPT, we have had a few little arguments lately and it's been really helpful where I can go to it and say, "Oh, Zoe's upset at me about this." And the biggest problem in any partnership is um people fight you on the surface about other things where it's like, you know, the classic line is like it's not about the dirty dishes, like it's about not feeling seen or it's about some childhood thing. And so where it's been really helpful is building empathy for one another and putting that into words. So, you know, we did this and I we were having a fight and it said um I was like, "What can I tell Zoe that will make her get this?" And it broke it down. It was like Andrew, you know, when you do this, it triggers this. And because of that, um it triggers this thing with his mom. And when he was a kid, he felt that when he did this, he was unlovable. And it was like the moment I said that she softened and we're like, "Oh my god, like that's crazy." So, um, it's just pretty profound. Like we're talking about the implications for programming. And this is obviously a great demonstration of its programming proess prowess, but even for psychology, therapy, anything. So, here's here's our relationship blueprint, how we match, where we're compatible, where it's difficult, that kind of thing. So, it's just it's super cool and I'm really excited to share this with everybody because I think it's a super useful tool. Um, and I'm just trying to figure out like what do I do with this? Do I do I use it as like a is it a corporate, you know, a tool you might use to figure out how to work together? Is it just about romance? Is it about everything? Do you use it to figure out if you should hire someone? Like, I don't know yet. But I'm I'm realizing like I have a software company here that I just dreamed up out of nowhere. \>\> Growth is good. Speed is good. But there's a specific tax that comes with growing fast. Every campaign needs a new landing page. \[music\] Every product launch needs copy updates. Every AB test sits in a backlog somewhere. \[music\] Your site becomes the thing that slows you down. And there's just not enough engineers to make up the difference. Framer is a website builder that works like your team's favorite design tool. With real-time collaboration, a robust CMS with everything you need for great SEO, \[music\] and advanced analytics that include integrated AB testing, your designers and marketers are empowered to build and maximize \[music\] your.com from day one. Companies like Perplexity and Miro and every use Framer because it eliminates the bottleneck. Marketing can test ideas. Designers can ship updates all without touching the engineering queue. Changes go live in seconds. One click, publish, and \[music\] done. And it's built to scale with you. Framer offers premium hosting with 99.99% uptime SLAs's and enterprisegrade security. Whether you're launching your first site or migrating an established.com, Framer \[music\] turns your website from a constraint into an advantage. Learn how you can get more out of your.com from a framer specialist or get started building for free today at framer.com/dan for 30% off a framer pro annual plan. That's framer.com/dan for 30% off. Rules and restrictions may apply. And now back to the episode. That's really cool. I think the answer is probably yes. And it's really easy to spin up five different versions for different verticals, you know. Um, which is which is pretty cool. Uh, and yeah, I think there's it's there's something it's something I've been doing like with GBT3. I was like feeding my journals into it or and stuff. I don't know if you remember that. And it's like it's so interesting that I think the overall discourse is very okay psychosis and all that kind of stuff. And look, if you have real mental and emotional problems, chy should not be encouraging your delusions for sure. And if you're thinking about a technology that can get um the best that things that we know about relationships and emotions and psychology into everybody's hands, there's no better tool than this. I think um I've seen a lot of therapists I think that you've seen at least some therapists and like it's hard to find a good therapist and a good one is 10 times better than than like your average one. \>\> Well, a great one is about 300 bucks an hour, too. If you go to like high end, it's very expensive. \>\> It's expensive and and and even if it's expensive, it's not a guarantee that it's good. \>\> And I think these tools are the best way to get that kind of thing. It's not the same thing as therapy, but the best way to get that kind of thing, getting people into that pipeline um uh uh that that has ever been invented. And it's also super useful. My experience is just taking some of this stuff to therapy and being like, "Hey, I was talking to my best friend Claude and like Claude said this." If you're some therapists are going to be like, "I hate this." You know, but um at least my therapist is very open to it because I think he's realized over time that there's actually some really good insights there and you can use it at the time that it happens instead of a week later you're like reporting it in therapy. You can just be like, you know, okay, this thing just happened. Yeah. Um, do you have a Limitless pendant? \>\> I don't, but some people that I work with do. \>\> You got to get one. Um, I've been wearing this thing. So, it's a little I don't have it on me, but it's this little black circle and you just pop it on. I usually wear it on my jeans pockets all day. And