Mike Taylor, who teaches Every's Claude Code for Beginners camps, joins Kate Lee (editor in chief of Every) to discuss what happens when beginners start building with Claude Code: how newcomers approach the tool, what works, and what surprises them. # Channel Identity Observations: Every (Live Stream) ## 1. Explicit Self-Description (Creator Claims) ### Channel/Organization Identity - **Publisher**: Every, described by its editor-in-chief as: > "Every is the only subscription you need to stay at the edge of AI. We offer ideas, um apps, uh and training." - **Format claim**: A hosted **live stream** — "this live stream that uh we are hosting from Every. Uh it's 3:00 Eastern time." - **Product scope claimed**: - A newsletter - "five as of last week, apps, uh productivity apps" - "trainings for both enterprise uh and corporate level and uh beginner level students" ### Named People & Roles - **Kate Lee** — self-introduced as "the editor-in-chief of Every" and stream host. - **Mike Taylor** — introduced as newly "full-time to be the head of tech consulting"; also writes the column "Also True for Humans" and authored a Udemy course with "300,000 students." ## 2. Claimed Scope of Offerings ### Education/Courses - **"Claude Code for Absolute Beginners"** — a live, cohorted, one-day course (~$1,000, per Kate: "Why spend a thousand dollars um when you could go to YouTube"). - Explicitly **not asynchronous**: "We teach it live. It is cohorted. We don't we don't do an asynchronous, you know, video version." - A more advanced follow-on course also exists ("a more advanced course… last week"). - Skill-level framing: audience is people with roughly **under an hour of Claude Code experience** — "if you've spent more than that… it might not be the right one for you." ### Consulting - Every's consulting team is described by Mike: Natalia runs consulting, Mike on tech, Brooker Belcourt (ex-Perplexity finance vertical) on finance-side clients (hedge funds, investment banks). - Mike's consulting focus: "working with engineering teams at tech companies… the kind of coding agent type stuff" and executive offsites ("teaching, you know, a room of C-suite people"). ## 3. Intended Audience (as Presented) - **Primary**: Absolute beginners to AI coding tools — non-technical knowledge workers, creatives, and professionals who "have literally never opened Claude Code." - **Secondary**: Enterprise/corporate clients and executives (offsite trainings, consulting engagements). - **Explicitly excluded**: Very advanced users — Mike: "I would tell like all my friends like not to join this course specifically because… you cannot like talk about agent swarms in the same breath as you're talking about like this is hey meet the terminal." ## 4. Observable Presentation Patterns (Beyond Claims) - **Promotional function**: The stream is itself a course launch vehicle — Kate closes by announcing the next cohort on **April 14** and directing viewers to a sign-up website. - **Interactive Q&A**: Kate relays live audience questions from the stream sidebar ("I see we've got a bunch of questions in the sidebar"), demonstrating a participatory live format. - **Conversational expert-interview structure**: Editor-in-chief interviewing an in-house expert about teaching experiences, positioning Every staff as AI practitioners. - **Editorial voice/tone**: Playful, informal banter between hosts (e.g., laughing about "Tech Consulting" capitalization, "AI degenerate like me"), suggesting a culture-forward, personality-driven brand. - **Anti-overclaiming stance demonstrated**: Mike repeatedly steers people away from the course when it doesn't fit ("I don't think they should buy the course" for the recording-only buyer), and Kate endorses this framing. ## 5. Content Themes Demonstrated - AI coding tools (Claude Code, Copilot, Codex, Gemini CLI) and adoption paradigms - Beginner barriers: terminal literacy, jargon ("why did they call it a pull request"), setup friction - Security/sandboxing, context management ("context rot"), AI agent management as "a new form of management science" - Practical use cases: inbox/daily-briefing automation, custom dashboards, Salesforce workflows ## 6. Claims vs. Demonstration — Summary | Observation | Type | Evidence | |---|---|---| | "Only subscription you need to stay at the edge of AI" | Explicit claim | Kate's opening | | Offers apps, newsletter, enterprise + beginner training | Explicit claim | Kate's opening | | Courses are live/cohort-only, ~$1,000 | Explicit claim | Kate's closing Q & "thousand dollars" exchange | | Audience skews non-technical beginners | Demonstrated | Q&A emphasis, course framing, "under an hour" rule of thumb | | Stream functions as marketing for upcoming cohort | Demonstrated | April 14 launch announcement | | Hosts present themselves as active AI practitioners | Demonstrated | Mike's consulting work, agents running 24/7, Every MCP usage | # Delivery Observation: Live Stream — Kate Lee (host) & Mike Taylor (guest) ## Format Two-person conversational livestream: Kate Lee structures with questions and audience-Q&A relay; Mike Taylor answers at length with anecdotes and examples. --- ## Kate Lee — Host Delivery ### Tone & Style - **Polished, welcoming openers** with housekeeping and organizational framing: > "this live stream that uh we are hosting from Every. Uh it's 3:00 Eastern time." > "Every is the only subscription you need to stay at the edge of AI." - Frequent **transition phrases** that summarize the guest's previous point before pivoting: > "Got it." / "Right. Right." / "Mhm." / "That's so interesting." > "Totally. Totally." ### Interview Habits - **Pre-frames questions with context** before asking: > "I'm sure people ask you this all the time, but can you see or say where is this all going a year from now...?" - **Relays audience questions** explicitly and tracks time: > "I see we've got a bunch of questions in the sidebar as well, which I will get to shortly." > "Um, we have just a couple minutes left." - Recurring **softening hedges**: "you know," "sort of," "I think," "maybe" appear throughout her questions. --- ## Mike Taylor — Guest Delivery ### Tone & Style - **Conversational, self-deprecating humor**, often at his own expense: > "I just uh cringed actually as I looked at my title, and I realized I capitalized tech consulting." > "if you're not already completely AI degenerate like me, you know." - **Balanced hedging before strong claims**: frequently pairs concessions with counterpoints: > "I'm able to forecast, but I don't know how much you should trust the forecast." > "maybe it's like somewhere in between product manager and engineering manager." ### Recurring Linguistic Patterns - **Heavy filler and repetition in rapid clusters** — a distinctive stammer-then-restart rhythm: > "so so so I would say um" > "I I I I can recognize my mistakes." > "you you you you like you're I I really learned how to code cuz..." - Signature framing phrase: **"to some degree"** (repeated at least 6 times): > "to some degree developers get paid a lot of money because they're willing to put up with all this complexity." > "in that very literal respect like coding is dead." - **"Get over that hump"** as a recurring metaphor for learner activation: > "so like it's getting him over that hump I think" > "I don't think you can get over the hump very easily on your own." ### Teaching Habits - **Concrete anecdotes before abstractions** — nearly every answer opens with a personal story: - Learning to code over "5 years evenings and weekends" - Paying a tutor "like 60 bucks an hour to teach me some like databases" - The Waymo self-driving car analogy for trusting AI - **Analogy-driven explanation**, especially management metaphors: > "it's it's almost like it's an employee that you're managing rather than someone that you're working together with." > "if you just hire a new employee today, you don't give them like a corporate bank account on day one." > "it's like almost like landing in a foreign country and everyone's speaking a language you don't understand." - **Direct address and reframing of questions**: twice he corrects the premise: > "I would say that the framing of the question is wrong. Like what will I use this for? I think it's... what what can't you use this for?" > "It's a persistence gap." (in response to "technical gap or process gap?") - **Memorable aphorisms** used as teaching anchors: > "there's no such thing as a technical person, just a persistent one." > "you shouldn't skip a step." > "you can use more AI to solve the problems that AI causes." - **Qualifies advice with self-aware honesty**, including admitting unhelpful answers: > "I know that's a terrible like not a useful answer." ### Humor Timing Humor consistently lands after a serious point, often flagged by laughter cues in the transcript: > "a developer that's $200 a month that doesn't sleep" > "you could also learn how to code in 10 years. [laughter]" --- ## Interaction Patterns - **Frequent laughter exchanges**, typically triggered by Mike's understated jokes; Kate often echoes agreement ("Right. Right. Yeah."). - **Kate's lightweight backchanneling** ("Mhm," "Right," "Yeah") punctuates Mike's long turns without interrupting. ## Uncertainty Notes - No timestamps beyond coarse transcript markers were available for precise delivery timing; all observations rest on textual cues (laughter markers, fillers) rather than audio/visual evidence. - Speaker attribution for brief interjections (e.g., short "Yeah"s) is occasionally ambiguous in the transcript. # Position Record: Claude Code for Beginners — Mike Taylor & Kate Lee (Every Live Stream) ## Participants - **Kate Lee** — Editor-in-chief of Every; host. - **Mike Taylor** — Recently full-time Head of Tech Consulting at Every; prompt engineering instructor (Udemy, ~300k students); former freelance developer; teaches Claude Code courses. --- ## Background & Roles ### Mike Taylor on his role at Every - Consulting team described as small: Natalia runs consulting, Brooker Belcourt (ex-Perplexity finance vertical lead) handles finance clients, Mike covers tech. - His focus: engineering teams, operations/marketing teams at tech companies; coding agents; automating knowledge work. ### Mike Taylor's coding journey - Learned pre-AI via bootcamps, free MIT/Harvard courses; ~5 years of evenings/weekends while running a marketing agency; ~6 months full-time coding before building anything useful (2020). - Was "just technical enough" to make a GPT-3 API call in 2020 — useful then, but says **"now you don't have to be technical to use these tools."