Real-World AI Adoption with Tracy Lee

Episode Description

Tracy Lee discusses practical AI adoption, changing developer workflows, enterprise constraints, automation, costs, human judgment, and AI-powered creativity.

Episode Summary

Tracy Lee joins Anthony Campolo and Dev Agrawal to examine how AI adoption is reshaping software development, leadership, and everyday work. Drawing on her experience leading This Dot Labs, Lee explains why organizations need shared AI literacy across executives, developers, QA, and operations, even as models, tools, and recommended practices change rapidly. The conversation covers rising inference costs, enterprise restrictions, shifting demand toward senior engineers, and the gradual human adjustment required to move from autocomplete to agents and automated pipelines. Lee shares examples involving forms, conference planning, presentations, CRM tasks, home design, and staffing, while the hosts discuss architecture, technical taste, framework selection, and responsible review. Their central recommendation is to begin using AI now, continually test its limits, automate repeatable workflows, preserve human judgment, and openly share lessons so teams can adapt together.

Speakers

• Anthony Campolo
• Tracy Lee
• Dev Agrawal

Chapters

00:00:04 - Introducing Tracy Lee and This Dot Labs

Anthony welcomes Tracy Lee back after earlier podcast appearances and reflects on how frontend debates have changed since 2022. Tracy introduces herself as CEO of This Dot Labs, a Microsoft MVP, Google Developer Expert, former GitHub Star, and RxJS core team member. She also describes the consultancy’s decade of work with companies such as Stripe, Twilio, Sentry, and Wikimedia.

Tracy announces that This Dot is hiring senior developers and engineering leads, gives contact details, and invites organizations to contact her about AI initiatives. Dev recalls following the company’s educational content and connects the episode to a long discussion they had at Chain React.

00:03:46 - AI Workshops for Whole-Company Alignment

Tracy highlights This Dot’s AI certification workshops, which are designed to align everyone from executives and project managers to QA, DevOps, and software engineers. The program begins with shared concepts and organizational alignment before moving toward technical practices such as specification-driven development, allowing different roles to participate at an appropriate level.

She also describes an executive-focused workshop built around practical, non-coding uses of AI. Leaders are effectively creating personal chiefs of staff that can handle administrative work, including conference registration. The hosts compare AI tools and discuss why Tracy consolidated much of her work into ChatGPT Work rather than continually switching among Cursor, Claude Code, and other interfaces.

00:08:56 - Rapidly Changing Tools and Enterprise Reality

The group compares today’s AI transition with the earlier explosion of JavaScript frameworks and command-line tooling. Tracy explains that workshop material comes from real organizational experience, where recommended practices can become outdated almost as soon as companies learn them. Specification-driven workflows, agent instruction files, and coding harnesses are all changing quickly.

They contrast public enthusiasm for tools such as Claude Code and Cursor with the constrained reality inside large companies. Some organizations are dropping GitHub Copilot, while others permit only a single approved product such as Kiro—or prohibit AI coding entirely. This creates a widening divide between independent developers with broad choice and regulated enterprises with limited access.

00:12:34 - Teaching Durable Skills as Models Evolve

Tracy describes highly regulated environments where developers may have no approved AI assistance, making demonstrations of modern workflows feel irrelevant. Anthony argues that training should emphasize capabilities that transfer across products, while Dev asks how instructors can teach durable practices when models grow smarter and once-popular techniques become unnecessary or counterproductive.

The conversation then turns to cost and model selection. Tracy recounts spending roughly $1,000 during her first week with Cursor because she used an expensive model for nearly every task. Attempts to substitute cheaper models revealed a practical tradeoff: simple changes may not need premium intelligence, but ambiguous research, planning, and implementation often benefit substantially from stronger models.

00:17:47 - The Human Adoption Curve

A viewer’s question about Microsoft Copilot training prompts a discussion of basic AI education. Tracy distinguishes Microsoft Copilot from developer-oriented tools and emphasizes that adoption is a personal learning process. Even if tomorrow’s technology is better, waiting does not eliminate the time required to build intuition, experience disappointment, refine expectations, and understand where AI genuinely helps.

Anthony recommends beginning with structured prompting, examples, stepwise reasoning, and reusable output templates rather than vague requests. Tracy extends this into a maturity model: people move from manual work to conversational assistance, local automations, reusable skills, and eventually agentic applications. Collaboration with developers is essential because nontechnical teams may not recognize which recurring tasks can be automated completely.

00:22:47 - Moving Higher Up the Engineering Stack

Dev explains that the stable trend across years of AI tooling is not any particular model or interface, but the developer’s movement toward higher-level thinking. Work has progressed from writing every line manually to accepting completions, generating functions, and orchestrating changes across entire codebases. Architecture, intent, and review increasingly replace direct implementation as the main human contribution.

Tracy argues that people who postpone adoption will still need to pass through this adjustment later, regardless of how capable models become. Anthony reflects on the major product advances of 2023 through 2025 and suggests that 2026 has focused more on refinement and broad adoption. Meanwhile, interest is returning to local and open models as organizations reconsider cost.

00:27:09 - AI Costs, Staffing, and Global Labor

The speakers challenge earlier assumptions that token costs would become negligible. AI spending can now represent a meaningful percentage of engineering headcount, forcing executives to seek measurable returns. Tracy notes that leaders may initially promise AI will not reduce staffing, only to change course after seeing productivity differences between employees who adopt the tools and those who do not.

They consider how this shift may affect outsourcing and junior work. Tasks once delegated to less experienced or offshore developers can increasingly be assigned to AI, potentially increasing demand for senior engineers who can specify, evaluate, and integrate the results. The group also observes that correcting an AI system is emotionally simpler than rejecting a human colleague’s work.

00:31:16 - Replacing Tasks Without Replacing Judgment

Anthony points out that AI has no ego or attachment to its output, which makes revision and deletion easier. Tracy shares how she responded when a longtime operations teammate departed: rather than immediately hiring a replacement, she challenged herself to reproduce as much of the role as possible with AI. The experiment produced strong results, although some physical and interpersonal work still required people.

The discussion shifts from personal productivity to organizational automation. Becoming a faster developer can create burnout if it merely multiplies active threads, so teams should identify repeatable processes that can run independently. Anthony describes teaching useful prompts first and then moving those prompts into CI/CD, where automated feedback can review every pull request without requiring constant manual execution.

00:35:20 - Organizing an AI-Powered Workday

Asked about her daily routine, Tracy says outside perspectives remain useful even when someone believes their workflow is already optimized. A conversation with Jeff Escalante led her to color-code active ChatGPT Work threads by urgency, duration, and personal category. This small organizational change made a large collection of concurrent projects easier to manage.

She then describes using browser automation for painful government forms and conference registration. Even when security prompts require repeated approval, AI removes much of the cognitive burden. Tracy connects ChatGPT Work to HubSpot, Slack, Google Calendar, Drive, and Gmail, while Anthony explains that his use of Proton services makes integrated context harder to obtain.

00:39:05 - Purchases, Conference Research, and Executive Assistance

The conversation explores expanding browser and messaging access. Tracy recounts authorizing ChatGPT to complete a $299 conference purchase through an existing payment connection, an experience that made autonomous transactions feel concrete. She also uses AI to identify relevant conferences, speakers, attendees, public LinkedIn posts, and potentially valuable meetings before an event.

Beyond logistics, Tracy treats AI as a business coach, CRM, and preparation partner for difficult conversations. It can gather prior emails, organize contacts, summarize obligations, and create board presentations from existing templates. Tasks that once required hours of mental setup can be reduced to minutes, especially when the system has access to the user’s calendars, messages, and stored documents.

00:43:16 - Design, Images, and AI-Generated Music

Tracy discusses using image generation while remodeling a house without a professional designer. Generated visuals helped her communicate cabinet and lighting ideas to a general contractor, although she still needed to watch for hallucinations and acknowledges places where expert help would have been worthwhile. She also compares the changing image quality of ChatGPT and Gemini over the project’s lifespan.

The discussion moves from visual design to songwriting. Anthony, a trained musician, recommends Suno for generating complete songs from prompts or transforming existing lyrics into music. Tracy describes discovering that a song she enjoyed was AI-generated and becomes excited about using the technology to express musical ideas without first mastering every part of traditional production.

00:47:16 - Reverse Centaurs, Robotics, and Physical Work

The hosts unpack the metaphor of a reverse centaur: rather than combining human intelligence with mechanical strength, automation may leave people performing physical legwork while machines handle intellectual tasks. Tracy connects the idea to practical limitations, noting that AI cannot yet pack boxes for an offsite, so she hired a nearby sixth grader to assist.

They wish robotics were advancing as quickly as language models and mention Hugging Face’s small robot project. Current demonstrations often combine impressive movement with spectacular failure, suggesting that embodied automation remains immature. The conversation also considers humanoid companions and whether conversational systems could reduce loneliness, while acknowledging that vulnerable people are already targeted by relationship and cryptocurrency scams.

00:51:29 - Job Anxiety and the Value of Human Taste

Dev asks why companies should retain people if AI can perform so many kinds of work. Tracy answers that beliefs about AI are often reinforced by whichever stories people choose to share, including reports of companies rehiring after unsuccessful automation. Her practical conclusion is that refusing to adopt AI creates greater career risk than learning to use it effectively.

They address anxiety among newer engineers and identify large technical decisions as an enduring human responsibility. The idea of taste becomes central: experienced people know which architecture, database, framework, presentation structure, or tradeoff suits a particular situation. Anthony connects technical taste to artistic judgment, while Tracy notes that humans must recognize when AI output is technically acceptable but poorly prioritized.

00:56:29 - Senior Expertise, Reviews, and Stack Selection

Tracy explains why senior developers remain valuable: experience creates an instinct for what can be delegated, what must be checked, and how a merely adequate result can be improved. She compares reviewing AI-generated marketing plans with inspecting generated websites, where a human notices duplicated data, inconsistent references, or changes that fail to propagate across pages.

This raises questions about the future of code review. Teams may need automated systems that distinguish harmless text edits from consequential logic changes so reviewers focus their attention appropriately. Anthony then asks whether AI should choose a project’s stack. Tracy argues that maintainability and team capability still matter, even if future systems may eventually need fewer frameworks or less human-readable structure.

01:00:34 - From Basic AI Use to Deeper Automation

The group discusses how improved models can replace standalone automation products. Tracy once relied on Zapier for conference workflows but can now manage some of those processes directly through persistent ChatGPT Work threads. This reinforces her broader message that sustained experimentation reveals possibilities users cannot anticipate at the beginning.

She distinguishes casual or intermediate AI use from deeper adoption. A developer may generate tests, code, or CI tasks yet remain unaware of how much further orchestration can go. Anthony suggests demonstrating ambitious examples, such as developers rebuilding complex infrastructure with agents. Both agree that tooling knowledge matters, but the larger obstacle is often a mental ceiling created by limited expectations.

01:04:04 - Credit Limits and the Rise of Shadow AI

Tracy emphasizes that broad industry transformation will take years because many people still use AI minimally or not at all. She then identifies a growing organizational problem: companies encourage engineers to rebuild their workflows around AI but later impose strict monthly credit or spending limits. Removing a tool after it becomes essential can sharply reduce productivity and morale.

Anthony says developers may respond by purchasing personal subscriptions across multiple providers, creating more unapproved shadow AI inside companies. Dev describes a different coping strategy: during rate limits, he writes detailed specifications, studies the codebase, and prepares reviews so agents can work efficiently when access returns. Completely removing the tools, however, would make returning to fully manual coding frustrating.