you know, a lot of the time I do meetings in person um, and I'll be working with somebody and I have ADHD, so I have very bad memory. So, like my girlfriend, for example, will say like, "Oh, um, there'll be something, let's say something around the house, right?" And I talked to the contractor for 20 minutes and he says, "Oh, we got to do this and the plumbing or whatever." And then Zoe says, "Oh, what did you talk to the contractor about?" And I'm literally like, "I have no idea." And and so having this perfect memory at all times has been incredibly valuable. I've done that same thing sometimes with therapy or um if me and my business partner are dealing with a hard thing, I'll record it and then I'll put that into AI and say, "What do you think he's getting at or what do you think he needs to hear in order for us to resolve this or whatever it is?" And so I think over time you you know, you can just see this world where right now I mean it first of all I just want to say it's hilarious that people are getting caught up on the psychosis stuff. It reminds me a little bit of like uh when the iPhone came out and it didn't have an app store and people are like well it's never going anywhere because you can't build you can only build web apps on it and it's like hm I think they might resolve that you know technology might get better. Um, but I I can see a world where we all get very comfortable wearing an Apple Watch that records everything all day and or or the new OpenAI device or whatever it is and suddenly we have complete context on all of our um all of our lives. And I think the challenge right now is the context window is too small. I'd say like cloud code does a very good job of context window management, but and so does Chad GBT, but it loses the plot too quickly. And I'm excited when we can have a 5 million, a 10 million context window because then you're going to be able to really incorporate every conversation you've had this year. \>\> Yeah, I think um the thing that I use for this mostly is Granola and um have you used them or heard of them? \>\> No, I use Otter. I I actually built a custom Lindy, which I'll I'll talk about later, but for meeting notes as well, but I generally I use Otter for recording meetings. \>\> Got it. The the interesting thing about Granola for me is I mean they did the whole thing where it just records everything you're in and there's no extra bot or whatever. It's just recording on your computer which is really nice. But the interesting thing about Otter is it has like a really good pretty good AI assistant inside of it that um you know I can put something in that's like uh I did this the other day because you know we're at we're at about 20 people at every and so I'm starting to think about okay what is like onboarding new employees look like and what is our handbook and like what is our culture and whatever and so I just asked Granola like if you're going to write um a handbook for new employees with here's how stuff works at that no one will actually say, but is like definitely how things work. What would you write and use like my last 25 meetings to do it? It was pretty good. Like I probably wouldn't just like put it into every employees hand like immediately, but um you have the ground truth for it to start working from. And and so it's it's amazing for that. It's it's really good like you said, you know, if you have a real personnel issue, it's like these conversations are so difficult and you're always wondering as a CEO or any sort of leader like, okay, how did I handle that? And is there a way I could have done better there? Um, and and and maybe and that's so that's like the postgame analysis and then there's also the pregame analysis like take a look through our previous conversations like what should I think about or how should I how should I handle this kind of situation and it's it would you know executive coaches are like 5K for a session or whatever like to get someone really good and it's so good for that like I'm such a better manager than I would be because I don't have to make every make things up all the time. I have um two things that I think would would you would get a lot of value out of. I don't know if you're already doing this, but so one is um I created a custom Lindy agent that records all my meetings. So it takes meeting notes, but the most useful feature that I added was um I do a lot of meetings with people that are pitching me on things or I don't know. And I have a tendency to really enjoy narcissists. I don't know why, but I really love narcissists. Um, I'm drawn to them. And, you know, I've dealt with all sorts of interesting characters over the years in business where, you know, psychology is probably the most important skill in business. And flagging something like covert or grandiose narcissism or a paranoid personality or someone who's being manipulative is very important. And so my Lindy agent actually texts me after a meeting if there's any red flags. So recently I had a meeting um with a contractor I was working with and I was quite upset because he had said he would deliver something on a schedule um and then he didn't he didn't do it and instead of and I was pretty calm but I was just holding him to account and instead he started like uh you know you've really you know you were so rude and and um kind of you know really really harsh to me back talk about it and I actually left the call feeling really upset at myself and then I got a text from my Lindy agent about a minute later and it said, "Hey, um, I just