** ## Is Coding Dead? **Mike Taylor's position (with internal tension preserved):** - "In that very literal respect, coding is dead" — he no longer writes code; his role is between **product manager** and **engineering manager**, though he still reads and reviews a lot of code. - Claims many early adopter companies report AI generating **100% of their code** — he says he's "seeing that for sure." - **Counter-qualification:** There's still "a ton of manual code" being written, and that makes sense for **legacy code bases** where the AI lacks organizational/historical context. "I still think there's a case for coding." - "What we're trying to do with coding is never going to die" — he learned to code to solve problems, not for the craft. People who coded for the craft are "mourning" its loss; he coded because he couldn't afford a developer, and now sees Claude as a "$200-a-month developer that doesn't sleep." ## Paradigm Shifts in AI (Mike Taylor) He frames adoption as distinct leaps, each roughly as large as the previous: 1. **No AI → ChatGPT** 2. **ChatGPT → Copilot** (AI can see the document) — "as big a change as no AI to ChatGPT" 3. **Copilot → agents on your computer** (Claude Code, Codex, Gemini CLI) > "It's almost like an employee that you're managing rather than someone that you're working together with." Further, "more experimental" paradigms exist beyond that, which he jokes might "scare people." ## Barriers for Beginners (Mike Taylor) - **Jargon/language barrier** is a huge problem — e.g., "pull request" is an anachronism since you *push* code. Comparisons to landing in a foreign country not speaking the language. His (joking) claim: "developers get paid a lot of money because they're willing to put up with all this complexity." - Terminal was formerly a blocker; the **desktop app** now reduces that barrier (Kate notes the shift happened within months). - You don't strictly need to code to use Claude Code, but you need someone to walk you through setup — though **once it's running, you can ask Claude itself** these questions. - **Course difficulty:** every cohort spans people who've never opened it to people with 10+ hours who "lied on their intake form"; hard to serve both. - **Surprising observation:** C-suite executives at offsites are often willing to get into the terminal — "a surprising willingness to get more technical" from people who'd never have opened one. - 80–90% of needed work is doable in the desktop app without looking at code. ## On the Course Structure - "Claude Code in one day": wouldn't have been possible a couple of years ago (references the old "Decoded" one-day coding course — which, he qualifies, didn't really teach coding in a day). - Actual outcome: being **comfortable** with Claude Code, so you ask it before jumping to ChatGPT or an app. - ~50% build time; TAs swarm the chat; breakout rooms for setup failures (with ~100 first-time users, "something always goes wrong"). - **Who it's for:** sweet spot is between "never opened it" and "more than 10 hours" (Kate: rule of thumb "under an hour"). If you've built a product or automated a workflow with Claude, "you've kind of aged out." - **Why a live course vs. YouTube/recordings:** the value is activation energy, social energy, and live TA help. Team unanimously advised a Europe-based would-be buyer wanting only recordings not to purchase. Recorded content works when you're already bought in (e.g., boss requires LangChain); the live course is for getting over the hump. - Claude is itself used to generate **custom practice data** for challenges (e.g., fake Salesforce data, role-personalized project ideas for executive offsites). ## Common Beginner Use Cases (Mike Taylor) - **"Organize my chaos"** — Gmail/Slack connectors, daily briefings, scheduling, meeting prep (his consulting team has no EA/ops people; Claude fills that role). - **Financial data analysis / custom dashboards** — previously required two full-time data scientists and weeks; now he can build one live on a client call. ## Skill Levels & Progression (Mike Taylor) - Very different mindsets between absolute beginners and agent-swarm users; "you cannot talk about agent swarms in the same breath as 'hey, meet the terminal.'" - He personally runs three agents 24/7 and uses Kieran's "compound engineering" plug-in; tells his advanced friends **not** to join the beginner course. - **Don't skip steps:** if not using ChatGPT yet, start with the chat interface; if not using Copilot/Gemini in documents, Claude Code may be a leap. Self-awareness of "where you are on the diagram" matters most. ## Persistence vs. Technical Skill > "It's a persistence gap… there's no such thing as a technical person, just a persistent one." - He failed to learn coding multiple times; succeeded only with 3 months of full-time time during COVID. - Attitude toward errors as learning opportunities — credits friend Rob Desmond's "great, we got another error" mindset; errors with Claude are content opportunities for Every. - Kate adds: paste the error message back to Claude — "constant dialogue." ## Security & Guardrails (Mike Taylor) - Personal setup: Claude with **root access, "dangerously skip permissions"** on a couple of computers — justified as trust built over time. - **Qualification:** for sensitive/production code bases he uses guardrails and developer sandboxes; security thinking must be "specific to the project." - Trust analogy: don't give a new employee the corporate bank account on day one; his first marketing job took 3 months to get publish privileges. - Waymo analogy (Kate shares the experience): terror at first, trust within 10 minutes. - **AI as security solution:** use more AI — e.g., a second session: "an intern on my team just wrote this code, can you review it?"; "brand police" AI reviewers; Kieran's plug-in deploys ~12 sub-agents including a security expert — he trusts that output more than his own review. - For regulated clients (healthcare): human/legal review layers are required. ## Desktop vs. Web - Prefers desktop because it can work in a **local folder** — enabling multi-session workflows ("do this piece of work, save it, start a new session from the folder"). - "Cloud Claude" (GitHub + pull requests) carries developer etiquette; using it well amounts to "learning how to be an engineering manager." He uses a combination. ## Keeping Up with Change (Mike Taylor) - "Nobody feels like they are" keeping up — including himself. - Grounding heuristic: walking down a Brooklyn street, someone from the 1970s/80s would notice only **iPhones** as different — filter for what's actually impactful; only ~10 truly impactful things in the past 5 years. "What still matters tomorrow?" — worst case you're a couple months behind the frontier. ## What to Use AI For - **Reframes the question:** not "what can I use this for?" but "what *can't* I use this for?" — find failures; "that's where we will continue to be employed." - Current known failure: **PowerPoint** is still "pretty bad" (and the deck template isn't AI-friendly); retry failures with each new model. - **Biggest mistake is almost always context:** too much (context rot, distraction, conflicting info — "also true for humans," his column's theme) or too little (making it guess what you'd brief an intern on). ## Creative Fields (Mike Taylor) - Surprises creative people (music, writing, design) most that **it can actually help** — they're not used to solving problems by talking to an AI. - Claims humans have "better taste than Claude," but Claude knows the APIs; AI music generation (Suno) has "a lot of potential." ## Forecasting (Mike Taylor, hedged) - "I'm able to forecast, but I don't know how much you should trust the forecast." Timing is the hard part; utopia/dystopia framing is wrong — "the answer is probably somewhere in between." - **High-confidence claim:** soon you'll work with more AI coworkers than human coworkers — **not** because humans are let go. - Skeptical of AI-layoff narratives: "I haven't really seen anyone legitimately fired because of AI"; companies announcing AI-driven layoffs are often "covering dysfunction" — an excuse like blaming the economy, since "other people [are] hiring in the same economy." ## Executives (Mike Taylor) - Teaches them much of the same content; existing managers already possess the skills (briefing, context, evaluation metrics) needed to manage AIs. - Example cited: **Toby from Shopify** "committing more code today than he did when he was solo coding Shopify." - Pattern: executives adopt **slower** (rightly skeptical — believing the hype has radical implications), but execute **faster and more aggressively** once convinced. ## Housekeeping (Kate Lee) - Next cohort of Claude Code for Absolute Beginners: **April 14**, launching officially the following day; details to be posted in comments/email. # Speaker Attribution Record ## Video Metadata - **Title:** What Happens When Beginners Start Building With Claude Code—With Mike Taylor and Kate Lee - **Format:** Live stream hosted by Every (3:00 PM Eastern) ## Identified Speakers | Speaker | Role | Evidence | |---|---|---| | **Kate Lee** | Host; **Editor-in-Chief of Every** | Self-introduction: "I am Kate Lee. I'm the editor-in-chief of Every." Acts as moderator/interviewer. | | **Mike Taylor** | Guest; **Head of Tech Consulting at Every** (newly full-time), columnist ("Also True for Humans"), instructor of Claude Code courses, former prompt-engineering freelancer, Udemy instructor | Self-introduction and Kate's introduction. | ## Roles/Parties Referenced (not speaking) - **Natalia** — runs consulting at Every - **Brooker Belcourt** — finance-side consulting at Every; formerly led finance vertical at Perplexity - **Kieran** — colleague, "compound engineering"/Corra plug-in work - **Rob Desmond** — friend who taught Mike Taylor to code - **Toby from Shopify** — cited example of an executive "vibe coding" - **Unidentified live-stream audience** — questions read aloud by Kate from the sidebar/chat ## Promotional Segments (Flagged) - **Opening:** Kate describes Every's subscription offering (newsletter, "five apps," trainings for enterprise and beginner