01:08:01 - Unfiltered Opinions and the Hallway Track

The conversation turns playful as Tracy encourages outspoken community members to appear on the podcast and share the candid opinions normally heard after conferences. She recalls the frank Chain React hotel-lobby debate about Claude Code’s harness and argues that such unfiltered technical judgments are often more useful than carefully softened public commentary.

Anthony and Dev affirm that the show aims to preserve honest disagreement rather than water it down. Tracy proposes recording informal conference conversations or creating an event centered on the hallway track. The group discusses practical challenges such as noisy venues, consent, sponsorship, and transcription, while noting that relaxed discussions outside formal talks often reveal the industry’s most consequential ideas.

01:12:02 - RxJS, Web Standards, and AI-Aware Development

Dev asks how established libraries such as RxJS fit into AI-assisted programming. Tracy says human expertise can introduce tools that an agent would not select independently, making libraries another expression of technical taste. While attention has shifted away from release-by-release framework news toward AI, standards groups such as TC39 and contributors such as Ben Lesh continue advancing the web.

The subject of standards leads to a broader reflection on how frontend priorities have changed since React’s rise and the 2020 wave of meta-frameworks. Anthony describes giving Bun release notes to an agent, asking it to inspect his repository, identify relevant changes, and produce an upgrade plan. AI can therefore make ecosystem updates more directly actionable rather than making releases irrelevant.

01:16:24 - Changing Attitudes Toward AI Coding

The hosts joke about rare appearances by community members and locate an older stream about local language models. This prompts Tracy to reflect on how dramatically attitudes have shifted. Developers who once defended artisanal hand-coding or publicly resisted AI have gradually incorporated it into their work, even when they remain uncomfortable with the culture and marketing surrounding AI companies.

Anthony distinguishes opposition to the technology from dislike of the industry’s tone, incentives, or public persona. The discussion illustrates the episode’s larger adoption theme: people rarely move from rejection to full acceptance at once. Instead, they experiment privately, revise their positions, and adopt selected capabilities while continuing to criticize the institutions producing them.

01:20:01 - Closing Advice and Total Duration

As Tracy prepares to leave for family responsibilities, she closes by inviting executives who feel uncertain about AI to contact her or attend This Dot’s executive workshop. She believes that showing real workflows in a safe setting can make the technology less intimidating, particularly for leaders who are removed from day-to-day development and may be handling sensitive information.

Her final recommendation is to share experiments openly so organizations and the wider industry can adapt together. The conversation ends with lighter remarks about bourbon, future guests, upcoming schedules, and Tracy’s children. The episode concludes at 01:23:14, marking a total transcript duration of 01:23:14.

Resources and Links

• thisdot.co
• workshops.thisdot.co
• thisdot.co/ai-executive-workshop

Transcript

00:00:04 - Anthony Campolo

Hello everybody, welcome to AJC and the Web Devs. We got new guest Tracy Lee who was on FS Jam back in the day, and I also was on her Modern Web—fantastic podcast—as well. Uh, really great to have you on, Tracy. How you doing today?

00:00:23 - Tracy Lee

Good, thanks for having me. It's been so long, life just Passes by.

00:00:28 - Anthony Campolo

I know, right? I was looking it up. I think it was 4 years ago. It was back in 2022. And I remember you were asking me, is Solid or Qwik gonna overtake React? 4 years ago.

00:00:38 - Tracy Lee

Oh my gosh.

00:00:39 - Anthony Campolo

Still talking about Solid versus React. Qwik, maybe not so much. Sorry, Misha.

00:00:44 - Tracy Lee

Yes, I know. Times have changed. Times have changed.

00:00:48 - Anthony Campolo

Yeah. But, um, give a short little introduction to who you are, what you do. And also, dev note for you, I wanna start doing the plugs up top, not at the end of the episode. So also, if you wanna plug plug things, do it right now since, um, this is what most people will see, not the end of the episode. So go ahead.

00:01:04 - Dev Agrawal

Yeah, totally makes sense.

00:01:07 - Tracy Lee

Well, my name is Tracy. I'm the CEO of a company called This Dot Labs. I'm also a Microsoft MVP, a Google Developer Expert. I've been a GitHub Star, RxJS core team member, and, um, who knows what's next. Uh, This Dot. We are a software consultancy, so we've been around for 10 years now. Started in the Silicon Valley, and we get to work on really cool, ambitious projects. So some of the customers you might recognize are people like Stripe, DocuSign, Twilio, Cloudinary, Sentry, Roblox, Wikimedia, etc. So we are hiring too. So yeah, yeah, we're hiring. We're hiring senior developers and senior engineering leads. And so if you're interested in applying, you can email me directly, tracy@this.co or jobs@this.co. It's confusing. So it's t h i s d o t dot co. Should have thought of that, right? Like when I made the made the name.

00:02:08 - Anthony Campolo

Like this, right?

00:02:09 - Tracy Lee

Yes, Tracy at this dot co. Yeah, yeah. Or if you need help with anything related to AI, always happy to have a chat about it.

00:02:19 - Dev Agrawal

Fantastic. Yeah, I so I um I've kind of been like watching this dot Uh, like the media that— or the content that comes out of this .dot YouTube channel for a very long time now. And so I'm a huge fan and I'm very, very excited to have you here and get to ask you a bunch of questions. Yeah, I think this kind of started with us, uh, hanging out in the hotel lobby at ChainReact.

00:02:46 - Tracy Lee

Yeah, for 4 hours.

00:02:47 - Dev Agrawal

And Chris Griffin. Yeah, that was an incredibly long conversation. Um, but yeah, we, uh, uh, I'm, I'm also excited to get, uh, Brandon on here soon. But, um, yeah, well, um, we would love to hear about, uh, kind of like how the, uh, how AI has been changing like kind of the software landscape and, uh, how you have been dealing with it as a consultant.

00:03:11 - Tracy Lee

Yeah, totally. Anthony, do you need to do plugs first? Because we can do that.

00:03:17 - Dev Agrawal

Yeah, we can do plugs.

00:03:19 - Anthony Campolo

Yeah, you should, you should do your plugs first on any links or things that you want people to know about. Go ahead and hit those up and then we'll dive into the questions.

00:03:26 - Tracy Lee

What, was I supposed to do a plug or are you guys doing—

00:03:30 - Dev Agrawal

No, everyone.

00:03:31 - Anthony Campolo

I mean, I mean the plugs for the guests. That's what I meant.

00:03:33 - Tracy Lee

Oh, oh, it's for the guests. I didn't even know that. Oh, okay.

00:03:37 - Dev Agrawal

Gosh, that was not clear. We get to plug every time. We get to plug every time.

00:03:42 - Anthony Campolo

It's, it's your floor. The show itself is the plug. We have nothing to plug.

00:03:46 - Tracy Lee

Oh my gosh. Okay. Let me plug, you can actually get AI certified through This Dot. So if you go to workshops.thisdot.co or just go to thisdot.co, you can see some of the workshops we have. The thing I like about the workshops is we actually, it's actually broken into 3 parts. And so it's this idea that executives everywhere from like the executive level down to the being a project manager to the QA to the, to the whomever, to DevOps. Everybody needs to be on the same page when it comes to AI. Hello. Um, everybody needs to be on the same page when it comes to AI. So this is like a, it's a 1.5-day workshop, but it's broken out into 3 parts. And so we believe that like the first 2 hours, the first part of the workshop should be about everybody being in alignment. How do you get your entire team aligned? How to you know, people who maybe aren't in the day-to-day sort of know what's going on? Um, and then, you know, they can drop off and maybe the, um, well, that's actually a different one. Um, but maybe some people can— it's actually workshops.this.co. Um, but, um, how do people like QA folks who maybe aren't like deep into development every day still get plugged in and still sort of get the value out of AI, right? Um, and then the third part is, you know, talking about spec-driven development and, you know, sort of like the newer standards on, um, how to use AI as, um, as developers. So I would plug that. And it's so funny because, Anthony, you, um, showed the executive workshop that I host. So if you're just curious how, you know, me non-development-wise, you know, uses AI in my day-to-day—

00:05:25 - Anthony Campolo

who's not tactical, that's what that is, is what you're saying.

00:05:27 - Tracy Lee

Yeah, I mean, you can be technical too, right? But this whole idea that I feel like every single person I talk to who's sort of in leadership is building their own chief of staff when it comes to AI. So I know we talk about how developers can get up to speed with AI very often since we are in that world. But, you know, it's crazy to see how you can use AI to do everything else. So I never— like every conference that emails me that I need to register, I don't even register anymore. I just have AI do it for me.

00:05:58 - Anthony Campolo

Oh my God, yes, that is the dream. Uh, are you doing that with what, uh, ChatGPT Work?

00:06:04 - Tracy Lee

Yes.

00:06:05 - Dev Agrawal

Yes.

00:06:06 - Tracy Lee

Okay, I'm fully ChatGPT Work. Um, you know, I was using Claude Code, but I feel like the, the sort of like interface, if you will, is so much better. I found myself, um, I don't know, I found myself just going back between like Cursor and Claude Code and ChatGPT Work and I feel like ChatGPT generally is sort of this more conversational vibe. Maybe it's because, you know, a lot of people started there, right, to have conversations, to brainstorm, etc. Um, and then I was using— I was basically using that and Cursor. And, um, then, you know, like, duh, why aren't I just using, you know, ChatGPT Work for both of them? Um, so now I'm using it for all of them. So, you know, anybody who has ADHD or ADD or anything like that, it's so satisfying. I mean, I'm not diagnosed, okay, but like I really enjoy doing everything. And so having like 10 different threads of things you're working on is like a dream to me.

00:07:05 - Anthony Campolo

No, it's true. I've got— I mean, I usually am running minimum 3 agents at one time, just pop back and forth between like one.

00:07:13 - Tracy Lee

It's the best.

00:07:14 - Anthony Campolo

I hop back over, do the thing. But, uh, anyway, for all these links, these will all be in the show notes and the YouTube description. I'm adding them as we go, but, um, take it away, Dev, with, uh, the questions you wanted to start diving into.

00:07:29 - Dev Agrawal

Yeah, I mean, uh, so who, uh, leads the workshops? Like you said, you do the executive workshop.

00:07:36 - Tracy Lee

Yeah.

00:07:36 - Dev Agrawal

And what about the other one?

00:07:38 - Tracy Lee

The other workshop is, um, typically our CTO, um, Elliot Fouts, who runs it. So, you know, not— and he's like, he still hands on keyboard. But, you know, not very often, right? Like, a lot of times you go to these workshops and you're like, uh, who's teaching me? I remember my, my first coding camp, my boot camp, this dude was teaching me JavaScript and he's like, I don't really like HTML or CSS. And I'm like, why are you teaching it to me then right now? You know? Um, but, you know, to have somebody like that of that caliber, like, actually teaching a workshop, I think is incredibly valuable. Um, so yeah, it's a good one. Um, and then the, the executive workshop that I host is more, you know, again, I mean, I, you know, 10 years ago when I started speaking at conferences, I start— I, I, I spent a lot of time live coding, right? And, you know, that was my whole thing, like, let me just live code on stage. And this was the land of, you know, oh, did you know that Ember CLI exists? Angular CLI? Let me show you how awesome it is. So I'm like, go live code like 3 different apps One in React, one in Ember, one in Angular. And then obviously, like, you know, as Vue came up, I came about, right? Like, I sort of started doing that. So I feel like this is kind of like the same thing but over again, but just with AI, right?

00:08:56 - Anthony Campolo

With that, that was like, right, FS Jam, which makes sense of all these frameworks.