wanted to make you aware this person was using some manipulation tactics. They were gaslighting. They were reframing in this particular way." And it was really helpful. And I've used that a couple times where I'm interviewing somebody and it just says, "Hey, this person is um highly highly highly anxious and needy." and that's okay, but you should know that that's going to be part of working with them or whatever it is. So, I found that incredibly useful. On the flip side, for employee feedback, I uh use Super Whisper. And so, I have the action button on my phone launch Super Whisper. And I don't know if you've used it, but you can basically prompt and I think you guys built a tool like this. We did. Um so, I should switch to yours, but um \[laughter\] \>\> basically you can create prompts. So, you know, the most simple one is just like, hey, turn this into good text. But I have one called uh good boss. And so I I basically can go into it and just be like, "What the You know, you didn't do this thing you said blah blah blah blah blah." All the like harsh stuff I think. And it converts it into what a good boss would say. And it just tones it down and it makes it more mature. And I found stuff like that really really useful. It's the best. I love it. Um on the narcissism thing to go to to go back there, A, that sounds awesome. be I'm curious um you know for myself one thing that I've noticed which I think is sort of important because I'm I'm currently grappling with this is sometimes I'll um put it in for maybe an interpersonal situation in my personal life or in my work life and you know maybe I'm feeling upset about something and I've noticed that it will maybe subtly reinforce my interpretation of things and then something will happen later that is like breaks my interpretation. I'm like, "Oh, wow. I was thinking about that totally wrong." And then I go back to it and I'm like, "What the fuck?" And it's like, "Oh yeah, you're totally right." Like blah blah blah or whatever, you know? And there's something about that that I think is an interesting there's an interesting question of AI hygiene, right? Like when we have these things that can that can do some of do this identification and also it's it's really hard even if it's super smart it's really it's really hard with a limited set of information to be able to draw like lots of conclusions and um that doesn't mean you shouldn't use it but it does mean there's there's this thing of how do I hold what it's giving back to me so that I I think of it in a in a skillful way where it's like sometimes they were actually being super manipulative or you know in my case maybe it's like there's uh their uh they don't want to be friends with me anymore or something like that like it's you know our relationship is strained um and sometimes it's like it's something else that that a further conversation or or more more context would like would reveal that how have you noticed that and how do you deal with it? Well, the you have to be very careful um to your point where you can't inject any of your own opinion into the prompt. So, for example, the prompt that I'm using is very much like high bar. You have to analyze every single word in this thing and you only flag it if it reaches a critical point where like I get one of these notifications maybe once every two to three months, right? So, it's a very high bar. Um and I think it's a little bit like body language. Have you studied body language at all? So there's a there's a um it's all about patterns, but it's like if you I I studied it a lot because I do a lot of inerson meetings and stuff and you know really my job is to assess people and then try and figure out what they're thinking. So for example, I'm meeting with a founder and when I ask them about a certain thing in their business, they're scratching their neck a lot and they're crossing their arms and you know getting defensive or whatever. You have to also understand their pattern of behavior in general. So, for example, my dad, he just always crosses his arms all the time. So, if I met with him, I can't say, "Oh, he's closed off." It would just be he crossed his arms probably because he's a big guy and it feels good or something like that. So, I think it's it's just data, but I think over time it just as it as the thinking models can do more and more analysis and have more and more data, I think it's just going to get better and better. And I I'd say like I probably wouldn't make a higher and fire decision entirely on that, but it will often confirm a feeling. It's it's usually like I feel like a cognitive dissonance and then I get one of those texts and I'm like, "Okay, that's what's going on." Yeah. Um Yeah. Yeah, I think there's also that thing about usually it's working from transcripts and there's a lot that's not in a transcript about how something was said or like how long the pause was between certain words or between when someone stopped speaking and someone started speaking like all that stuff gets lost or like what's on the person's face like all that kind of stuff which I think we're we're getting there and there are some models that are actually doing they can do real time um detection of emotion which is pretty freaking cool based based on your facial expression and how you say things. Um, and they're they have an API. So, I think you would probably be into this. It's called Hume. Um, and uh, I think that's that's an another level of yeah, they're crossing their arms and obviously you have to take into account what are their