levels). - **Closing:** Announcement of next **Claude Code for Absolute Beginners cohort, April 14**, launching "tomorrow," with sign-up link promised in the comments; mentions course price of **$1,000** during Q&A. - Recurring promotion of Every's course/consulting offerings throughout (taught cohorts, executive offsites, curriculum). ## Referenced Third-Party Products/Sources - Claude Code / Claude Co-work / Claude desktop app, ChatGPT, GPT-3, GitHub Copilot, OpenAI Codex, Google Gemini CLI, Copilot Studio, Suno, Waymo, Salesforce, Gmail, Slack, Google Data Studio, Harvard/MIT free online courses, "Decoded" coding course, LangChain, Udemy, Flatiron boot camp, Every MCP. ## Third-Party Claims (attributed to Mike Taylor, unverified) - "Companies are saying that AI is generating 100% of the code now" — reported claim among early adopters. - AI-related firings often "cover dysfunction" — Taylor's opinion. - Toby of Shopify "committing more code today than when he was solo coding Shopify" — cited by Taylor. ## Comment/Interaction Availability - **Live audience Q&A confirmed:** Kate reads multiple questions "from the sidebar" / "from people in the stream." - Post-stream sign-up link promised "in the comments." ## Uncertainty Notes - No curated metadata supplied; all attribution derived from transcript self-introductions. - Audience questioners are unnamed and recorded only as relayed by the host. this live stream that uh we are hosting from Every. Uh it's 3:00 Eastern time. We're excited for um more of you to join us as you come on. Um I am Kate Lee. I'm the editor-in-chief of Every. Every is the only subscription you need to stay at the edge of AI. We offer ideas, um apps, uh and training. And uh that includes a newsletter, that includes um four, well, five as of last week, apps, uh productivity apps, and trainings for both enterprise uh and corporate level and uh beginner level students. I am here with Mike Taylor, and I'm really thrilled to be here with Mike, who's uh just joined us full-time to be the head of tech consulting. Um Mike's actually been a writer for us for several years, so I've had the pleasure of working with him for a long time already, and I'm now uh even more excited to work with you on a full-time basis. Why don't you go ahead and introduce yourself, Mike? Yeah, thanks, Kate. And um I just uh cringed actually as I looked at my title, and I realized I capitalized tech consulting. I know that you would hate that. \>\> \[laughter\] \>\> Warms my heart. Thanks I I I I can recognize my mistakes. No, um yeah, so so yeah, I've been writing the column um Also True for Humans uh for the past couple of years, which has been really fun, uh and I've been freelancing as a prompt engineer, I guess, uh depending on whether you want to call it that, uh for the past 5 years. Um but yeah, it's only been a couple weeks now, but uh but yeah, full on uh in the consulting team. I've been already a couple weeks now, but uh but yeah, full on uh in the consulting team. I've been already been to San Francisco, Miami, uh done a bunch of projects already, uh but um also kind of in the lead and did did a few projects uh over December and January as well, so feels like it's been longer. Great. And can you just say can you share any more about what it means to be the head of tech consulting? What does that actually mean? And you will be writing more about this, but like what does that mean for your job? Yeah, so um head of tech consulting, we we have uh relatively small consulting team. People might be surprised, but we have Natalia who runs consulting. And and then working with Natalia is myself on the tech side, and then we have a guy called Brooker Belcourt who's on the finance side. His background is in finance. He used to lead up the finance vertical for Perplexity. So, he deals with hedge funds and you know, investment banks, companies of that nature. And then for me, I'm mostly working with engineering teams at tech companies as well as operations or you know, marketing teams at those companies as well. So, I focus much more on you know, the kind of coding agent type stuff as well as you know, how do we start to automate knowledge work for the type of work that tech companies run into. Great. Well, let's get started. So, Mike has been teaching some classes in addition to doing trainings for for big companies. He's been teaching classes particularly around cloud code. He's taught cloud code for absolute beginners and then a more advanced course. And even just in those two cohorts, I think he's learned an incredible amount about what is working, what isn't, and what is the state of AI adoption. So, we thought this would just be a good opportunity to talk about what what he's learned and what he's seeing out there and what you might be able to pick up from this as well. So, to get started, can you just talk Mike about your own coding journey and what is different in pre AI you know, between pre AI times and now? Yeah, so I don't know if it's I was fortunate or unfortunate to learn code before AI. But you know, I did a bunch of bootcamps over the years. I did like some of the MIT courses and Harvard courses that were free online. Um I was I was running a marketing agency before I learned how to code and I spent like 5 years evenings and weekends trying to do this. Uh then uh finally like quit, went full-time, um quit my agency. I was uh you know coding full-time for about 6 months really before I started to build anything useful. Um and then and that was in 2020 and I was just happy to you know get into um a situation where I had time to play around with GPT-3. That you know so so like I was you know just technical enough to be able to make an API call basically, right? So uh that was super super useful. Um and uh now obviously you don't have to be technical to use these tools. Um but back then you really did and um and I would say that to some degree there is an advantage to being technical uh with with these tools. Like they're not always very beginner-friendly or user-friendly. Um and uh there's this big barrier there. I I wouldn't say necessarily that um you have to know how to code to use Claude Code um but um you definitely have to have someone to kind of take you through that and push you through and say like this is the terminal and like here's the you know here's how you navigate to a folder and here's how you you know here's what Claude is doing for you under the hood. I think like that's uh that's still a big barrier for people. Um and it took me years to to learn that uh the the hard way. Uh but now um you know the funny thing is like once you get Claude Code running um you can just ask Claude a lot of these questions. That's right. \[laughter\] Uh so so like it's getting him over that hump I think like uh you know that's that's the big the big blocker. Absolutely and and we can talk more we'll talk more about what what that actually look like. Um so I'm sure a lot of people ask you or at least the you we've all read pieces about like is coding dead? What is your view on that? Yeah, it's funny because um to some degree uh like I don't I don't write code anymore, right? Like I'm not like I I uh am probably closer to product manager in terms of what I'm doing day-to-day. Um but I still read a lot of code. I still review a lot of code. Um so yeah, maybe it's like somewhere in between product manager and engineering manager. Um but uh so so in in that very literal respect like coding is dead. \>\> \[laughter\] \>\> I think a lot of companies are saying that AI is generating 100% of the code now and I'm seeing that for sure amongst early adopters. Um that said, you know, there's still a ton of manual code being written and I still think in a lot of industries that makes a lot of sense, especially when the code base is um you know, very um legacy and and there's like a lot of context that the AI doesn't have about your organization and historical things that you tried. I think that that can be really really important. Um so so I still think there's a case for coding. Uh that said, um like what what we're trying to do with coding is is never going to die. Yeah. \>\> Uh like you you you like you're I I really learned how to code cuz I wanted to solve problems and uh the problems I wanted to solve uh just like required me to write code to solve that problem. Uh and now if I can just use cloud code to to do that step, uh then I can solve more problems. And that that's I I think people who learned how to code for the craft are really struggling right now because they're they're almost like mourning this loss of uh of of of something they loved. Whereas like for me, I only coded cuz I had to because I couldn't afford a developer. You know, that \[laughter\] was that was the main thing. And so like now that I can afford a a developer that's $200 a month that doesn't sleep and and can write better code than I could ever write, then then I'm I'm off to the races. But I can totally understand that like a lot of people wouldn't be. Right. Right. That makes a lot of sense. Well, you know, you have also taught the class on prompt engineering. You taught the Udemy class on you know, several hundred thousand students. You've taught them. Yeah. And I know we've talked a lot about like what you know, is prompt engineering still a role, still a job? These things have just sort of morphed and changed and tend to be a little bit more horizontal than vertical in terms of the skill set that they that that that they enable for across different jobs and different roles. But now you have been teaching Cloud Code for Absolute Beginners. You led one cohort in February in late February and then you led a more advanced cohort not for beginners. So just clear that was not for beginners the second one. Um last week and curious what are you know, we've we've just talked about sort of the the what is what is what are some common blockers that you have seen from this past cohort and just boot camps in general for people to get started doing this? Yeah, this has been a really interesting experience because previous to this most of my courses were not like live. I had done some corporate training for a few big companies where they wanted like a custom thing but but like you know, I had 300,000 students in my Udemy course but like you know, I never met any of them. They just