00:09:00 - Dev Agrawal

And I saw, right?

00:09:01 - Tracy Lee

Yes, I love that. It's like history repeating itself 10 years later.

00:09:07 - Dev Agrawal

Yeah, in the— in just new forms.

00:09:08 - Tracy Lee

Yeah, yeah.

00:09:10 - Dev Agrawal

After AI agent, I hope that there'll probably be another new thing to do. Um, but I'm guessing that the, the material in these— sorry, I'm guessing that the material in these workshops kind of comes from like some sort of real-world experience that you might have with companies. So that's what I would love to hear more about. Like, I mean, certainly if, if people in Silicon Valley are having trouble with AI or like are still trying to figure out how exactly to make use of AI, I can't imagine that the enterprises are having like any better luck.

00:09:44 - Tracy Lee

It is wild because, you know, you see on— you see online, right? And, you know, um, you know, Dev, I, I really enjoyed our conversation in the lobby at ChainReact because I was like, hold on, you know, the, the industry has just rallied around spec-driven development and that this is the way it should be. And like, Dev is having this conversation with me about how, you know, it's changing and this and that, and here's the standards. And I'm thinking, people are just learning the best practices that now you are saying are anti-patterns. And it's the same thing with, um, you know, Claude Code, right? Like, you know, Dev and, you know, Chris and Brandon and I were having a pretty lively conversation. And, um, you know, there was a lot of a—

00:10:28 - Anthony Campolo

when, how long ago was this? Conversation.

00:10:31 - Tracy Lee

This was like a month, month ago, right? Like just about a month ago.

00:10:33 - Anthony Campolo

Just make, just making sure.

00:10:34 - Tracy Lee

Yeah.

00:10:35 - Dev Agrawal

I like, yeah, it was end of July in Portland.

00:10:38 - Tracy Lee

Yeah. Yeah. But then, you know, I was like, hold on, you guys are telling me that Claude Code is like the worst one to use, but like everybody just—

00:10:48 - Anthony Campolo

it's the worst with the smartest model. Is it? That's the thing that makes it so fresh is Fable is the smartest model, but Claude Code is the dumbest harness.

00:10:55 - Tracy Lee

Yeah. But, but let me tell you, everybody, you know, sort of like, right, everybody this year has, you know, 8 months ago sort of has rallied around like, oh, we all use Claude Code.

00:11:07 - Anthony Campolo

Because, you know, a year ago, that was the only agent thing they'd heard of was Claude Code, right?

00:11:12 - Tracy Lee

Well, Codex, a few years ago, maybe like 2 years ago, or maybe, gosh, even a year ago. I mean, time is just moving so fast, right? A year, year and a half ago, everybody was kind of saying, Everybody's actually using Cursor. You know, it's like Cursor or Claude Code.

00:11:27 - Anthony Campolo

Or they were still stuck on Copilot.

00:11:30 - Tracy Lee

Yes. And then now it's like, no, actually everybody is on— a lot, a lot of people are on Claude Code. And then, gosh, I never, ever, ever thought that Copilot licenses would be going away. But I'm hearing in companies that they're getting rid of Microsoft Copilot. And I'm like, oh, GitHub Copilot, I'm sorry. And I'm like, what?

00:11:53 - Anthony Campolo

It's like a credit thing now.

00:11:55 - Tracy Lee

Yeah, but just like, how is that even possible? Like, I would have never thought that that would be the reality. And they're mostly saying we're keeping our Claude Code licenses, but like we're getting rid of our GitHub Copilot licenses. So now this is, this is like not everybody, but a lot of people I talk to, this is sort of the pattern that I'm seeing.

00:12:15 - Anthony Campolo

Yeah, that makes sense to me. That's good jobs as well for, especially at the, at ADP, cuz it was, we were mandated to teach either Copilot or get this Amazon Q. Okay. Which now does not even exist anymore. That has now turned into Kiro, which also no one uses.

00:12:33 - Tracy Lee

Mm-hmm.

00:12:34 - Dev Agrawal

But, um, so except for literally everyone at AWS. Yeah.

00:12:37 - Tracy Lee

But also Anthony, what's so funny is there are some people, some companies, some enterprises I've talked to who only have access to Kiro. No, so terrible. Yeah, but, but that's the crazy thing, right? Like, we on the internet, you know, think these things. We have choice, right?

00:12:55 - Anthony Campolo

And then choice.

00:12:56 - Tracy Lee

Yeah, right. And then you go into a lot of these companies and you're like, oh my God, like, you know, something— you know, I, I work in a lot of highly regulated environments and it's like they don't even have access to GitHub Copilot, you know? I don't have anything.

00:13:10 - Anthony Campolo

Um, no AI coding at all.

00:13:12 - Tracy Lee

Yeah, no, and, and, you know, it comes to the question, right, like us as software developers, uh, if you are a software developer, what does that mean? Should you be learning things that a lot of these sort of enterprises are stuck on, or should you be leading the charge and saying like, hey, by the way guys, you know, do you see, you know— and so with this executive workshop, right, I'm just like, you know what, I'm just gonna use ChatGPT Work and see how it goes, but There are gonna be so many people that attend that and say like, look, this is completely irrelevant to me because like, I, I can't, like, I don't have access to it. So what do you do? Like, we're, we're kind of split as an industry, if you will.

00:13:50 - Anthony Campolo

Yeah. You wanna, as much as you can, demonstrate the agnostic abilities that like any agent could do. That's, that's kind of how we approach it with our workshop.

00:13:59 - Tracy Lee

Yeah.

00:14:01 - Dev Agrawal

Yeah. But how do you best demonstrate those capabilities when the model keeps changing when the model keeps getting smarter, workflows keep getting obsolete.

00:14:09 - Tracy Lee

Yeah.

00:14:10 - Dev Agrawal

Like there were a lot of, like there were huge kind of, I don't know, workshops or just courses on how to write AGENTS.md file, how to write, like how to, what skills you should be using. And then Claude, like what Fable, Opus, and Sol come out and all the providers are like, oh, you should not be writing these huge AGENTS.md and skill files. You should just get rid of everything and just let the model do its thing. So things keep changing all the time. How do you keep up with that? How do companies keep up with that?

00:14:40 - Tracy Lee

Well, I think that's really hard because, you know, as, as we lived through the token maxing days, we are in the token minting days, and, um, there is a constant conversation to be had about cost, right? And so, um, I know there's probably like a lot of startups— I want— not even probably, I know there's a lot of startups who are trying to solve this problem. But, you know, beforehand, right, it wasn't like Gosh, I remember, was it like a year, a year and a half ago or so where people like, oh, I use this model for this and this model for this and this optimizes this and this is whatever. Um, but, um, and then, and then it was like, no, actually don't even worry about that. Just, just go, right? We're in the token maxing days. And now it's like, hey, you know, best practices on which model to use when. So, you know, I'll, I'll give you guys a story. So like when I, the first week I used Cursor, I spent $1,000 and my CTO came to me and he was like, you know, you just spent $1,000, right? And I said, well, it doesn't sound so bad to me. You know, I got a lot done and you know, I was using the highest model and he is like, look, for text changes, you probably should have just been using like a smaller model. And I was like, what? What'd I say? A few hundred dollars? He's like, probably could have been like tens of dollars versus like $1,000. Um, so then I got on ChatGPT Work though, and, uh, I made the same mistake, right? So I spent like about $1,000 in, in a week. And, um, but here's the thing, so I was using Sol, and then, you know, I, I was like, okay, okay, maybe I don't need to use Sol, like let me use Terra or, or, or, or Luna or something like that. It's just dumber. It does not get the job done for, for just sort of like very, like, I don't know, I don't know if the right word obscure is, but for the sort of like, can you find this thing for me? Can you do this thing? Like, it's not, you know, Sol is the best model for that.

00:16:44 - Anthony Campolo

Even when you like write a plan and you have it do the plan step by step, as it's doing the plan, it's gonna run into shit. Like, Yeah, it's gonna find edge cases as it's implementing the plan. So you still want the smart models even implementing, because that's usually the pattern people say is have a smart one write a plan, have a dumb one implement the plan. But that doesn't really work in my experience.

00:17:04 - Tracy Lee

Yeah, I think, um, I mean, I think in development, you know, I mean, here I am just, you know, submitting PRs left and right and saying like, guys, make sure to check the logic before you guys Press merge, please. Um, but you know, when it comes to text changes, obviously, like, you know, who, who cares, right? Um, but I feel like when it comes to text changes, sometimes I'm sitting there, I'm like, oh, you know, like, this doesn't sound right, make it sound— I literally say make it sound better, like, of a, of a, you know, like, great directions, Tracy. But you know, Sol can typically do better than that, right? Than Terra or Luna who are going to be like, uh, you know, and go back and forth.

00:17:47 - Anthony Campolo

Yeah, we got someone at LinkedIn who's saying— oh yes, yes, we like, we like dumb it down real quick.

00:17:52 - Tracy Lee

My company wants my L&D team to develop training on MS Copilot and I'm lost as to AI basics and use cases. Can you help? I mean, I'm always asking. Yeah, and I'm always happy to have a private conversation with folks. So feel free to message me anytime. Microsoft Copilot is a little bit different, right, than GitHub Copilot, if you will.

00:18:16 - Dev Agrawal

Yeah, it's closer to ChatGPT than to Claude Code.

00:18:20 - Tracy Lee

Right, right.

00:18:22 - Dev Agrawal

ChatGPT Work.

00:18:23 - Tracy Lee

Yeah, yeah. But training, gosh, I think it's just, you know what, you know what I've actually found so crazy is We talk about the adoption of AI and you think, right, like, okay, well, how long is it going to take for somebody to adopt AI? But I don't know if you guys have seen in your own personal journeys, but like, as a human, you just have to go through a path and like, you don't know what that path is. Everybody's path is pretty much the same in terms of like the sort of like, you know, trough of disillusionment, blah, blah, blah, whatever, right? But it's like, you know, a lot of— some people have even said to me like, well, why should I adopt a technology now when I know it's going to get better tomorrow? But like, you can't—

00:19:11 - Anthony Campolo

it's already better than what you're doing, right?

00:19:14 - Tracy Lee

But then also you underestimate the amount of human— like, the time it takes. The time it takes for a human to go through that adoption curve is not going to change, right? Like, it might get like marginally better, but they still have to go through that adoption period. And I think that's the hardest part to, um, like, you can't, you can't rush that, right? You can't just like shove, shove tooling down somebody's throat and say like, hey, you know, everything's gonna get better.

00:19:42 - Anthony Campolo

Yeah, because there's a period where you have to learn new concepts, apply those concepts, see how they play out in real time, refine them from there. So what I would say to this question is that you want to start If you wanna start at the very basics, like start with just prompting. Like you have like chain of thought and few-shot prompting. Those are like the 2 most important concepts. I think like few-shot prompting is when you give an example of what you want the output to look like. And then chain of thought is when you tell it to like think through step by step, like look through these tests. What are the inputs? What are the outputs? How are the inputs processed before they get to the outputs? That's like a very simple chain of thought, 3 steps, you know? So if you can get your engineers to understand that, because then you can get at least consistent outputs. And that's what's most important because people will just prompt saying like, fix my docs. It's like, it's going to be different every time. Everyone is going to figure out how to do that. Like you, you want to have ways of actually giving it structure. So like generating reports with like a report template, like. That's something I do a lot. Yeah. And then you can at least feel confident like you're gonna get something consistent each time and then you can iterate on that and you can refine it.