personal habits and and and what's the context of the situation, but also yes, in general crossing your arms like there's something you're closed off and and maybe that's about this person or maybe there's something else going on. It's it's such a fascinating world. I've had times where I'm depressed or anxious or vulnerable or sleepdeprived or just stressed out about business and I make bad decisions usually during those times. So to me it's like that extra buddy who's sitting there watching and maybe watching out for me a little bit. But yeah, I don't I don't know I would make a lot of decisions entirely based on AI analysis. Like recently I was like, "Oh, can you audit this company um and just look for any suspicious transactions?" like I you know I realized we hadn't done an audit on it and you know hadn't been paying attention to the company or whatever and I made the mistake of prompting it as if I was suspicious and it was like oh there's like mass fraud in this company there's been $90,000 \[laughter\] of fraud and I looked at it and it was like it was like an employee has purchased a bunch of Starbucks gift certificates and it's like no that a bunch of employees went to Starbucks and bought coffee you know \[laughter\] they like I prompted it the wrong way so you got Yeah. You gota be really careful with this stuff. Um, do you have any like Claude code workflow things that you've you've picked up over the last couple weeks that have helped you or things that you've learned not to do? \>\> Um, I would say the thing I'm playing with a lot right now is skills. Um, you know, skills basically for those that don't know, you can kind of say like um whenever you do a thing like this, reference this skill. So, a skill might be copywriting. So, some of the principles I talked about earlier, I've been thinking about um turning into a skill where every time I ask it to write copy, it just knows all the books I like and the tone I like and that kind of stuff. Um I'm trying to think of what else. I mean, the biggest thing I've done is I've created a sync across all my computers so that uh a cron job runs on every Mac that I have and just make sure that if I open clawed code, it gets back to where I was on any computer. And that's been huge. Um, and then I've been using all sorts of MCPs and stuff. I mean, MCP like I find it it's still early and some of the integrations suck and they're too slow, but you can really see how it's going to be this kind of tentacle monster that touches all of your software. And what I wonder about is like for example like why do we need software at all eventually? like Claude will be capable of building anything you need and probably doing anything you need uh and so will CHBT and so ultimately I really wonder what what software continues to exist versus just gets built on the fly for you. \>\> It's a really good question. Obvious the rules are changing a lot on the skills uh topic. Do you have you tried our compound engineering plugin? No, I just read about it yesterday actually, but I was like, I don't do a lot of like pull requests and stuff because I'm just working by myself. So, I was like, it's probably overkill and I'm gonna mess my system up. But, I'd love to hear about it. \>\> You should check it out. I I think you could use it for specific things that won't mess your system up. So, specifically, for example, um as you're getting more complicated in your app, um planning is really important. And the plan function in this thing is like it spits off like five different sub aents that um uh that research all the best practices for the problem that you're working on and research your current codebase and like find other code examples from like open source and kind of puts it all together into this big plan that has you know technical requirements and acceptance criteria and all that kind of stuff. that helps kind of keep it on track as you're getting it's overkill for for you know the first 20 hours of a project but as you're getting into a big codebase it can be it can be really helpful. \>\> So it' be like it'd be like keep security in mind, write tests, that kind of oh that's cool. I'm definitely going to check that out. \>\> Check that out. Um and yeah, I think to your point on like future of software stuff. My my like quick quick quick take is that there are these there's a new architecture for software that I've been calling agent native architectures and you can think of that as being every piece of software is just cloud code in a trench coat where um when you uh are using an app you are you know you press a button in the UI that's just kicking off an agent to go do the thing in the background and the really interesting thing about that kind of software is that every feature is a prompt And if every feature is a prompt, then um it's much easier for users to customize their features. Um and it's much easier for developers to make their software more flexible and grow. Uh because you know, you can be like, okay, like I'm seeing a lot of people are customizing our software in this particular way with these kinds of prompts. I'm just going to build that as a feature. And really all I'm doing is taking the prompts that they've created and pushing it into main and pushing it out to everybody else. So there's this whole new kind of like more flexible way of um building software that's that's really agent native where anything that a user can do in an app the agent can do and once you're there it just opens up this whole new thing and so that's that's kind