watched the videos that I recorded. So it's a very very different skill set. And what I realized this gave me a little more empathy for you know, like going slower and getting to the point where like some of these things are just not not obvious in any way and they're poorly named. Like one that came up recently in the course was why did they call it a pull request when you're checking in code? You're actually pushing the code to GitHub, right? So so these these things are like anachronisms and and to some degree developers get paid a lot of money because they're willing to put up with all this complexity. \>\> \[laughter\] \>\> And most most people are not. So any any time someone dives into the more technical role or or more technical topic they they bump up against the these these unfamiliar terms. They They actually like pretty straightforward to understand once you get over that, but it's like almost like, you know, uh landing in a foreign country and everyone's speaking a language you don't understand. It's like very difficult to do anything like feel competent in what you're doing when when you don't speak the language. So, I think like the the language side of it is still a huge huge problem. Right. And I think even in the past couple of months, uh and different different cohorts that we've taught and you've taught, so much has changed. You know, 6 months ago or 4 months ago, you were teaching in the terminal, um and now in this last, you know, in this most recent one, you can just teach in the desktop app. Um and I think that that that kind of change is happening really fast and that something that is the terminal, which is not necessarily uh native to non-coders, can be a hard hump to get over. And now that's not even something that um is really what we're we're orienting around. Yeah, we we try and get this trade-off right, um and I think it's hard to get right because in every class we have someone who's like, you know, barely opened never opened code code, right? Um and then we'll also have someone who's spent more than 10 hours and they've they've they've lied on their intake form. That's not a beginner. That's not a beginner. \[laughter\] Yeah, they're just like they just want to see what you guys are teaching and and and unfortunately like it's it's very hard to create content that matches both of those those groups. So, so we we we try and try and get get it right. Um I I would say that, you know, as well as these courses, I've also done a few executive offsites recently where we've been teaching, um you know, through the consulting kind of org, I've been teaching, you know, a room of C-suite people like who who, you know, haven't you know, they've had people to do tasks for them for a long time, right? Because they've been, Um, know, working their way up through the management ladder. Um, and I've been I've been really surprised that they're actually like quite often willing to get into the terminal. So, so I would say that we we try and make it easier for people to use Claude and get like the I guess get get the outcomes that they want without having to add extra complexity. And you can do 80 to 90% of what you need to do in Claude Co-work in the desktop app. Um, without even having to kind of look at the code. Uh, but I would say that I'm seeing like a surprising willingness to like get more technical. Um, from from people who traditionally like wouldn't wouldn't have ever opened a terminal. Yeah, well, I would count myself among those people who now regularly in the terminal, but but wasn't not not so long ago. So, we've we've said that this course or what you've what you've been teaching most recently and what we're going to be doing again in the future is really like Claude Code for beginners. I think the question I would say is one, why do you why do you think it's it's best to start there? And and is there something you've observed from that cohort that is like particularly amenable to starting this kind of class? Yeah, so I think that there have been these larger paradigm shifts within AI. Like obviously there's something new happening every week and it's hard to keep up, but broadly speaking there was a big leap from no AI to chat GPT, right? And then another big leap from chat GPT to Copilot. I mean that both in terms of GitHub Copilot in your code, but also just the Copilot app. Like basically this paradigm of it can see the document, too. And I think that going from chat GPT to Copilot is actually like as big of a change as going from no AI to chat GPT. Uh but then you have this next leap, which is uh basically working with an agent that's on your computer. And that's where I'd put Claude Code. Uh that's where I'd put Codex. Uh there's also the Google uh Gemini CLI version. Um but uh there's a there's a really big difference, and it's it might be not uh not necessarily obvious to someone who hasn't tried it, um between like what Claude Code can do uh and what like Copilot type app can do. Uh Claude Code can uh do um it's it's almost like it's an employee that you're managing rather than someone that you're working together with. Uh I think that that's like a very different uh paradigm. Um uh and uh I think, you know, even people who are really into AI, it might not be obvious to them until they see what what sorts of tasks are are are are now achievable, and how many different things can you pull together into the same into the same session. Um so so I think I think like starting with that and getting getting them to have that aha moment um is is really important. And there's we can get into it, but there's like, you know, there's further paradigms that are like more experimental past that. Uh if if we want to like uh scare people. Hopefully we'll have time for that. Um so just thinking about this course this sort of, as I said, the two things we say is it's for absolute beginners, and you're going to learn Claude Code in one day. Um can you take us through, Mike, what does this day look like, and what is what is the scope of what you're actually teaching? Yeah, so first of all, like this basically wouldn't have been possible a couple of years ago, right? Um which is is pretty nuts. Um uh like I I remember there was a there there was a a a course, uh I think it was company was called like Decoded, uh that was really popular for a while where they were teaching you how to code in a day. And um it was a great great course. I I know a few people who did it, but it you wouldn't didn't really learn how to code in a day, right? Like uh you learn what JavaScript was and what HTML was and CSS and you could have you had like to their credit you had you put something live on the internet by the end of the day. But, I would say that you could you know you could learn to how to code in a day but you could also learn how to code in 10 years. \>\> \[laughter\] \>\> So, it's like it depends on how what you mean by learn how to code, right? I think to some degree there is an element of that with this course. But, what we actually I think what you actually do have by the end of the day is you are comfortable with Claude code. And and once you're comfortable with Claude code, you get into the habit of asking it to do things before before you you know jump to chat GPT before you even jump into an app yourself. So, through the course we basically like get you over that hump. We get it installed on your computer. We have a bunch of TAs that swarm the chat that will basically like walk you through how to get things set up and we we even kind of push people into breakout rooms off the main thread if they if they need extra help cuz something always goes wrong. Like, you have you know 100 people who have never set this stuff up before and and like they might have something different on their machine that like we've never seen before, right? So, so so I think like that's a really big part of it. And then we we make it it's about 50% build time. So, so a lot of it is like we're giving you the direction and then we're we're you know giving you the the introduction to the concepts and we're showing you what we would do but then you get to actually build something. And then you're you're putting it like you've got it you got it like a complete done by the end of the day. And and once you've done one of these things then you start thinking about all the other ideas. So, I would say that it's just that necessary step to get you to the point where you have it running you've done a task. You maybe set up some connectors as well so that the results you're are useful from Claude. Um and uh uh and now you can just you're you're armed. You're ready to reap for it. And uh by the end of the day once you once you've done all of that, um you don't necessarily even need uh an instructor after that because Claude becomes your instructor. Like you can you can ask Claude like when you get stuck next time, right? So So I I I think like it's um it's it's a really big It's more of a mindset shift, I think, than than around the like any kind of secret information we have. Um like a lot of people ask me, "Hey, um why don't I just do YouTube?" And I'm like, "If you want to go do YouTube, you should 100% go follow a YouTube thing and work through it." Um and um and and and you know, be our guest, right? Like you should. But But like I would say that um you know, all of the information um for like the Harvard uh computer science course is online for free as well. Uh so why would you go to Harvard, right? Like I think it's like I'm not comparing us to Harvard, but but I think the big reason why you do a course like this is for the kind of um physical kind of emotional um like motivation like reaction that you get uh when you're in a room full of people all trying to do the same thing. And uh you have like people who actually care about the outcomes of the day uh there on the chat who can who can walk you through it. So So I think like that if if that's what you're if that's what you need, that kind of activation energy to get started, then it's a it's it's worth it for you. Mhm. Can you um Can you recall what are some of the things that maybe someone said they wanted to start out to do at the beginning of the day? I would love to learn Claude code to be able to do this in my job or something similar. You know of an example of that? Yeah, we have a bunch