00:20:55 - Tracy Lee

Yeah. I, I think that one of the, one of the best parts about working with engineers though, right? So like a lot of, a lot of departments are adopting AI, sales, marketing, et cetera, et cetera. But, you know, I have the privilege of having a technical background. I have the privilege of having, a technical team that I can just talk to and, you know, right? Um, but you don't know what you don't know. And so I think like, you know, for me, for example, lately I've still been kind of like going through and saying, okay, you kind of have to— it's almost like you're doing it manually now, okay? And then now you're doing it with something like just ChatGPT or, um, you know, Microsoft Copilot or Claude Code or whatever, right? So you have your your sort of things that you have in your— on your own machine, right? Processes or automations that you might create. Then there's like another step of, you know, creating, um, more automation, like whether that be like a skill or like just something a little bit deeper than, hey, remind me when XYZ. And then you have like the other step, which is actually building agentic applications or agents. Right, um, to fully automate something away. And I think that without like talking to a developer about it and talking through your workflow, you might not actually know what is actually possible, right? Until you actually talk to a developer, they examine what you're doing, they're like, dude, like, I know this is easier for you, but like, I can actually make this so that you don't ever have to touch this. You know, so I think there's a lot of value in doing that, um, and I hope teams are sort of reaching out to their developers and trying to figure out those processes.

00:22:47 - Dev Agrawal

Yeah, it makes sense. And I think to the earlier point of like, why should I use AI today when it's going to get better tomorrow? I'll just use it then. I think, yes, model capabilities keep changing, uh, they keep getting better. Some best practices kind of like disappear over time, but there's also certain things that don't change over time. If I think back to when I started using the Copilot autocomplete in VS Code 3, 4 years ago and today to orchestrating multiple agents, the thing that has remained the same is that my work keeps moving a little higher up the stack, going from directly writing all the code to writing most of the code and getting some of it autocompleted, to getting a lot of it autocompleted, like entire functions, to getting entire code bases autocompleted. The thing that has remained the same is that I am not thinking of all the small technical details, but I'm thinking a lot more of the high-level details. That's something that's not going to change. It's something that's only going to get even up, like it's going to increase. I'm going to move even up, even higher in the stack as the models get better. So that is something that we need to get a feel of as like immediately.

00:24:03 - Tracy Lee

Yeah, you just need to wait for that. You just get started, right? Because Dev, you couldn't, you couldn't start, you know, if, if we're talking like 2 years later and it's the highest models, you are still going to have to go through that process of the autocomplete to this, to the this, to this, to this. So really, I think people are who are like pausing on AI for whatever reason are really underestimating that sort of like human transformation that needs to happen.

00:24:32 - Dev Agrawal

Yeah, I think the people who can maybe just, uh, skip the, like, a few weeks of AI, like take a 3-week vacation, are the people who already know where the AI capabilities are. They know that, okay, if I take a 3-week break, I know, like, okay, models are gonna, gonna get better, but once I come back, I'll catch up in, in a couple days and I'll And I'll be all good.

00:24:54 - Anthony Campolo

Yeah, I mean, I honestly feel like the main, like, innovations happened in 2024, 2025. Like, I feel like people throughout 2026 have been catching up on stuff, but I don't feel like things moved as fast this year as it did last year. That's just my personal opinion.

00:25:10 - Tracy Lee

Hmm.

00:25:11 - Anthony Campolo

Because Claude Code came out last year. Like, these, all these CLI tools came out last year. They ain't come out this year. They've been refined over the course of this year, but really the main innovations happened in 2025 and we've been living with the consequences ever since. And the, the big innovations in 2025 were built on all the innovations that happened in 2023 because ChatGPT itself came out November 2022. Let's not forget. So 2023 is when everything important actually happened.

00:25:37 - Tracy Lee

Yeah. Yeah. It's, it's crazy to see how, like how fast things move, but then all how, like how slow things move also at the same time. Um, and it's also crazy to see, like, just again, you know, in the past year, year and a half or so, how many things have changed, right? Like, everybody was saying, no, we're only going to use the big, uh, big models. He started using GitHub Copilot maybe like 5 months— yeah. Um, so, you know, if you— so before, in the beginning, it was like, oh, look at all these other models. Like, look at this thing I can do on my local machine. Look at this. Like, I'm using these different models to optimize. Right. But then it was like, no, everybody just used like Anthropic or OpenAI. I mean, really Anthropic and OpenAI, right? But now, now that cost is becoming a thing. Oh my gosh.

00:26:26 - Anthony Campolo

Now we're thinking about local, uh, local LLMs on our machine again, but like hosting open models on cheaper inference, like GLM and KIMI and What's the other one? DeepSeek.

00:26:43 - Dev Agrawal

Gwen.

00:26:45 - Anthony Campolo

Gwen.

00:26:45 - Tracy Lee

Yeah.

00:26:46 - Anthony Campolo

Yeah.

00:26:46 - Tracy Lee

Yeah. But like, we got it. I mean, I thought that conversation was over about 8 months ago, maybe. Right. I thought we were never going to have that conversation. And I thought that we were going to have a conversation about it's going to get cheaper and tokens are basically going to cost nothing. Do you guys remember that conversation?

00:27:04 - Anthony Campolo

Like 0xAlpha, which was free for a week.

00:27:06 - Dev Agrawal

Yeah.

00:27:06 - Tracy Lee

Yeah.

00:27:08 - Dev Agrawal

Yeah.

00:27:09 - Tracy Lee

But now here we are talking about how it's just gonna get more expensive. I mean, you look at sort of the executives that I work with and, you know, if all of a sudden, like, the cost of AI for a development team is somewhere between 10 to 30% of their overall headcount, right? Like, it's a real thing, right? I mean, there are changes that need to be made and I think a lot of people are still saying, you know, especially in like companies that I feel like are just adopting AI, the way they get people to be less scared about AI is to say, we're not gonna fire people. We're just, it's just gonna make it better. And then people are like, okay, I'm not scared. I'm not scared. You know, and then they adopt, but then guess what? Those people that didn't adopt while everybody else adopted get fired. Right, because, you know, AI might not take your job.

00:28:05 - Anthony Campolo

Then yeah, there has to be, uh, there has to be something. That's right.

00:28:08 - Tracy Lee

Yeah, or ROI shown or something like that. Um, but, but, but it's also, you know, I feel like when executives say that, they're also walking the same path like Dev talked about. You know, as a developer, you walk that path of like autocomplete to this, to this, to this. An executive is like actually truly believes in their heart, we're not gonna fire people. This is just gonna make us better, just gonna make us faster. But then they get to the next path and then the next path, and then it starts becoming the conversation of like, okay, actually we need to let people go. Or like, this person isn't adopting. My favorite quote is like that whole thing where they say, I mean, in every industry you hear this, like AI isn't gonna take your job. But AI, a developer that's using AI is gonna take your job.

00:28:56 - Anthony Campolo

I've said that. Yeah.

00:28:57 - Tracy Lee

Mm-hmm. They say that for marketing. They say that in sales. Yeah. You know, they say that everywhere.

00:29:03 - Anthony Campolo

I mean, it is kind of, it is kind of true. Yeah. The way I would do it if I was an executive, you start by firing a bunch of people. Then in a month you're like, hey guys, AI. So you've already taken care of it. You just fire people cuz they're incompetent. And then you add AI.

00:29:22 - Dev Agrawal

Yeah. Yeah.

00:29:24 - Tracy Lee

It is. I'm, I'm, I'm, I'm curious about sort of different overseas markets though, and what that's going to mean. Right. And, you know, again, you know, we can, we can sort of think about it here, but we see how fast the industry's changing and, you know, how, how fast things change just in 6 months, um, even a month. Um, but, you know, offshoring, is that gonna be, you know, before it was like, hey, we're just gonna give it to a bunch of offshore developers and it's gonna be better. So we don't need our senior people. People. And right now it's actually we need our senior people and we don't need the offshore developers because you can basically tell AI to do things that— I'm generalizing here, okay? But you can do things, you can, you can, you can give it— if you can give AI things that a low-level developer, you might give a lower-level developer and it will do it better. So that's the assumption, right? Um, but I mean, who knows? Who knows what's— what the conversation is going to be in the next 6 months about that.

00:30:24 - Dev Agrawal

Yeah, and the best part is that when the AI QA comes back with a, with a bunch of issues, you don't feel bad just telling it to go, go to hell, which you do when it's a human on the other side.

00:30:36 - Tracy Lee

Let me tell you, so we hired somebody to do our websites, and, you know, he was— it was me and him vibe coding. It was mostly him vibe coding, but it was me like sort of whatever. And, you know, as we were going through the design, you know, even though I really love him as a person, everything like that, There were definitely pieces where I was like, man, this is like way too much animation. This is completely useless. But he's like, man, you know, I spent so much time on this animation. I'm thinking, okay, let's just move it further down the page, you know? And I kept it. Whereas if it was AI, I would've just cut it, you know? Like, I don't need that crap. So unfortunately when he left, I did cut it. But you know, it's the same thing, right?

00:31:16 - Anthony Campolo

Like AI doesn't have an ego. It's, Yeah, that's one of the best things about it. And it doesn't have all— yet, like, this code is obvious. Like, it will treat all code from just a pure, you know, kind of mindset.

00:31:29 - Tracy Lee

Yeah, yeah. And, and, you know, I, I was very sad when one of my teammates recently left me. So Calypso, she's Calypso Coding on Twitter, um, but she was with me for a very long time. She was, she was in an ops role I met her at a meetup actually, and she was trying to get a developer job. And so, you know, she hung out with me for a little bit, as she did, 2 years. I mean, a little bit. She hung out with me for 2 years, worked for me for 2 years. Um, and when she left, I said, you know what, I'm not gonna hire a person back. I'm just gonna like see if AI can do it for me, right? And I have to say that if you really challenge yourself, and I think this is some of the advice that, um, we've given people as they're adopting AI as well. If you're using AI for autocomplete or something like that, push the boundaries. Actually see what you can push AI to do that you do manually every single time you decide to do something manually. That's how you get better at using AI. But that was what I did. I said, you know what, let me just— everything Calypso did, let me just try to figure it out myself. And honestly, I feel like AI is, you know, I love Calypso. Okay. But like, it's doing a better job because AI can work 24/7. You know what I'm saying? And like, it does.

00:32:50 - Anthony Campolo

It never gets tired.

00:32:52 - Tracy Lee

Yeah.

00:32:52 - Anthony Campolo

And it, and it, they don't just have human energy that they're plugged into the matrix, you know?

00:32:56 - Dev Agrawal

And AI, yeah, there's just rate limits. That's, that's the form of tired.

00:32:59 - Anthony Campolo

Yeah.

00:33:00 - Dev Agrawal

There's 5-hour rate limits.

00:33:03 - Tracy Lee

Yeah, so it's great. I mean, there are some things that, um, you know, I, I wish I had a human to do because there's still things that humans need to do. Um, but it gets me so far and, um, it— God, it just makes me like so much more productive. But again, right, like a few, a few months ago we were talking about how like for, for us it's been the conversation like for the workshops, for, you know, you guys know how to use— yeah, like if you can, if you can become a 10x developer now with AI. Amazing. But then, you know, we all hear the conversations about burnout and having to do this and etc., etc., etc. But, you know, then it goes to the, well, okay, you're improving yourself as a 10x developer, but what are the processes that you can improve with, you know, within your team or whatever, or within your pipelines that can actually be more automated? So Talking about the agentic applications, talking about scripts, just talking about things that you can run more that are not as much of like, oh my God, I have like 20 threads in my head and I'm 20x more productive. But there's smarter ways to do things, right? And again, I think that's just the journey.