of like all we're doing in 2026 is like agent native architectures and I think that's that's a big thing um and then yeah which what needs to be software versus what what should just be something that chatbt does is another really interesting open question \>\> you guys are building in production and I know you guys have a whole suite of software. What are you guys finding is the current gap at 4.5? What are you hoping for five or six to be able to do? I don't think that I don't think that we know what the gap is yet because we are just like holy we're just flying. Um I I would guess the one of the bottlenecks right now is you can produce so much code that um the bottleneck is reviewing what it built to make sure it works. And a lot of what we do is uh you know today for example I have I'm building this like personal re reader app to help me read better. And it's an agent native app. It's super cool. I built it myself in between meetings. It's like it's it's the best. And um uh and I have all these like little things that as I'm using it every day, I'm like I want this to be better. So I have a Apple note with like a bunch of little things. And I was at breakfast today and before I started eating, I just like popped off a bunch of cloud codes on my phone to like do like 15 different \>\> You just SSH into your computer. What do you do? So we have that but the uh cloud code like cloud in the cloud mobile app there's a code section and you can just like talk to it and say hey I want you to go do this and it will pull down your repo and start it in a virtual machine and then like create a pull request for you. \>\> Oh I haven't tried that yet. \>\> You should check that out. It's pretty cool. Um it's I think what you have is actually it's actually a bit better but for quick things that you want to do on your phone without like messing around it's the cloud app is actually pretty good. So I spun up a bunch of parallel things to fix a bunch of little little bugs that I found and little features that I wanted changed and I finished breakfast and I was like cool like now I'm going to test all this but it's that's a lot of work to to test everything and to make sure each thing is like done right and so especially for a mobile app it's like hard to wire that up where cloud tests itself but I think cloud is good enough to do it. So I think the big the big bottleneck is closing the loop on everything that you build so that by the time it reaches a human it's pretty much good and um and it's working as intended and it's it's you know it used to be the bar was can it get to the human without any errors and that's like basically solved. It now it just has no errors. But the next bar is um does it actually work well? And that's a I think that's still an open question. Yeah, it seems like the big problems that I notice are like like right now I'm using um Superbase and Versel and it's just like there's things that you still need to go and configure and I I do wonder if over time Anthropic can just launch their own database product and their own hosting product and then you just have this one sandbox you work in and it does absolutely everything. That would be really incredible. \>\> I think they probably will. One thing that I've I love is um I've been using Chacht's Atlas browser if you've never tried it. \>\> I love it. I use it every day. \>\> Isn't it sick? And so for anything where you're like, "Oh, I'm configuring Versel." I just open up the HMO mode and I'm like, "Go do this thing." And I just never touch a settings dashboard ever again. And it's so good. \>\> It's amazing. It's so cool. \>\> Alice is amazing. Everyone should try it. I know you have a couple of more automations to show. I'd love to see them. \>\> Sure. Um, here. Let me Can you see my window? \>\> I can. Have you seen the movie Clueless? No. So, Clueless, it's like this rich girl and there's this funny scene where she goes into her wardrobe and she has this computer and the computer automatically matches all her outfits and stuff like that. And I've basically created the modern AI enabled version of that. So it's called personal stylist. \>\> Amazing. \>\> Let me pull it up here. \>\> You're really good at doing this for your personal life, \>\> dude. I It's really where I use it the most. So So one problem I've had is I um like so many guys, you know, you want to look good, you want to dress well. I don't understand color theory. I don't know what fits. I don't know what looks good. And so, uh, probably four years ago, I was newly single. I got divorced and I was like, I got to got to look good. And so, I hired a personal stylist, which is actually not that expensive. Um, it was like a couple thousand bucks. And this lady basically made a bunch of outfits and sent me stuff, but it was a total pain in the butt because she's shipping me stuff that doesn't fit. I, you know, I'm not choosing any of it. And so, what I did is I basically created this um AI automation. So you can see uh every day at 7 in the morning it goes to um it goes and checks the weather in Victoria and then it generates outfit recommendations and it has access to a um a Google sheet called Andrew's wardrobe. And you can see in here I basically just took photos using Claude and then had it um turn it all into a CSV file. So you can see these are all the all the clothing that I own. And then what it'll do \>\> is it goes to Nanobanana. It it creates four outfits. It goes to Nanobanana and then it generates a rendering of those outfits. So let me see if there's