of examples that uh commonly come up. Uh one is like organize my chaos. Uh basically, um you know, plug it into my Gmail and to my Slack. Like give me a daily update. Tell me, you know, what what I'm what I'm doing today. And um that's a really common use case that we use a lot in the consulting team. Like you know, we don't have an EA, like we don't have any kind of ops people in the team. We're literally just like using Claude for that. And we're you know, doing scheduling, we're doing meeting notes, we're doing like you know, prepare me for my meetings. Like where do I need to go and when? And like what what what do I need to look at in Slack? You know, I think like that that's like a really big one that usually blows people's minds when they can do it. Another one is like financial data analysis or Excel type stuff or like kind of building custom dashboards. That's a really big one that comes up like hey, just export this data. I want to kind of see it in a custom dashboard. And that's the sort of thing that I used to have a team of two data scientists who you know, were full-time like doing this stuff all day every day. And you know, it would take weeks for them to give me a custom dashboard for a client. Whereas like now or I'd have to play around Google Data Studio myself like trying to do it in between meetings. And and like now I can just like I just ask Claude. I'm like here's some data. You know, tell me like make a custom dashboard. And then I could like I could do that live on the call quite often with the client. \>\> \[clears throat\] \>\> And and that's that's pretty rare. Like how how is the you know, what's the NPS score from the latest thing? You just go okay, let me just ask Claude and then boom like it's there, right? So I think like that's that's another big one that comes up. Got it. And so over the course of the day, is it just in terms of getting people set up and getting started? Is that something where everyone is sort of working together on the same problem which is getting set up and then they you know, and you said there's 50% is devoted to build time that then the rest of the day everyone is sort of working on their own project but being guided by you and the other TAs. Yeah, exactly. So we we we set specific challenges. We you know, we might change this up Um over time. We're always changing after every course really also because Claude keeps releasing new stuff. Yes. \[laughter\] But but like I think you know, one one of the one of the things we're starting to do more with courses is create like a wider selection of challenges they can work on because when it was me just manually doing this for previous training sessions, I would have to like if if you want to create a financial dashboard, then you have to like provide the students with some financial data to use, right? And unless you want to share like I I think for a previous every course you guys have shared some NPS data from from dashboard and you're happy to share that but like what do you do if you can't share that, right? And so so one thing that Claude is really great for is customizing like data for for those ideas. And even for like some of the executive offsites we've done, we've even made like three custom ideas that they could work on like each based \[snorts\] on the job role like personalized to the job role. And like that's something that you know, I would never have the time to do without Claude but but I feel like we're trying to make these sessions more personalized if possible and I give them a wider range of options to build from and then giving them some custom data that they can use to plug in like if you want to automate I don't know like a salesforce workflow, then you might not have access to salesforce in that session, right? Like IT team might not have given you access in time. So we come prepared with like here's you know, here's a bunch of example salesforce data that you know, when when you do get the connector it'll work the same way. Mhm. Now if someone is thinking about this course or wants to take this course and is like I have literally never opened Claude code. Yeah. Is that where do they start on, you know, the morning of the class? Perfect. Yeah. It's Yeah, I think it's that sweet spot between I've I've literally never opened it and like I've spent more than 10 hours. And so I would say um you should like in order to avoid any of the potential chaos on the day and and like missing out some of the material, you should come with like you should try and get that code code running. Like literally just prompt and say hello, like what's the weather like in Hoboken or whatever, you know. If you've done that, then if you go into that point, then then you're going to be able to follow and follow along the whole day and you'll get I think a ton out of it. But it might might take a bit of time to get to that point. So so we we try and try and tell people to kind of walk through that and we we have some steps in our boarding guide there. But that said, if you do turn up and you've literally never opened it before and it's not running on your machine, like the TAs can help you work through that. That's part of the value that we provide. But I'd say if you've if you've built a product with Claude or you, you know, have already automated a workflow with Claude, like I would say that then you've kind of aged out of this course as well. You don't want to like I think it it would be you know, probably better for you to do one of the more advanced ones. Right. Right. Yeah, we've sort of been using the rule of thumb of like under an hour with Claude code is is this and if you spent more than that, you know, then then it might not be the right one for you. I see we've got a bunch of questions in the sidebar as well, which I will which which I will get to shortly. So we you sort of referred to this a little bit, but I'm just wondering you know, over these past couple of cohorts, have you you know, in sort of teaching people across skill levels, like the people who've never opened ChatGPT or Claude to people who are orchestrating swarms of agents you know on their own. Um, are you finding that there is sort of um, can you speak to both of those groups in the same time or is this really they're very different mind sets? Yeah, very different mind sets. Um, so uh, you know, day-to-day like one of the things I'm doing is I have like three open claw um, like AI agents basically running 24/7 for various reasons. You know, I'm I'm using like Kieran who we work with on Cora like he has compound engineering this kind of plug-in that that we use day-to-day to to basically scale up the amount of development work we can do without introducing too many bugs. So like that's where I'm at on a day-to-day basis. And and when I'm kind of sharing advice between friends that's that's kind of where we're living. And I would tell like all my friends like not to join this course specifically because it's not you cannot like talk about agent swarms in the same breath as you're talking about like this is hey meet the terminal, you know. Like if you're not already completely AI degenerate like me, you know. So so so I would I would say um, it really depends on your personal comfort level but but like that I would say like a lot of that like the really advanced stuff like doesn't exist anywhere actually. Like it's it's \>\> \[clears throat\] \>\> and like people are talking about it in events and things like that. But but like we want to try and get more of that into into courses and and into the content as well. Like you know, definitely for my column like getting getting more of that material out there. But yeah, I would say I would say yeah, broadly speaking you want to really think about like where you are at on this journey. And it is totally fine to be you know, wherever you are on the diagram. The important thing is that you just like know where you are, right? Like you are you are you at the point where you just like you you've not you're not really using Chat GPT, in which case I wouldn't go straight to Claude Code. I would say like start with the chat interface and start getting used to that, right? Like if you don't have, you know, Copilot open in your documents, or if you're not using Google Gemini in your documents, I would also say like probably like it might be a bit a bit of a leap to go to Claude Co-work or Claude Code. And I would say you you don't want to skip a step. Like get comfortable with that step. And I like I had the benefit of doing that incrementally because I started in 2020, and I had to make an API call to GPT-3, which was a pain. You know, but but then Chat GPT came Chat GPT came out, and then I started experimenting with more stuff. And I've kind of seen these things in progression, so they make sense to me. But I think if you if you're now just getting deep into AI, like just allow yourself that that that progression as well. Otherwise, I think it's it's pretty easy to spin out and and like feel like you're overwhelmed and not getting value from from from the material that you're looking at. Got it. We have a couple questions from from people in the stream that I'm going to go to, and then we can come back to some other things. One question is do we need to download the desktop app, or can we work with it entirely on the web? Yeah, good question. I I like to download the desktop app specifically for one reason, which is then it can work in a folder locally on your computer. And I think that's one of the major benefits of using something like Claude Co-work. So, you know, not even even if you're not getting into Claude Code yet, I would say you can experiment with Claude Co-work, and you can connect it to a folder. And that that will give you a bit more of a sense of why this is special because you don't have to do things in one chat session. You can say, "Okay, do this piece of work and then save it in the folder and then you start a new session and you say, 'This piece of work is in the folder. Like, start from here and keep going.' So, I think like that's a really big paradigm shift and and and without having the desktop app or having the terminal on your own computer, you kind of miss that. And you also have this extra setup step of like you have to learn GitHub which is a whole topic in itself. And some of these still