00:34:13 - Anthony Campolo

No, that's such a good point. And this is a big theme that I tried to draw out in my workshops I was doing where we would basically spend the first week teaching them a whole bunch of prompts. That are like very, very useful. And then the second week we would be like, put these prompts in your CI/CD. Like don't just sit, sit there on your computer running these over and over again.

00:34:32 - Tracy Lee

Exactly.

00:34:33 - Anthony Campolo

Have these happening on like every PR or something like that. And then you get that feedback immediately. You improve the PR, you push it up, and then all of a sudden you have this like great review flow. And it's not just you on your computer, like spending hours and hours a day running all these fricking prompts.

00:34:49 - Tracy Lee

Totally, totally agree. Totally agree.

00:34:52 - Dev Agrawal

So it's fun that we were, we were earlier talking about frameworks and while we were in this call, Remix 3 just released their release candidate. So I'm sure that'll be a topic for a different time. Yeah, I'm sure that'll be a conversation for a different time. But yeah, speaking of pushing the limits of AI, I'd love to hear about how you are pushing the limits of AI, what your workflows look like. What's the role of AI in your daily routine?

00:35:20 - Tracy Lee

Yeah, well, I'll tell you, I really started off going about things and I think this is why it's so important to just talk to people through your workflows, even though you think you're productive. My friend was trying to get me to— sidebar, my friend was trying to get me to like, he was like, maybe you should talk to a therapist. And I was like, I know my problems. I know how to solve my problems. I don't need a therapist. I'm not pushing away therapy, right? But like, I feel like I just know, and I would know what I need help with, right? But it's actually not true in all cases. And it's the same thing with AI. Because, you know, you think you know, you're being incredibly productive. You're like, well, how could I be more productive? I don't need to talk to somebody about it. But like, as I was, you know, having tons of threads, so I had like probably 20 threads going on, you know, and then I was talking to Jeff, Jeff Escalante from Clerk. And I was like, well, how do you— yeah. And he's great. I don't know if you guys have had him on the podcast, but he is amazing when he—

00:36:18 - Dev Agrawal

just last week.

00:36:20 - Tracy Lee

Oh my God, really?

00:36:21 - Anthony Campolo

On the show?

00:36:21 - Tracy Lee

Yeah, literally one of my favorite people to talk to about AI. Yeah, I know. Yeah, he's the best. Um, but you know, he was like, just, oh, you know, I actually color code my whatever. And so I started doing that, right? Like the things that I just sort of have going on today you know, I have like labeled in white. The things that are incredibly urgent that I will delete soon, I have in red. The things that are like ongoing that I sort of need to keep track of, maybe I have in green, you know. Things for family stuff, I have in pink, right? But like, just even that sort of like small little tip has really helped me organize my thoughts, something that I like wouldn't have thought I needed, I suppose. But that's why I think it's so important to, to, to talk to people. Um, so I've been doing a lot of things. Like, for example, I've been filling out a lot of government contracts lately. Like, no, sorry, not government contracts, government forms. Okay. And it is so painful.

00:37:17 - Anthony Campolo

My least favorite activity in the entire world.

00:37:20 - Tracy Lee

But guess what? I'm just having AI do it. And even though I'm still filling out these forms and I'm not done yet, and I started at 7:30 in the morning and was still not done by 1:30, How ridiculous is that, guys? Okay, because, you know, you have to like press okay, okay, okay every single time, you know, um, because they're sensitive documents. But like, the fact that AI is doing it for me, like, if you— if I had to do this manually, I would have quit in the first hour. But it's okay for me to just keep saying yes to something or like brain, you know, troubleshooting with it or whatever, you know. Um, so those types of things, again, I don't even Like when I get an email, so you have to connect everything obviously, but like when I get an email in, um, to register for a conference with like a code or something, I don't even do it. I mean, I just have AI register it for me and it just does it for me.

00:38:12 - Anthony Campolo

I mean, even why you've been so into ChatGPT Work, and I'm curious what systems is connected to ChatGPT Work?

00:38:19 - Tracy Lee

Um, well, I mean, I connected to, you know, I'll connect it to HubSpot, I connected to Slack, I connected to All my Google, like my Google Calendar, my Google Drive, my Gmail.

00:38:31 - Anthony Campolo

Part of my issue is all my shit's on Proton. I don't use Google anything.

00:38:35 - Tracy Lee

Oh yes, yes, yeah.

00:38:37 - Anthony Campolo

So like there's a lot of my, my context I can't bring into ChatGPT Work unless I like find a way to ingest all my Proton stuff.

00:38:44 - Tracy Lee

That is a little bit tough.

00:38:45 - Dev Agrawal

Yeah, I was just gonna say this. Browser use, computer use. It's not as efficient, but it's there.

00:38:52 - Tracy Lee

I, I do this a lot, actually. I do a lot of browser use. Um, I wish you could read— okay, so I wish you could read my like text messages as well, but at the same time, like, you kind of want a little bit of separation, right?

00:39:05 - Anthony Campolo

ChatGPT just cooked into iMessages, they just said like a week ago.

00:39:10 - Tracy Lee

Well, dude, I'm telling you, controversy. Well, here, here's the thing, right? If I wasn't talking to you right now, I would be waiting for that thing to arrive And so this is again why it's so valuable to just like talk to people. Um, but okay, let me, let me tell you. So today I registered for a conference that I had to pay for, and it was asking for authorization. And apparently it was like already connected to one of those like Stripe accounts or Link accounts or something like that. And it was like, okay, do you want me to charge the $299 for this? And like in ChatGPT I said yes, whereas before I like, you know, I wouldn't have said that. I'm like, oh my God, like it's actually gonna make purchase for me. And it did. And I was like, oh my God, it actually— I mean, I really had doubts that it would work, but it actually worked, you know. And I thought that was incredibly valuable. I think there's things like, you know, find me all the tech conferences that I should, you know, apply to, or at this conference, you know, what's missing from this, whatever, or, hey, I'm about to go to this conference, who's going to be there, who's been posting, and it will just basically go and find all the public LinkedIn posts of something and say, hey, these are people maybe you should meet, right? Like, where are all the speakers? Hey, all these speakers are there. Like, who should I actually meet? Who was at— who would actually be interesting for me? What talks should I actually go to? Um, those are really valuable. I think having just like it as a therapist or as like a, uh, like, I don't know, business coach or, you know, whatever you want to call it, but maybe CRM. Right. Yeah. But like, you're about to go into a hard conversation, let's say. And, you know, how do you have that conversation? What are the things you should talk about? The CRM thing is a huge thing as well. Like, I even have like a, a thread of, you know, everybody I talk to about random AI stuff. And then, you know, I have this like AI workshop that I just released, right? Um, so I just have a list of people that I just need to invite to this one, but it's just stuck in that thing. I think my favorite thing, I do a lot of community stuff. And so like, I sit on a few boards. And like, let me tell you, it's very difficult sometimes to have to like create presentations, like present this blah, blah, blah. I'm like, dude, I don't even have time to look at this freaking email. Yes, I basically just said, oh my God, I have this board meeting tomorrow, read this damn email and tell me what to do. And it takes the template, you know, and basically reads all my emails of what I've done and, and bullet points the things that are necessary and creates this deck for me. But it's just like, my God, you know, the thing that I didn't want to do that would have taken me like 2 or 3 hours to like get my brain around took me like 10 minutes.

00:41:55 - Anthony Campolo

No, it's amazing. Do you do, um, do you have, uh, AI listening to your meetings that then takes like meeting notes?

00:42:02 - Tracy Lee

Yes, I do. I do have that as well.

00:42:04 - Anthony Campolo

'Cause I've got a lot of leverage from that recently. I'm building this weird-ass project with my friend and we'll meet every week to do it. We started transcribing all of our conversations and so like when we're talking, we'll just like speak to the transcriber like, remember this, remember that, remember this. And then we'll just have it feed it all into, to Codex at the end. Just have it create us a list of to-dos, you know?

00:42:26 - Dev Agrawal

Yeah.

00:42:26 - Anthony Campolo

Yeah.

00:42:26 - Tracy Lee

And you do have to sort of troubleshoot it, right? Like you do have to sort of play around with the AI, which, you know, again, is sort of this, you know, this thing that we've been repeating on this call of, you know, the human transformation and really understanding the tool and stuff like that is like, you know, still necessary to start early, right? To stay on the curve or ahead of the curve.

00:42:52 - Dev Agrawal

Yeah, for sure. What's the reverse centaur book?

00:43:00 - Anthony Campolo

Who knows? I'll have to look it up like usual. Yeah. What else you got, Dev?

00:43:11 - Dev Agrawal

Yeah, I'll have to look through a little bit.

00:43:16 - Anthony Campolo

Are you messing around with image or video generation yet?

00:43:22 - Tracy Lee

Yes. So another silly project. Like, I decided to remodel a house and, you know, I decided not to hire a designer. I have regrets in a lot of areas. But, you know, getting AI to basically generate all the images for me, be able to talk to my GC, right? And say like, here's the pic, you know, because how are you describing what your cabinets want to look like, right? And, you know, AI can get it pretty good. And it's so crazy because I started this about probably about a year ago. And before, ChatGPT was awful at creating images. I would have to go to Gemini to create the images. And then one day magically Gemini sucked and ChatGPT got better. It actually, you know, for ChatGPT Work, I actually don't find it very good. I feel like just using ChatGPT is actually the best for image generation. Um, but I've done that, you know, it taught me how to be an architectural lighting designer, another thing I should have paid money for. Um, but it's, it's, it's pretty, it's pretty fabulous. You have to, you know, obviously like watch for the hallucinations and things like that. Um, I haven't done the video generation yet. But, um, I, you know, another thing I've always wanted to do is become like a songwriter. You know, but like it's not like I'm the most amazing.

00:44:46 - Anthony Campolo

Are you using Suno yet?

00:44:47 - Tracy Lee

Well, here's the thing. Well, isn't Suno going away or like?

00:44:52 - Anthony Campolo

No.

00:44:52 - Tracy Lee

No.

00:44:53 - Anthony Campolo

Okay. Wait. What's this?

00:44:55 - Tracy Lee

What's the thing?

00:44:57 - Dev Agrawal

No.

00:44:57 - Tracy Lee

What's the thing that's good? What the the the video? It was the video generation. It was um Sora. Sora.

00:45:03 - Anthony Campolo

Sora is going away, which is I'm so sad. I love Sora. I generated so many amazing videos with Sora. I generated a great video with Sora where I showed up. At the Oakland airport and got instantly beat up and robbed. So funny. But total tangent, Suno is the best AI music generation model. I, I'm a music major. I'm actually a musician by training.

00:45:24 - Tracy Lee

Oh my gosh.

00:45:25 - Anthony Campolo

Suno music generation is better than 99% of living musicians. Dude, like, it is so good. It's really good.

00:45:34 - Tracy Lee

I'm gonna do it because like, I've always, you know, I've always wanted to, but like to, to sort of like learn this and learn that. I mean, you, you know, Anthony, if you have like a, if you have like a, you know, a background, right? Yeah. But, but the thing about it, right, is so I listened to this song last night. What was it called? It was called this like bourbon, uh, bourbon something. And I was like, wow, this is such a great song. And I sent it to my friend and he's like, that sounds AI generated. I'm like, is it? Is it AI generated? Turns out it was. But it was awesome. And I'm like, man, if I could like get my thoughts out and like generate songs using AI, I mean, that just sounds amazing to me.

00:46:16 - Anthony Campolo

Um, yeah, we should do a whole stream about AI-generated music. But with Suno, you could do 2 things. You can prompt it to generate a song, or you can give it a set of lyrics with styles that will turn the lyrics into a song. That's what I like doing.