one. Here's here's one with me. So I basically gave it reference images of me and then it sends me this. I get this as a text using Twilio and it says wear this watch and this outfit and here's the different clothes. And it has really helped me up my game. Uh, it's been it's been awesome. And it's just one of those really dumb fun things that, you know, everyone kind of goes, "Man, I wish I wish I could do that." And now it's like it's like Uber. It's like everyone has a personal driver now. Well, everybody has a personal stylist if they want one. So, that's been a really fun one. \>\> That is amazing. And I will say you dress well. \>\> Um, thank you. \>\> Yeah. Like I like \>\> chatbt though. \[laughter\] ChachiT's got got your wardrobe, got your face, your Yeah, it's it's amazing. \>\> Yeah, I've also got a custom GPT and so I can just take a photo of like, hey, what goes with these jeans or you know, tweak this outfit or tell me it'll be like, oh, you should French tuck that shirt or do these things. So that's been that's been really really cool. Another one that I created that I absolutely love, um, so you don't have kids, right? \>\> No. Those of those of the listeners with kids will relate to this, but my kids school, it is a full-time job keeping up with the emails. There's emails about um you know, trips and things I need to sign and lunch on certain days and all this stuff and I really get overwhelmed by it. And so, as part of my email automation, I basically have it uh ingest all those emails and then it texts me and says, "Hey, heads up. you know, Peter needs a packed lunch tomorrow and here's the link to sign the uh you know, the the field trip notice. And then it also puts it all in my calendar and having it just automatically get added to our parenting calendar and everything being organized has just been incredible. Uh and it's one of those things where you probably don't get it, but like if you talk to a parent, they're just like, "Shut up and take my money." Like, "How do I do this?" Amazing. How has Okay, so now I'm I'm I think we can get a little bit philosophical. Um, how has this changed your whole view of the world and of software and of businesses? So, a couple things that I think I'm interested to hear from you. one is, you know, as we've talked about this over the last 6 months, I think this is actually the first time we've talked about it where you seem so energized and so psyched. And um most most of the other times I think that um I think you've always been excited about it, but you've also been anxious about it. Like how does this affect the world and how does this affect business and are things changing uh in a way that like our jobs going to exist in 5 years more or less? Um, so how do you feel right now about that? Like how is your how is your position? How is your um how's your thinking evolved and then how does that um layer down into software businesses and how you think software businesses are are going to exist or work and and what that means for example for tiny where you own a bunch of software businesses. So I I'd love to hear your take on all that. \>\> So yeah, I think I talked to you maybe a year ago. I called you and was like, "Dude, like are you freaking out or what?" And I think I was really struggling with the idea that all knowledge work could go away or radically change. And so I carry two different feelings. One, as a creative person and entrepreneur, I'm incredibly energized and excited. I feel like I'm more competent and I can create so much more and I can communicate better and and so I'm very excited about building things and how we can use the these tools in our businesses but um it definitely changes things in terms of buying businesses. So we have really slowed down on buying technology companies and software companies. Uh I think unless you have a distribution moat or a hardware moat or something like that I think most software businesses are just thin wrappers like if your business is like a database call or an AI call I just I think it's not that it's going to go to zero it's just it there's going to be a lot of competition and I think about it like um pizza restaurants right so pizzeras are they are they well let's say that somebody makes a machine that can incredible pizza, the world's best pizza, and anyone can buy it for $10,000. Are there going to be any more pizza restaurants? And will the world of the pizza business go away? No. Um, consumers will benefit because there's going to be amazing pizza everywhere, but business owners will really struggle because the margins will go down, right? Maybe you used to be able to charge 10 or 15% margins on a slice of pizza, but when the cost of the pizza goes way down and the quality goes way up, the consumer benefits and the business owner uh you know, their margin goes to like 1% or 2% basically. And so I think that's going to happen all over the place. I like I remember like uh a year and a half ago people were getting very excited about all these like calorie tracker apps like you know kid vibe codes a calorie tracker and makes a million dollars a month and it's like that's fine like he picked up some pennies in front of the steamroller but what happened in the next three months everyone who could just copied him and the the moat for software used to be that there's only so many programmers programming is hard to learn and takes a long time to learn and it's very expensive to hire them and now it's basically free and so your moat has to come from something else. You have to have