like a lot of senior engineers mess up. So, So, so yeah, I would say that like cloud Claude, as I call it, where it's like running from GitHub and and then kind of giving you a pull request, there's like a whole like developer etiquette around that that you need to like learn actually how to be an engineering manager really to make use of that very well. I use a combination of the two, but but but yeah, I think I think it starts with the terminal and starts with you know, the folder on on your computer. Right. Um, another question is that Mike, in all of your teaching, you've seen people go from zero to building, you know, building with tools and with confidence. And what's the difference between the ones who have been able to actually ship a project and the ones who stall out? Is it a technical gap or is it a process gap? It's a persistence gap. Um, yeah, I I I like to say there's no such thing as a technical person, just a persistent one. Um, because like you become technical just by like running into these barriers and and like not stopping. Um, and it's easy to get discouraged though cuz I I tried and failed to learn how to code myself multiple times over over the years. So, it was only really when I like had 3 months full-time to do it that I was like, "Well, I've got to do this. I have literally nothing else." And it was during COVID, so it was like actually \>\> \[laughter\] \>\> I was like, you know, I've got to become employable now you know, cuz we don't know what's going to happen. So, so uh, it was uh, you you have to have the either the motivation to uh push through those those barriers when you hit them and and be like happy to see an error message uh because you're like great now I'm going to learn something new. Um, actually like I have a a friend who taught me a lot of how to code a guy called Rob uh Desmond and he um uh he uh he used to say like it was like almost annoying cuz he was like he's like great we got another error and I'm like dude like how are you so happy about this? feel like a failure and you know and and now I think about that myself I'm like oh great like we got another error like when I when I get an error now with Claude I think oh great like there's an opportunity to figure out something the models aren't good at yet uh and like that's content for every right? Like \>\> \[laughter\] \>\> so so yeah I feel like like having that positive attitude and and kind of keep going like the persistent kind of uh approach I think like that's actually uh the important thing not not necessarily like where you are right now technically. Right. I think that's something I've learned as well which is you get an error message you just send that error message to Claude and say what does this mean? What do I need to do? So you're sort of in a constant dialogue and and putting going back to it and saying no you tell me you tell me what I need to Exactly yeah. Um, I know there are a lot of questions in general around some security and access and things like that. What kind of access do you give Claude code in your own systems? Yeah I think again this is one of those things where you shouldn't skip a step uh but like my personally I have it like I have like Claude running with root access uh dangerously skip permissions. \>\> \[laughter\] \>\> Uh on on at least a couple computers uh and and the reason is that I built up that trust over a long time. I've seen like things it can do. I I mean I I still think a lot about security um in terms of like uh like if I if for example I have access to a code base uh that is more sensitive or production level code base um then uh you know for that I would I would like tend to use more guardrails and maybe like a developer sandbox which is like a way to set aside part of your computer that cloud can only run in that part and then it can stop it from deleting the rest of your files or sharing your API keys or something. So I think a lot about security but I think I think it's it has to be specific to the project and you have to build the trust over time. Because obviously like if you just hire a new employee today, you don't give them like a corporate bank account on day one, right? Like I remember when I used to be in marketing it took us three months in my first job to get they call like publish privileges where like I could only send my boss like here the campaigns I want to put on Google AdWords and then my boss would put them on. Like I didn't have the right to like publish anything to the account yet. I kind of think about it that way like you know once you once you get to the point where you're just getting bored because cloud is so good at this stuff that you're just like fine let me it's it's actually safer than me like I'm just going to run it. Yeah. I had a similar experience with Waymo as well. I don't know if you've driven one of the self-driving cars. Yeah, it's it's like I was like oh my god this is kind of terrifying. Like I'm about to get in the car that's driving itself and and I noticed like within like 10 minutes I was just checking my phone and like not even looking at the road. It was like felt so safe because I was like I feel I feel that way with cloud as well. Interesting. Along those lines we have some people who are curious for non-technical people which is what this class is for is how is how can we do proper guard railing or sandboxing ways of developing solutions dashboards and apps? Are there any rules of thumb? Like giving it access but not full access. Yeah. So, um the uh this is a big question. Uh like we I have some clients in like that like a client in the healthcare space, for example. Uh so, they they have to think really carefully about this and they have to have uh certain levels of human review um for for things like that like legal review as well. Um so, so I would say um One thing that people don't really think about is that you can use more AI to solve the problems that AI causes. Um so, for example, uh if you're using AI to generate code um and uh and you're like worried that it's not going to be uh you know, correct or like it's going to cause like, you know, it's going to take the database down or something. Um you can you can then create another session and say uh an intern on my team just wrote this code, can you review it? Uh right. So, so like Uh it might actually pick up some things that you might have missed as a human. Um So, so I think like the best uh Uh people who are best at this uh like building trust with AI, it's it's quite often because they've set up these systems to basically spend more tokens. Uh like Kieran we mentioned, but you know, for his engineering plugin uh command engineering, we have like it deploys like 12 sub agents, I think he's up to now, where it's um like one of them is a security expert that looks at like the security app. Another one is like, you know, uh best practices for the coding language \[clears throat\] you're using. And there there's, you know, 10 others. So, um I I would say that like actually when I run that, like I trust the code more than if it was just me reviewing it. Um you know, so I would say that um you know, I've seen companies have like a brand police uh AI that that runs every time they try and uh send an email or every time they try and uh publish an ad, uh the the brand police AI runs and it just tells them whether it's on brand. Um you know, things like that. So, so I I would say that like don't look at AI as like the cause of the problem. Uh you can look at it sometimes as the solution to security problems as well. Mm, that's fascinating. Um we have a couple of questions just asking really that skill level, which I think is really important and I know and I know we've talked a lot about that as well. Um one is if there is a course that's between like zero and one hours and a course that's for 10 hours plus, what is the difference between the A, what is the difference between those course? And is there what what's in that gap? Is there a gap between that one and 10? \>\> \[laughter\] \>\> Yeah, good question. Um so part of it is I would just say that like if you're doing a full day, you actually have time to build stuff yourself. Um and the biggest complaint I get or not really complaint, but like thing that people mention as a friction point with AI for the consulting work I do, they say that they don't have time to play with this stuff. Like they don't have time to save themselves time. Uh yeah, I'd love to automate my job. I think maybe you you even said the same thing from the editorial perspective, but but but like that's one sneaky thing that a workshop does is it actually just buys you a half at least half a day or a full day to like step outside of your normal day-to-day role. Like nobody's going to bother you if if you're in a cloud code camp all day. Maybe they still will, but you can ignore them. And then and then you you actually get to build something without like oh, I've only got five minutes for my before my next meeting. So so I would say like that's that's a that's a really big difference is the build time. But I I I tend to think like pretty much anything could be taught in five minutes or five years or somewhere in between. It just depends on how deep you want to go. Like um I don't know, like I I'm pretty sure I could I could teach a five-year long university degree in cloud code even though it's only been out for just over a year. There's just so much like cuz to really understand cloud code, you have to understand actually like almost all software development and maybe even all knowledge work, right? Like to to like really go deep on like how do we train skills and like like one of the big things we we talk about a lot in the office is that this is like a new form of management science, right? Like learning how to manage AI agents is just as hard like just as difficult as learning how to manage humans and nobody's had that management training yet. So I I I really think that could be like a university degree in itself. It's fascinating. We just had a new a new colleague start today and did a lot of preparation with onboarding, which is what you would do with a human, but is not necessarily something we thought about with an AI. So speaking of someone and we're talking about all of these other tools and again back to levels, we just have a question as