00:46:30 - Tracy Lee

That's what I want to do.

00:46:32 - Anthony Campolo

Yeah, yeah. So if you already got lyrics, like, go to suno.com, create an account, mess around with it. It's $10. Like, it's very, very cheap.

00:46:38 - Tracy Lee

I'm, I'm literally like, like, I need another side project. Um, this song was called Cigars, Bourbon, and No Things, and it's by the Lincoln Collective. Um, I heard it on my Instagram yesterday, and I totally fell in love with the song. Um, but it is AI generated. But, you know, it's a dude. It's a dude sitting, smoking a cigar. You know, talking about, uh, Pappy. I was like, who is this new artist?

00:47:04 - Dev Agrawal

How cool.

00:47:06 - Tracy Lee

So I guess, I guess you have to be in the know to know. So you can still probably make a ton of money, um, doing so.

00:47:16 - Dev Agrawal

Nice. So AI can generate code. Oh, sorry, go ahead.

00:47:20 - Anthony Campolo

So this is the centaur thing. Take the original centaurs, mythos, Throw it into a box, twist it till it's so contorted that the horse is now the head, as you could take it in the hay and you have the human doing all the legwork. I'm still so confused, Fuzzy. Does anyone understand what's happening here? So I—

00:47:40 - Dev Agrawal

okay, my understanding is that reverse centaur, basically instead of a human head and a horse body, it's the horse head and a human body. So it's like the brain, you don't have the human brain.

00:47:53 - Anthony Campolo

Stupid horse brain.

00:47:55 - Dev Agrawal

Exactly. That's kind of the point. That's how, that's what AI is going to turn us into. Like we are not going to be doing brain work. We're going to be doing legwork.

00:48:04 - Tracy Lee

Yeah. Yeah. I mean, it's, I mean, I do, I do feel like that is the case right now. Right. Like for example, Calypso leaving, I can't get anybody to pack my boxes for me for my offsite. I have to do that or hire somebody. I hired, um, a 6th grader down the road to help me do that.

00:48:23 - Anthony Campolo

Where are you based at, Tracy?

00:48:25 - Tracy Lee

I'm in Atlanta. Atlanta. Yeah, I mean, if you wanted to come over and help me with these boxes, like, we could help you figure something out.

00:48:33 - Anthony Campolo

Yeah, if we were right down the street, I'd be like, let's fucking go.

00:48:39 - Tracy Lee

Oh my God, I'd be so happy. I would be dying. I'd be like, thank you for saving my entire life right now. Uh, but yeah, yeah, my—

00:48:50 - Dev Agrawal

I, I do wish that we, we got similar advancements in robotics like we have in, uh, like AI models so that we can actually like skip some of the legwork as well. I don't know, did you guys see Hugging Face just released like a little duck robot, uh, a few—

00:49:06 - Tracy Lee

I think I saw that.

00:49:08 - Anthony Campolo

I saw something though of like AI, like I saw the one where AI was like a in a sprinting conversation, in a sprinting competition, and just ran really fast and then it just like ate shit and immediately exploded. Like, that's, those are all the AI robot videos I see. Something like doing something really cool and immediately failing.

00:49:30 - Dev Agrawal

Yeah.

00:49:30 - Tracy Lee

Yeah. But if it can get you, you know, I mean, you think about like, I've definitely seen some things about like, you know, these, these robots, or I don't know, are they called humanoids, right? Like being built and, um, you know, being your companion. So I think that's kind of interesting, or like just having a friend to talk to, if you will. I think, uh, being lonely— I mean, maybe, you know, actually this would be really good because I don't know if you guys have had family members that have gotten conned into those like Bitcoin scams. Like, you know, I've fallen in love with somebody and he's made a lot of money on Bitcoin scams.

00:50:04 - Anthony Campolo

Shut your mouth.

00:50:05 - Tracy Lee

You made a— well, more what—

00:50:07 - Anthony Campolo

hey, I got into Bitcoin 2017, so I was ahead of the curve.

00:50:12 - Tracy Lee

But that's not a scam though. The scam I'm talking about, which is my favorite scam right now, is on LinkedIn. Sometimes I don't know why I'm being profiled by Asian dude profiles, but sometimes an Asian dude will add me on LinkedIn and he will, uh, he will, you know, start talking to me. About things. And I'm like— and, you know, he's like handsome, you know, I mean, you know, try to be handsome anyways. And like, you know, he has a really good, you know, like, oh, I worked at, you know, Netflix for 10 years as a blah blah blah or whatever. He starts trying to have a conversation with me, and, you know, you know, he's just trying to get me to fall in love so that he can get some Bitcoin something. Like, I'll buy Bitcoin because I'm in love with him or something. Type vibes.

00:50:56 - Anthony Campolo

The closest thing I ever had to being scammed by that is when I tried to buy weed in New York City and went to a fake website and actually paid a bunch of money for weed that never arrived. That's the closest thing I have.

00:51:10 - Tracy Lee

Yeah, well, maybe AI can help though with loneliness, was my point, right? Like, if a lot of women and men are getting scammed by just being lonely and, you know, getting into these sort of like whatever, like AI can kind of replace that that pain for people, you know, or that, that desire for people.

00:51:29 - Dev Agrawal

Yeah. So if AI can do all these things, what are humans for anyway? Like, why should companies not be firing human employees? I don't know, it's a kind of a weird left-turn heavy question.

00:51:46 - Tracy Lee

I mean, yeah, but I mean, honestly, like, you look at this and it You know, here's, here's what's funny. We amplify as humans, we amplify what our actual, what our actual thoughts are. So when I see people posting about like, oh, you know, IBM said they made a huge mistake laying off all these developers because of AI. And now they're starting to hire developers back. Right? You can kind of get an idea of like what this person believes, right? Or like, Or like, um, another thing is like, oh, you know, AI is taking away all of the like basic principles of, of management. You know, you can kind of think, okay, well, I don't know, maybe like you aren't using AI as much, right? You can kind of like make some assumptions about what that person's doing. So I think like there's, there's so many things in the world that you can kind of be like, yes, this is going to happen. Or like, you know, it just depends on like what sort of media you decide to gravitate towards, if you That will, I think, form your beliefs. Um, but I mean, one thing I do know is that if you're not adopting AI, then like, just like we said earlier, right? Like, if you don't adopt AI, AI is going to take your job. Uh, well, maybe I am going to take your job, but somebody using AI will take your job. And so what do you do, right? Like, it's— I think it's really important for everybody to just sort of like get going and then evangelize as much as possible what you're doing. Share it with your teams, share it with your, you know, whomever. Because, um, I mean, it's crazy. I mean, you see these things on Instagram, right? I spun up 3 businesses that are making me $100,000 a month, and I just am using AI to do it. But you know what? It's real. Like, that is actually reality. And, you know, we should probably capitalize on the, uh, the gold rush. As quickly as possible.

00:53:47 - Dev Agrawal

Yeah, that makes sense. I think, uh, yeah, I'm, I'm trying to figure out like, okay, if, uh, there's a lot of people trying to get into tech or people who got into tech recently and are feeling a lot of anxiety around like AI is going to replace their job. And to me, to me it's like very obvious that that's not gonna happen as long as you know what you're doing and you adopt AI yourself. But I think it's hard to communicate the value of the human in the engineering process. Like, for me, it's like, uh, it's not just— okay, yeah, the first thing is that AI is just bad at like large technical decisions or architecture decisions. Like, those are something that you gotta get better at and make. But apart from that, there's I don't know, people keep talking about this thing called taste, which I don't know if I fully get.

00:54:41 - Tracy Lee

The best keyword ever these days, right? Humans have taste. Yeah.

00:54:47 - Anthony Campolo

Um, can I just jump in real quick?

00:54:50 - Tracy Lee

Yeah.

00:54:50 - Anthony Campolo

I think that's an interesting point about taste because I prided myself on having good taste because I was into art before I was into tech, and art is all about taste, you know? Like, yeah, you don't have good taste, you're not gonna be good at And so like taste is like the high, it's like Dev was saying, you gotta be good at architecture decisions. So it's the higher level thing, like knowing when to use Postgres versus when to use SQLite. Like that is taste. I think when people say taste, they're really just talking about like what is the higher level decisions that needs to be made that AI can't quite do. That's what I would say.

00:55:24 - Dev Agrawal

Yeah.

00:55:25 - Tracy Lee

And I mean, it even goes, you know, if you, if you talk on the sort of like more ops level thing, right? It's like, you know, you put together a plan, and, you know, AI can put together a— so you can always tell, like, somebody who just started using AI versus somebody who, who hasn't, right? Because probably even maybe similar to like a junior developer or an intern, right? Like, AI will sort of like one-shot something and do a pretty good job. But like, it'll bury the, the lead. And so somebody with you know, taste will say like, actually, it should be presented this way and tell AI, actually present it this way to more clearly communicate my, my thing versus like, hey, here's a 10-page document that, you know, people who are using AI know. And like, they don't want to read through 10 pages of slop. They'd rather know like, blah, blah, blah. I think that's also sort of like an example of taste that I think is so important to have the human in the loop. Yeah, I don't remember your first question. You said something about taste and then me and Anthony derailed.

00:56:29 - Dev Agrawal

What's the role of the human in the process?

00:56:32 - Anthony Campolo

Yeah, what does taste mean in terms of using the AI? Yeah.

00:56:36 - Dev Agrawal

And I guess more importantly, how, like when you go to organizations, when you go to teams, like how do you teach them this or in workshops?

00:56:44 - Tracy Lee

Yeah.

00:56:45 - Dev Agrawal

Like, okay, this is what you should do by yourself and this is what you should delegate to the AI.

00:56:50 - Tracy Lee

Well, I think that really becomes the, um, this is why senior developers are so in demand right now because they sort of know, right? Like, you know instinctually what you should do, right? And, and kind of like what can happen. So it's the same as any, you know, document that can be generated, right? Like now I think the thing about it is, you know, the lowest barrier of doing something So for example, like a marketing— like I just recently put together a marketing plan, right, for, for, uh, for a local conference. Um, and, you know, AI could have, you know, did generate something for me in the beginning, and I was like, okay, well, this is pretty good, but I know that actually you need to, you know, announce the speakers here and do one post per whatever, and you should announce the panelists. And I sort of know, like, the path to which, you know, to just to like make it a little bit better, if you will. And I feel like that's the same in development, right? Like, you know, AI can do a good enough job and, you know, maybe we shouldn't, you know, deviate and start getting into conversations about like, does code quality matter these days or not? But it's the same thing, right? It's like, you know, you sort of know. Like another good example is like, you know, we had AI generate like VibeCode a bunch of these websites, you know, I can look at it and say, okay, on the main page you have all these events, on the workshops page you have all these events. I make a change on the workshops page and it's not changing here. So like, I know they're not using the same, you know, the same reference, like in the code, if you will. And then I'll say, okay, well, go ahead and do that, right? Like, AI is not going to help me do that. AI is just going to say, yeah, it's perfect, that's what you wanted, right? Um, so I think it's like some of those small little things that like still matter for humans to, to be checking. Um, but again, we can go through that. We can go through conversations of, look, like, you know, the amount of code that needs to be sort of reviewed these days. Like, when, when is it going to get to the point where like humans aren't doing code reviews anymore? What actually matters to be reviewed versus like what actually doesn't matter? To be reviewed. Um, makes me feel like actually I am missing something in my CI/CD to, to sort of tell my team like, these are just text changes, but these are actually logic changes. Actually review this code, but merge the other one versus YOLO, you know.