a brand or a distribution mechanism or hardware lock in or something because I just don't see them being very good businesses in the long term. \>\> I think that makes sense. I I really like the um I like the pizza analogy um especially because like what constitutes great pizza is also sort of up for debate, you know. Um, and yeah, maybe it can make great Neapolitan pizzas, but maybe you're you you like deep dish and so there's like endless complexity to the economy and so it doesn't get rid of pizza restaurants, but yeah, it sort of changes the landscape. How do you think about that? You know, I think probably a lot of people listening uh own software businesses or are five or 10 years into software businesses that are not necessarily AI native. How do you think about for for your own businesses um how to face that future where things change like this? Well, I think it's a little bit like owning a um water powered factory. You know, if you have something that's powered by um you know, a windmill, you better convert it to electricity pretty quick, otherwise you're going to be out of business. And it's really hard to do that because you have a lot of sunk cost. The beauty of a software business is you can actually replace a lot of the human work. But the question is where do the humans go and who are the consumers that are going to be buying all this pizza and all this software and how do they make money and in you know historically the way that it's worked is that the bar just gets raised. So, in the pizza analogy, um, pizza, you know, pizza shops all look the same, um, but someone's really creative and they spend more money and they make an amazing space and they get on TikTok and, you know, they're able to kind of build something different and the standard of what people expect when they enter a restaurant just goes up. So, that's certainly a possibility, but the problem is that AI compounds in all industries at the same time. And so what I worry about is not so much um long term like I think in 20 years we're going to be fine. What I worry about is another great depression where you have um suddenly all these white collar employees getting laid off from previously very lucrative jobs like lawyers um you know probably not doctors but maybe doctors anything that that is knowledge work programmers and where do they go? I think a lot of them will probably go to bluecollar work, right? So maybe you go start an HVAC company because people need um air conditioning, but what happens in HVAC when a hundred people all start an HVAC company in your city at the same time because they've all been laid off? Well, uh the margins go to zero or they go very low. So again, they'll they'll have jobs and then you start thinking about robotic automation and you can go down a whole other rabbit hole. So I'm not there yet. I I'm but I am thinking about um how do you protect yourself from that and I really believe that if you have money to invest I think owning compute is probably or computer power is probably one of the only ways to own a tollbridge to the future. Um so I think part of the reason I feel commerce is I've also made a bunch of investments um in frontier models and in um data centers and that kind of stuff. So, I think my um my butthole relaxed a little bit, but you know, I was pretty I was pretty freaked out and I think a lot of people should be pretty freaked out. Um but you know, I live in the Pacific Northwest and there's everybody knows there's going to be a super thrust earthquake some point in the next 50 to 100 years. We live with that every single day, but there's very few people I know who are obsessing over it. Um you know, in my basement I've got earthquake stuff. I've got a Starlink dish. I've got food. And so what I've realized about AI is all you can really do is educate yourself and prepare and then not worry about it. And so that's kind of what I've been doing and just enjoying it. \>\> For someone who like maybe has some money to invest but doesn't maybe doesn't have access to a lot of private deals that you might have access to, what are ways to buy, for example, compute? So, I actually um just made a large investment in a company called Iron in January. And Iron is just a it's actually a Bitcoin mining company that had a shitload of data centers and power. And they realized, oh my god, this is really valuable for AI. And so, they pivoted to AI. And so, I was able to buy it at a valuation where it was being valued like a Bitcoin miner, not very highly. Um, I think I invested at a $10 share price and now it's at 35 bucks and recently it was at 70. So, uh, that I look at as a bit of a hedge that anyone can access. Uh, and there's a lot of those a lot of those businesses. You just got to be careful to figure out which ones are real and which ones are kind of made up. \>\> I love that. Um, Andrew, I love hanging out with you. I always learn so much from uh from our conversations. Thank you for sharing and we'd love to have you on again soon. \>\> Yeah, that was awesome, dude. Anytime. \>\> Oh my gosh, folks. You absolutely, positively have to smash that like button and subscribe to AI and I. Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure unadulterated knowledge bombs about chat GPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show. It's a journey into the future with Dan Shipper as the captain of the spaceship. So, do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now, without any further ado, let me just say, Dan, I'm absolutely hopelessly in love with you.