what's the class for someone who has created agents in Copilot Studio, GPTs in ChatGPT and websites in GitHub, but hasn't doesn't still doesn't know how to use Cloud Code. Yeah, interesting. So they already know how to code, but they haven't specifically adopted AI. Yes. \>\> \[snorts\] \>\> Yeah, interesting. I would say that um it it's like weird weirdly like all this information is online, right? Like um so so the way that I learned how to use Cloud Code as someone who was a developer is I just installed Cloud Code and I started asking it to do stuff and if it failed, then I looked at the code it wrote and I went, "Oh, that's weird. Why did it do that?" And then I realized, "Oh, you need to make a change like my prompt should have been better because I didn't include this context or oh, it can't do that because it doesn't have access to my Gmail or you know, whatever it is. So I'd say to some degree the reason you join a course is really to hear it from the people who are like living it and and get their direct attention to your problems so that like you can you can get over that hump. So I would say like first and best thing is like you just do it and start playing with it. If you already are technical, but I know that's a terrible like not a useful answer. So I would say the second best thing is make friends who gotten over that hump already. Like I I learned a lot from like different people I've met like in the office or you know at events or just online like actually in a really bad habit of like meeting strangers from the internet now \>\> \[laughter\] \>\> now that I moved to New York. So so like a lot of people I've followed on Twitter for a while I'm like trying to talk to them now in person. So so like you know and I think that you get a lot of information that way. But but then I would say like if if if neither of those avenues are working for you then um then I would say like rather than a course in particular I would say you probably actually might be better with like one-on-one tutelage. So so like I actually paid a guy who used to work at Flatiron the coding boot camp. I paid I paid him like 60 bucks an hour to teach me some like databases it was because I didn't understand databases pre-AI and he like made like a custom curriculum. I think I only paid him for like 10 12 hours something like that. So like for 600 bucks you know which is you know three months of full code I managed to like learn databases right and I think that was more engaging. So so I would say like yeah if you already have a lot of the technical skills and you don't have you know and you have like specific problems that you're running into it might be better to not do like a broader course if that makes sense. I I want I want like people to like basically do only do this course if it's really right for them. Totally. Totally. Oh, yeah, definitely. Um another question we got is for for teams and not engineers who are using cloud code, um can you recommend best practices for keeping shared business and strategy context updated? Um so, the context updated, not just the code. And is it Do you use GitHub? Do you use Google Drive? Something else? Yeah, this is like the million-dollar question at the minute that nobody has ever solved because to some degree it's You only really know that the context is stale when AI makes a mistake, right? Um and even then, like you have to have that context in order to to solve it. Like actually, we we had this problem at the marketing agency I ran. Like we had all of our company docs were like constantly outdated and and and you know, like even we had a bunch of standard operating procedures and the team I'd find out that someone in the team wasn't using them cuz they're like, "Oh, it doesn't really work that way anymore." And I'm like, "Have you updated it?" "I know, I don't have time." You know, so so I think it is like a universal problem that doesn't go away with AI necessarily. But I would say one thing you can do is aim for connectors over over over like content. So so um I don't know, like one thing you can do is like download like say say for every right. Like we have every MCP internally, right? So I could just download everything I've ever written on every by by like, you know, like it would just like keeping it in a GitHub repository. Or today, I just asked the MCP, like, "Hey, can you just get like my latest posts?" And it did. So so like getting directly accessing that data directly or accessing that context directly from from the live um uh tools, I think like is is is is one one way to keep it up to date. Mhm. Okay. Um so, one thing uh some things some of the the people in the chat have talked about that you and I know talked about a lot is like keeping up with the pace of change. It's very overwhelming. We have even said we have to update our curriculum uh for for everything new on a non-infrequent basis. How do you keep up with the pace of change \{slash\} are there recommendations that you have? Uh am I am I keeping up with the pace of change? It doesn't feel like it. \>\> \[laughter\] \>\> I mean probably better than most. I I mean honestly like nobody feels like they are. I I think um is is the truth um but but I I would say that um like one one way to ground yourself in this and I I you know I'm I'm always hyping AI but uh but you know if you I'm just in Brooklyn at the minute like we're both here but if if you walk down the road like if you're from the 19 uh 80s or 1970s um like if if you if you walk down the same road like what would you notice that was different? Um and like the only thing I can think of is iPhones. Right? So so even though there's been a ton of technological change with the internet and everything like I actually like very few things that are really that impactful in your day-to-day work and and that's how I keep myself grounded with AI. I just look for like what actually is impactful uh that I keep coming back to uh that like my friends keep coming back to and they say it's become like a core part of their life um and uh there are when you use that filter there are like only really like 10 interesting things that have happened in the past 5 years. Uh not not like you know because like when when when things come out like there's a lot of failed projects. There's a lot of um like interesting things that happened that we thought were going to be very impactful and they turned out not to be right. So I would say like you can cut across the noise uh just by really just like leaning back a little bit and saying okay like what what's still It's like No, like oh, this matters today." It's like what still matters tomorrow? I think is is is something you can you can use if if it is getting a little bit overwhelming. You can pull back a little bit and you worst you'll be like a couple months behind the frontier. Right. Right. It isn't too bad. Yeah. Yeah. Right. So, um I noticed that some people it's just hard to know again where to start. What am I going to use this for? What are some um things you've seen like even underrated ways of using AI that may not seem the most obvious but but are actually really useful? I I would I would say that the framing of the question is wrong. Like what will I use this for? I think it's um like what what what can't you use this for? Right? Like I think actually you go to try and find cases where where you where it's not working. Um I think that's a that's a more useful thing to discover uh because um that's where we will continue to be employed. Right. Um you you know, frankly like that's uh that I I look for cases like that all the time and I'm like actually kind of happy when I find something that code is bad at uh um still, you know. Um so uh so so what the the way I would flip this is I would say like try and use it for everything. Um and uh and then and then note down where it fails. And then when the next model comes out then then try that thing again and see if it works now. Um so like right now for example uh it's still pretty bad at PowerPoint. And uh that annoys me because I would like love to offload my PowerPoint deck creation to AI but um but it like you know, I I know that I need to spend like a good couple of weeks to to really like optimize that and get it kind of working okay. And to the point where I I need to like even modify the deck template that we use um because it's just not like it's not very like AI-friendly. Um but but like in 6 months maybe that problem goes away. In which case, you know, like great great. I can I can stop worrying about it. Um so so yeah, I would say try and use it for everything if you can. And then and then uh and then and then like when it when it doesn't work and you'll have you'll have learned a lot more than uh than if you just sit back and wait for a use case to materialize. Right. So, what are some um maybe some some like non-obvious mistakes that you've seen people make with with AI? I think we we can all know some obvious ones or or or know when we are we are seeing that. \>\> Yeah. Yeah, um so uh one mistake I would say is uh almost everything ends up being the same mistake, uh which is context. Uh so, either um you gave it too much context and uh distracted it or filled up the session and um uh there's this concept called context rot, uh where where basically like if the more information you put into the that chat session, uh the more likely it is to go off the rails. Like it's going to get distracted, it's going to get um confused, you're going to have conflicting information in there. Um, you know, so uh I would say that that's also true for humans, right? Like there's my my uh the name of my column. Uh but but um but I I I I really believe that. Like uh at the end of the day, um if if you fill it up too much with information, uh then uh it's going to do a bad job uh in the same way that if you give someone too much like you know, overwhelm them. Um, but uh also too little uh context. Uh so, if you haven't given it something that a um you know, someone uh you would give an intern uh to do that that work. Um uh then then you're you're like asking a lot of it. Uh you know, without you know, it has to kind of guess that context. Um, so so I I I would say almost always comes down to like what was in the context window at the time that you asked it to do that. Mhm. Okay. Um we have another question