00:59:19 - Anthony Campolo

I want to hit something that's kind of similar to that, which is— and this is near and dear to both me and Dev's hearts— which is What do you think about AI in terms of making decisions around a stack? Like, AI is going to pick React unless you tell it otherwise. Should you tell it otherwise, or should you just let it pick React?

00:59:40 - Tracy Lee

I think, like, you know, there, there could potentially be a reality at some point in time where, you know, developers are less in the loop, right? We talk about code reviews or not code reviews, right? But I mean, I think there is something to be said about like some level of human maintainability for some period of time. Who knows if that's 7 years, 10 years, forever, or 6 months, right? And so I think it's still important to rally around technologies that your team can manage, if you will. But, you know, I mean, we've all had those conversations of like, well, you know, frameworks were built for, for humans and, you know, AI doesn't need them. And it, you know, it's better to just, you know, write to the DOM or like blah, blah, blah, whatever, you know.

01:00:34 - Anthony Campolo

Um, and, um, I've removed most of my dependence. I have Bun doing almost everything except like, I basically, I, I would say Bun, Solid, BetterAuth, Stripe. I don't know if I need any other dependencies.

01:00:49 - Tracy Lee

Yeah, yeah, yeah. And it's, it's, it's crazy, like, as certain tools become like better and the models get better, how much it just— like, I remember, um, gosh, I remember using, um, what's this, Zapier. So I was using Zapier for a few things beforehand to like sort of automate some of the things I do for conferences and things like that. And, um, now it can just be like a little, you know, one of my little threads happening in ChatGPT Work, and I don't actually need Zapier anymore. So I don't know how much I was paying Zapier, but it was probably like $100 or $200 a month or something, you know, gone, right? Um, but again, I think, I think we, we go back to the, the sort of theme here of this whole like walking, walking down the path. Of AI is so important just to see what— you don't know what you don't know. And you know what I think is really weird though? I don't know if you guys have run into this, but like people sometimes think they're using AI and you're just like, no, you haven't unlocked. Like you can kind of, you know, when somebody hasn't unlocked, but like you don't know how to— I don't know if you guys have advice for that, but if you've ever run into a developer and they're like, yeah, I'm using AI for this and this and this, and you're just thinking No, you're not. You think you are, but you're not.

01:02:11 - Anthony Campolo

A little more specific, like, why do you say no, you're not? Like, I can imagine, you know, like, you're just asking ChatGPT a question every now and then, but like, you're not actually asking code, you know? And so you can't have it touch your code base. Like, that I would imagine would be an issue.

01:02:25 - Tracy Lee

It's not about that. Like, it's even, even getting into, hey, I'm using AI to write my tests. Hey, I'm using AI to automate some of my CI/CD stuff, right? Like Hey, I'm using AI to write some code. Like there's just, you know, like another level. Like once you—

01:02:42 - Anthony Campolo

yeah, they're not thinking deeply enough about what they're having it do. They're just saying, go do this and then have it comes back.

01:02:49 - Tracy Lee

Yeah. And, and I think that like, you know, I've definitely seen some people get stuck right there where it's like you've adopted it, but like, you know, everybody on this call, you know, has, has gone like to level you know, 100, and you can just see that these, you know, some of these developers are like stuck on like level 30, you know, like they've gone deep enough. It's not like they're just using autocomplete, but they just haven't—

01:03:16 - Anthony Campolo

for a couple more levels, right?

01:03:19 - Tracy Lee

And but, but the thing is that they think that they've reached the mecca.

01:03:24 - Anthony Campolo

Like, yeah, they don't know that they're at a lower level, right?

01:03:27 - Tracy Lee

But what do you, what do you do? Like, what is the advice you give to some of those developers, smart developers?

01:03:31 - Anthony Campolo

Yeah. No, that's a really good point. I would say you try and explain or show them what they could be doing. And that's why there's a, there's a dude who I really want to bring on the show. His name's Jeffrey. I forget his last name, but he, he builds all these skills. He builds all these projects and he's like the second, like literally like the number 2 top committer on GitHub globally. And he just has like the craziest projects that he's like running through AI. And like, you could point to that and be like, bro. You could rebuild Redis right now. Like, you don't even know. You don't even know.

01:04:04 - Tracy Lee

Yeah, yeah, it's, it's really hard, right? Because like part of it is understanding the tooling, part of it is just the human mental block, you know? So I think, I think again, you know, like people, people talk about, you know, there's a lot of, you know, we've all heard the fear in the industry of like, oh, AI is going to take our jobs, etc., etc. But like Dude, I mean, what, in 7, 10 years? I mean, it's, you know, transformation takes a long time. So many people aren't even using AI right now. Not to say that you should be happy with not using AI because so many people aren't using AI, but it's like, there's, you know, there's a lot, a lot of work to be done to transform, um, an industry. I've heard some people say that like we're just gonna go back to like regular coding, and I'm thinking Okay, and here's a— here's another thing I have to ask you guys. So before, when we were token maxing, right, and you just got access to everything, uh, you know, life was great. But a lot of companies now are pulling back and saying like, actually, you're limited to X amount of credits a month or X amount of dollars a month spend. Um, if you've gotten your developers to use AI in their workflow now And then now you're saying like, actually you can't, you know, hey, I, I hit $500. Like, you know, you can't use AI anymore. Then what? Like, I feel like that's so crazy, right? 'Cause like you actually change somebody's workflow, it's a tool, and then you just take it away.

01:05:38 - Anthony Campolo

Yeah. I mean, I, I pay for a subscription to ChatGPT, Claude, Grok, and Gemini now, so I never run out of AI credits. So I'm thinking, so I feel like if, if a developer hits that, they might just go buy their own subscription to be like, I'm gonna drop $200 just to get me through this week. Mm-hmm. Because I have a problem I need to fix and I need the AI to do it. So I'm gonna go buy my own subscription.

01:06:07 - Tracy Lee

Like, that's what I'm thinking.

01:06:07 - Anthony Campolo

Yeah.

01:06:07 - Tracy Lee

So like more shadow AI, more shadow AI basically will just pop up.

01:06:11 - Anthony Campolo

Yeah. That's what I would do. I don't know about other devs.

01:06:13 - Tracy Lee

Yeah.

01:06:15 - Dev Agrawal

I think for me, what I usually end up doing is I just start writing down things because I know that once my rate limits reset, or basically once my rate limits reset, I have a very detailed specification of what I want and I can just let it go to work. Whenever I have downtime, I just start writing down whatever thoughts I have or looking at the codebase myself and trying to figure out how something is built, how some— because I need to at some point, I need to go review the code anyways. So I might as well do that when I don't have AI credits. Or like in the worst case, I just go and touch grass. Yeah, which is pretty rare. But yeah. Yeah, but this only happens because I know that eventually it's coming back or like eventually the rate limits will be reset and I'll have access to it again. Um, if you just completely took away the AI tools, then I don't really know. Like, I mean, obviously I would go back to hand coding, but I don't think I'll, I'll be enjoying it as much.

01:07:22 - Anthony Campolo

I think most people quit at this point, right?

01:07:26 - Tracy Lee

Yeah. Yeah. So I have another question for you guys before you ask me a question, Dev. When are you getting JLarky on this podcast? Because obviously he has a lot of opinions.

01:07:37 - Dev Agrawal

I was just going to DM Anthony about with that exact same question.

01:07:43 - Anthony Campolo

Get them all out of the way at once.

01:07:47 - Tracy Lee

Love it. I love it so much, though. Okay, Dev, sorry to interrupt you.

01:07:51 - Anthony Campolo

No, no, all good. Bring on Jay Lark. We're going to bring on Fuzzy. We're going to bring on Brad. We will bring on anyone. I don't give a fuck.

01:08:01 - Dev Agrawal

Yeah, everyone has fun things.

01:08:04 - Tracy Lee

But you know the thing about it, though? Dev, when we were talking at Chain React, right, sort of like that 4 hours after the conference, I was like, Dude, you guys have a lot of— like, for example, saying that like Clawd Co's is like a super shitty, like the shittiest harness out there. Who's gonna say that publicly? But I'm like, guys, can you please— like, why am I not recording this conversation? Because they were such good— it's like, yeah, but it's true, you know? And, um, you know, I'm not saying that, okay? But I'm just saying that that was what I heard. But like, nobody's sharing those opinions. And so whenever, like, Jay Laker, you better get on this, and Fuzzy Bait, like, you guys better get on this podcast and share your opinions. Don't start getting like watered down just because you're actually live now, you know? It's, it's not safe in the comments.

01:08:50 - Anthony Campolo

Fuzzy Dev has been on before. This will be— if he comes on again, it'll be his second appearance. Yeah, he already— and he's not— he was on FS Jam too. He was one of the very last guests on it. Yeah, actually.

01:09:00 - Tracy Lee

But is he watered down when he comes on though?

01:09:03 - Anthony Campolo

No, of course not. No, no. And me and Dev, probably the opposite. Like, we, we will always speak our thoughts. That's one of the reasons why I think that people who do watch our content enjoy it is cuz like we, we will never ever water it down.

01:09:18 - Tracy Lee

I love that.

01:09:19 - Anthony Campolo

We don't, we don't believe in that.

01:09:20 - Tracy Lee

I love that so much. Yeah. I'm just gonna start taking a recording to conferences and be like, can you just say that really quickly? And then get people on podcasts and be like, let's just play this back, guys. You said this, give me your reaction.

01:09:32 - Anthony Campolo

This is why you gotta ask the conference host, like, can I sponsor this as a podcast host? And then just record. Like, that's what we did for JS Jam. That's, uh, I don't think we ever did it for FS Jam, but, uh, me and Dev have been trying to do this, but, um, we, we haven't been able to line it up recently.

01:09:46 - Tracy Lee

Yeah.

01:09:47 - Anthony Campolo

Um, yeah, but it's, uh, we're, we're, we're as free as we can be with our sponsors.

01:09:53 - Tracy Lee

I love that. That's so important though. It's so important. It's so important to like have that in the industry because those are the opinions that matter, right? That's the— those are the opinions that are going to push the industry forward, not the whole like let's be nice blah blah blah BS, but like the real true opinions that, that matter.

01:10:16 - Dev Agrawal

Yeah, conference hall recordings are basically impossible to hear or to listen, uh, probably transcription is gonna be more helpful. Uh, this was in the hotel lobby though, so it was, uh, relatively quiet.

01:10:28 - Tracy Lee

Uh, yeah, it was pretty chill.

01:10:29 - Anthony Campolo

Oh man, I wish me and Scott were there putting hot mics on.

01:10:35 - Tracy Lee

I know. Yeah, somebody should do like a conference called like Hallway Track. Wouldn't that be so cool? Like have a conference.

01:10:41 - Dev Agrawal

I think Tanner—

01:10:42 - Anthony Campolo

yeah, the conference track is what you want on a podcast. Like, because that's where people are free, they're like kind of speaking their minds, you know?

01:10:50 - Tracy Lee

Yeah, yeah.

01:10:51 - Dev Agrawal

I think Tanner Linsley had something like that kind of like going around. He even made a Discord server and invited like a bunch of people to, uh, who might be interested in either sponsoring or doing something. Uh, and I think that the idea was basically that like it's mostly going to be just, uh, a hallway track, like discussions and less of talks. I don't know, it was a while ago. I don't think there is any activity there anymore. But might be worth trying to restart that conversation at some point. Actually, the question I was going to ask was the same one as Anthony. I was just going to ask about RxJS more rather than React. But yeah, kind of, I guess, in the similar vein, like when it comes to like tech stack choices, like you mentioned, you have been a core contributor to RxJS. Obviously, you work very closely with Ben Lesh and some other folks. So what— I don't know, what's the state of like libraries like RxJS these days? Is that something that you use with AI or like benefits the AI coding somehow or makes it worse?

01:12:02 - Tracy Lee

I think— I mean, I think, I think, I think everything benefits, right? I mean, this is like the, the whole thing when it comes to taste, right? Like If there's something that can help AI that AI doesn't, I mean, this is like going back to like skills and this and that and whatever, right? But if you're able to introduce something to AI that it wouldn't maybe necessarily pick in the first place, it's incredibly valuable because you know, right, with your taste, with your opinion, with your human, as a human, that that is going to improve something. So I think that's incredibly valuable. I think that, uh, you know, it's funny because everybody cares about AI now. Nobody cares about like what are the latest up— I mean, some people might, but like most people are not reading the whole like what's the latest in Remix, although I'm excited to hear what you guys say in the next podcast, or like what's the latest in React, or what's the latest release and blah blah blah, right? Before, that was like the most important stuff that came out. Now AI can do a lot of it for you. So a lot of people are not reading that anymore. But standards are still being pushed forward, right? Like TC39 is still having their conference here soon, right? The standards committees are still meeting and talking about things. So, you know, RxJS and Ben are still very active in those conversations and pushing standards forward. Not much has changed in terms of the, like, framework or format in which those committees and things happen, right, in terms of their internal or external processes. But what has changed is a little bit of like what conversations are actually happening because the web is changing. And so I think a lot of those conversations is like, okay, what matters to the web now with AI, with these new technologies, versus like I feel like starting in 2020 or so. Thank you, COVID. Like, think— or maybe just thank you to, hey, JavaScript has been around for like— well, I didn't, didn't mean JavaScript has been around for 5 years, but I, I sort of think of the timeline like 25 years at that point. 2020, 2015.

01:14:17 - Anthony Campolo

Yeah, React came out in 2015. By 2020, it was 5 years old. 2020 was the year of the React meta frameworks. That's when Blitz, Remix, and Redwood came out.

01:14:24 - Tracy Lee

Yeah. Yeah.

01:14:25 - Anthony Campolo

But like, 'cause I did my bootcamp.

01:14:28 - Tracy Lee

Yeah. But like the, the, you know, I, I think, I think like, you know, actually I forgot my point, but you get my point. You guys get my point.

01:14:41 - Anthony Campolo

Well, it was actually, I was gonna tangent real quick 'cause Jalen, you're sitting here, I get excited. I get excited for new web releases. Your agents can't do it for you. I'm gonna push back and say we had like the, the Bun 1.4 release and like it's a big release. Like they add all this different crap. Oh, we got Nicky T in the chat.

01:15:02 - Tracy Lee

Yes. When is Nicky T coming on the podcast?

01:15:06 - Anthony Campolo

Nicky T was just on like 2 months ago.

01:15:07 - Tracy Lee

Oh my God.

01:15:08 - Anthony Campolo

Nicky T is the, the most featured guest on this podcast.

01:15:12 - Tracy Lee

He's been on, oh, that's the best.

01:15:12 - Anthony Campolo

We figured out last time he had been on for like 6 times. Because me and Nikita are such homies. I love Nikita. But, um, what I was saying is you, I gave the 1.4 Bun release to my agent and said, read my whole repo, figure out what all of these updates mean, write me a plan to tell me what should be upgraded. Like, how is this new release relevant to my codebase? And if it's a well-written release, then it has all the information it needs to do that.

01:15:44 - Tracy Lee

Yeah, yeah, yeah. And, uh, Fuzzy Bear, I do remember when I kidnapped you, and that was so much fun. But yeah, people have to be close. Like, I'm really into bourbon right now, and so, uh, one of my bourbon friends, like, does not— like, he's like, no, I don't want to be on camera. But like, you know, then he gets a little bit too close, and then he starts like talking to me about bourbon, and I'm like looking at him thinking Man, this guy's been talking for five minutes, like, and there's a camera in front of me. And I was like, "Didn't you say you don't want to be on camera?" And he's like, "Oh shit!" And then like he walks away. You know, I'm like, "I'm not cutting that, bro," because it was good. So yeah, it's my favorite thing to do.

01:16:24 - Dev Agrawal

I think I've only seen Fuzzy in Fuzzy's face in a video once, and it was an interview. Was it you or was someone else? I don't remember. I mean, it's like an in-person interview.

01:16:36 - Anthony Campolo

FS Jam was audio only. He— right on screen, no one else got to that special moment.

01:16:44 - Dev Agrawal

Yeah, no, I remember this was like an, uh, in— like in-person interview with 3 people, and, uh, like one person was asking questions and Fuzzy and another person were answering. I don't— it was a while ago, I don't remember.

01:16:57 - Tracy Lee

Yeah, um, he can send us the link.

01:17:00 - Anthony Campolo

For us.

01:17:00 - Tracy Lee

You can send us the link if he is like, you know, not featured often. He probably has the like 3 links in which he's been on, uh, his face has been shown on video.

01:17:09 - Anthony Campolo

He was on— he came on my stream. I think he went— I think he was on camera for— I'll have to figure it out because it was a— it was a— it was my buddy Monarch. I— this is before I was doing streams with Deb. My buddy Monarch did like 4 or 5 streams with me, and Fuzzy, we were in the chat and he was talking about how much he hates AI, and I was like, Fuzzy, like, come on the stream, let's talk about this. Like, we need to work this out, buddy. He actually came on, and that was like— I mean, that was like 2 years ago. It was a while ago.

01:17:38 - Tracy Lee

I mean, it is absolutely crazy to just sort of see how things have changed, right? So like, you know, one of our lovely friends Ken Wheeler, right, before he was like—

01:17:46 - Anthony Campolo

what'd he say?

01:17:47 - Tracy Lee

He's like, no, you know, if, if I have to use AI, then I don't even want this job because like he likes the artisanal Blah, blah, blah of, you know, of, of coding. But then, you know, slowly over time you see Ken, you know, using AI and, you know, sort of like overcoming his own little whatever.

01:18:05 - Anthony Campolo

He was just kind of full of shit. I feel like he was always talking shit about it, but he actually was using it and he just, he hated the vibe that the, that the AI companies put out, which I totally get.

01:18:19 - Dev Agrawal

Yeah.

01:18:19 - Anthony Campolo

He hates the aura that AI people put out, but he didn't hate the tech.

01:18:26 - Tracy Lee

Let's see where Ken Wheeler is working now. Ken Wheeler.

01:18:31 - Dev Agrawal

Is that even public info?

01:18:33 - Tracy Lee

Um, well, no, I'm not, I'm not saying anything, but, um, I'm just looking to see, uh, see what his, uh, Instagram or his Twitter says. But, um, no, dude, bro. I wasn't gonna say anything and you just said something and I didn't say anything and you didn't say anything. We, but we all didn't say anything collectively. Anyways, cool.

01:19:02 - Dev Agrawal

Uh, yeah, I mean, what else do we want to talk about? We can, uh, we can probably go on.

01:19:07 - Anthony Campolo

I need to do this and share my screen and show Fuzzy. So everyone can see him in his glory. Not this one. Oh, this one.

01:19:17 - Tracy Lee

There we go.

01:19:18 - Anthony Campolo

Fuzzy. Look at him.

01:19:23 - Tracy Lee

And that is, that is the face he's currently making. That is the face he's currently making.

01:19:30 - Anthony Campolo

So Auto Show and Local LLMs streamed on May 21st, 2024. You wanna see Fuzzy? There you go.

01:19:37 - Dev Agrawal

And you were working on AutoShow 2 years ago. I did, I did not know that.

01:19:41 - Anthony Campolo

Yeah, I know. Yeah, AutoShow is over 2 years old. I think I started April of 2024. Yeah.

01:19:47 - Dev Agrawal

Cool.

01:19:52 - Tracy Lee

We got—

01:19:53 - Dev Agrawal

I mean, we could talk about new hot AI tools, hot frameworks.

01:19:58 - Anthony Campolo

I mean, I want to be respectful of Tracy's time. Um, do you have a course?

01:20:01 - Tracy Lee

I do have to run. I'm so sad. I apologize. I need to put the children to sleep.

01:20:06 - Dev Agrawal

So of course.

01:20:08 - Tracy Lee

Yeah.

01:20:08 - Anthony Campolo

This was great though. Really, really fun conversation. So, um, feel free to wrap it up and tell our guests any, or our listeners, anything else you got to mention.

01:20:18 - Tracy Lee

Yeah. Well, I mean, you know, if you, if you see anybody at the executive level who's like struggling with anything when it comes to AI, just point them at me. Because I would love to just, you know, again, it's this whole like this vibe coding thing, right? Like I'm just gonna open up my, like, I can't record it because I'm showing you really sensitive information, right? But just seeing how I work through things, I think is so empowering for people. And I also feel people that like are a little bit further removed from, that aren't developers, are a little bit scared, you know? And so if you just give them a safe space, and say like, hey, whatever. So you can point them to the, um, the executive workshop that we have on our main page, this.co, or you can just like feel free to drop me a DM or an email or something like that and say, hey, can I introduce you to this person for whatever? Like for me, you know, in 2015 when CLIs were coming out and Angular was so awesome and Ember was awesome and React was awesome, I was just all about sharing, right? Because like the more you know, the better you are off in life. And I really believe that. And so like the more that we can all share, and I encourage everybody listening, can just share what they're doing with AI, you know, the better off we're gonna be as an industry. And, you know, the faster we can sort of like go through this change together. So that would be one thing. If you like bourbon, let me know because I'm happy to either send you some or, you can send me some if you have like a Pictures 20 running around. I definitely know the guys, and I've been doing a lot of bourbon shares, and I have been like, guys, look at the bourbon that's at my house.

01:22:00 - Anthony Campolo

And so obvious.

01:22:02 - Tracy Lee

Yeah, they, they have said— yeah, I mean, we've— I've partnered with them quite often in the past. Um, not so much for podcasting. I mean, I've been on their podcast, but like, you know, we've collabed on a Like on a, you know, collab level, if you will.

01:22:16 - Anthony Campolo

Yeah, they're showing great big—

01:22:17 - Tracy Lee

yeah, yeah.

01:22:21 - Dev Agrawal

Honestly, the way that you described your workshop kind of sounds like therapy for AI psychosis, but yeah, I'm sure it's just practicality. Yeah, yeah, for sure.

01:22:34 - Tracy Lee

Yeah, yeah.

01:22:36 - Dev Agrawal

Cool.

01:22:38 - Anthony Campolo

Awesome. We'll wrap it up here. Um, we just sent out a bunch of scheduling links to people in the chat. So if you want to come on in the next couple of weeks, check that out. Um, I'm going to be going on vacation soon, but we do have slots open in the next week.

01:22:52 - Tracy Lee

Yes. Yes.

01:22:54 - Anthony Campolo

But thank you, Tracy.

01:22:56 - Tracy Lee

Yes. Yes. And the twins are doing good, by the way, Fuzzy Bear. The twins are doing good. They're getting older. They're 2 and a half. So. I will always have my hands full, but they are very cute.

01:23:10 - Anthony Campolo

Awesome.

01:23:12 - Dev Agrawal

I'm sure they will be forever.

01:23:14 - Tracy Lee

Yeah.

01:23:14 - Anthony Campolo

All right, we'll call it here. Thank you everyone for watching. And Tracy, just stand up for a second before we end it, and we will catch you all next time.