for uh for you about in terms of what you actually taught. So, for people who are coming from creative fields like music, writing, and design. This is what someone says. What's the thing that tends to surprise them the most about working with Claude Code from your own experience teaching? Yeah, that's a good question. Yeah, and I love it when people have like use cases like that. Cuz you know, obviously it's it's really cool like the first time you see it organizing your inbox or you know, giving you a daily briefing, but but like yeah, when when when someone's like making music with with Claude Code, I think like that's very cool, you know. So I would want to encourage people with maybe less corporate like interest to to to try this stuff. And and I think like that's where the jagged edge is right now. Like it'd be really interesting cuz like I think humans have better taste for example than Claude does. But but but like you know, Claude you know, knows all the APIs to like program you know, different things and and like AI does a pretty good job of generating like music like I don't know if you tried Suno. There's a lot of potential there, I think. So yeah, I would say that um I think that the thing that surprises people is that it can actually help. Because if you come from a creative background, you're not used to solving your problems by talking to an AI, right? Like it's it tends to be like you use software to make music and stuff, but but I I I think it's it's a little bit more alien than it would be for a knowledge worker who that maybe they've used Excel macros before or maybe they've like worked with a tech team before to like build a dashboard or something. You know, like I feel like it's it's more obvious for engineers and for knowledge workers than it is for creative teams and and they're usually positively surprised at how much it can do. Mhm. Got it. Um so, if you I'm sure people ask you this all the time, but can you see or say where is this all going a year from now, 5 years from now? Are you able to forecast in any way? Uh I'm able to forecast, but I don't know how much you should trust the forecast. Um yeah, I mean to some degree I saw like this coming because, you know, in 2020 I was messing around with GPT-3, and I was like, "Oh, this is kind of crappy right now, but I can see where where it's going." \>\> \[snorts\] \>\> Um I I think uh if if I the the timing is is the hard thing, right? Like understanding it is is it going to happen in 6 months versus 1 year versus 2 years. Um and and uh also, if you zoom out and like follow current trends, like you get into this weird place where you're like talking about utopia or dystopia. Um I think that the the pro- the answer is probably somewhere in between, right? Like um it'll be it'll be hard to know, but but I would say like one thing I'm really confident on is that um pretty soon uh we'll have um you'll be working with more AI coworkers uh than you are um human coworkers. And it won't be I don't think it'll be because they let all the humans go. I think it'll be uh I I actually haven't really seen anyone legitimately fired because of AI. Um I've I've seen a lot of companies announce that it was because of AI, but really I think a lot of that is covering dysfunction. Um uh and and it's a good excuse, right? It's like it's like firing people because of the economy or whatever. Like um I think uh you know, there's other people hiring people in the same economy, right? Um so uh so I would say I would say like that that's actually the bigger thing is that um we'll have like as soon as AI coworkers become more stable and and usable, uh like it might be open core, it might be something else. Uh, then then you'll see it very quickly proliferate where you might be not just like managing uh, working with an AI, but like managing a whole team of them. Mhm. Mhm. So, I know you do a lot of this for executives and and companies um, as well. So, what is like you know, what is the version of this that you treat that you teach to executives and how is it different? Yeah, I think for executives um, I actually try to teach a lot of the same stuff. Um, just \[snorts\] because I think it is important for them to really see what their people are doing. Um, and what what the like what the hype is about. Um, because at the same time that um, now everyone is in is a manager of AIs, uh, I also think that like people who are already managers uh, are pretty well uh, established in terms of like the skill set they need to manage AIs. And you're seeing even in some cases you have like executives who are who are like vibe coding. Like Dan, we could we could we could talk about, but uh, but but like uh, you know, Toby from Shopify is a good example. Uh, like he's actually committing more code today than he did when he was solo, you know, coding Shopify, right? Like before he was an executive of a publicly traded company. Uh, you know, so so I I would say that like um, all the things that people struggle with with AI, like context switching, understanding how to like brief them properly, uh, you know, like um, setting good evaluation um, metrics. How do we measure good performance? Like actually managers have been doing that forever um, in their careers. And and so I think they actually have an advantage when it comes to managing AIs. So, I I think like once I ex- they're usually um, take a bit more uh, to to like learn the technical side and and to get Claude um, like really like doing something useful for them. Uh, but once they get there, they immediately understand it and then they go much more aggressive, much more quickly than the average person. Um, because I think um, executives are I I think uh, you know, for a good reason, like uh, they want to be skeptical of this stuff um, to some degree because um, if if if they believe the hype that actually um, that that that's like a pretty radical thing like for it has a lot of implications for their whole company. Um, but then as soon as they they can see through the hype and they try it for themselves and they see like, "Oh, okay, I get it. It's like it is actually pretty amazing, but like it can't do this, this, and this yet, and we need to upskill here, here, and here." Uh, then they get like quickly into the action. So, I think like that's the difference is like they're slower to adopt, but also like faster to execute once they've adopted. Got it. That's so interesting. Um, we have just a couple minutes left. Um, so you spoke a little bit about this in the beginning and there are now a lot more people on the stream, so it just might be helpful to to talk a little bit more about this, which is um, you know, we're we're teaching this class called Code for Absolute Beginners. Um, we teach it live. It is cohorted. We don't we don't do an asynchronous, you know, video version. Um, you know, why take a course like this? Why why spend a thousand dollars um, when you could go to YouTube or you could I'm sure find an asynchronous version to to take yourself through on your own time. Like what's your perspective on that? Yeah, it's funny actually, and I really like how there's so much alignment with the team. I saw um, someone on the team saying uh, "Hey, uh, I think the customer service kind of came in and said, 'Hey, someone just wants to buy the course and they're not going to be able to attend cuz they're in Europe and it's going to be really late.' Um, and they they're willing to spend a thousand dollars, but like they just want the recordings." And we're like, "I don't think they should buy the course." Um, and and it was pretty unanimous actually through the team, which I I I really loved because I felt the same way about it. Um, I I I think that um the the value of the course is really um the energy and the and the motivation you get from being there on the day and carving out that time to do it together and I like the social aspect of you're in the chat, you're seeing what other people are saying live. Um, I I found that you you really can't replicate that just by watching a recording. And I've created a lot of recorded content, right? Like uh you know, my my Udemy course is 24 hours at this point um if you wanted to watch it end to end. \>\> \[laughter\] \>\> Um, you know, and I think there's a you real place for that. But I I'd say recorded content works when you're already very bought in to to that thing and you're like, "Hey, I I I I know for sure that I need to create a LangChain agent, right?" Because my boss has told me I need to learn LangChain and we're that's what we're using. And and and so like it's well worth say doing my Udemy course and then and then like watching the LangChain stuff. And you pay a lot of attention to that because your job depends on it, right? Um, whereas uh I would say that like a course like this like a live course is much better at "I'm excited for this. I'm bought into this. I've identity-wise I feel like this sort of person that should be using code code more." Or like I I really want to be like I I've seen other people do it and I I'm amazed by it and I want to be part of that. But like I just haven't been able to get over the hump. Mhm. And I don't think you can get over the hump very easily on your own. Um, that's what like being there on the day is for. Great. Um, well, we are at time. Thank you everyone for attending. We are launching our next cohort of Cloud Code for Absolute Beginners that Mike Taylor will teach. Um, it will take place on April 14th. We are officially launching it tomorrow, but we will share with you in the comments in a moment where you can um, sign up or just learn more about it. Um, we have a a website that uh, explains everything that that we've talked about here and um, we would love to see you there and of course if you have additional questions you can always um, you know, email us at every or find us online. Um, but we uh, we're we're really excited to see what you guys all end up doing with Cloud Code. Is there anything more Mike that you would want to say? Uh, thanks guys for all the great questions. I just seen them all streaming in um, but uh, it's um, it's an exciting time. There's a lot of fun stuff and even if you don't take the course I think just download Cloud and start messing around with it. Great. All right, thank you so much. All right, thanks Kate. Bye.