Action absorbs anxiety (Friends)

29:11What we're doing at Auth0 is using our 10 plus years of identity developer tooling to make it simple for developers, whether they're working at a Fortune 500 company or working just at a startup that right now came out of Y Combinator, to build these things with SDKs, great documentation, API first types of products and our typical Auth0 DNA. Friends, it's not if, it's when. It's coming soon. If you're already building for this stuff, then you know. Go to Auth0.com slash AI. Get started and learn more about Auth4GenAI at Auth0.com slash AI. Again, that's Auth0.com slash AI.

29:58I was a little worried about open source there for about a week, maybe a week and a half, and I started to see more and more of, like when we had the conversation, Jared, a little while back, just talking about if I could generate code so quickly, does it make open source no longer valuable? Because we tend to open source full frameworks, full ideas, community-led things that have a lot of different facets and bug fixes, and it's very, very rich. It's a very rich thing. and I had been like really, really worried, I suppose, almost like deeply sad in my heart, like is something going to change fundamentally with open source now that it's so in quotes, so easy to generate some new code.

30:44And I sat there for a little bit thinking that until I began to play more so. And maybe you, maybe you feel this way. Tell me if you do, but I kind of feel, no, I fully feel like the change that's happening right now is going to influence even more open source because more people will develop more things that are uniquely valuable to them. They're uniquely valuable to the workflows they decide to use. And I feel like we're going to have a major, major boom in a proliferation of more open source out there, more developers, you know, more people using software, more developers contributing more software.

31:20I was worried for a bit there. How do you feel? Do you feel like it's, are you worried about less open source or you think it's for sure more open source being produced? I think it's definitely leading more towards more open source. And I'll give you my theory behind that. So if you think about the pendulum swung too hard towards LLMs to begin with. Oh, 400 billion parameter model, 500 billion, a trillion parameter model. And those are the ones that we really need. Those are all closed source models. You should worry only about the performance part of it and we are good with that. But now we are seeing kind of pendulum kind of coming back.

31:59You know, those LLMs are great, you know, at all the things that needs to be done in the world. But in order to solve my problem, I don't need 99 % of that functionality. I need like a small language model that does my thing well and really well. So if you think about that pivot is coming down back from LLMs to SLMs, small language models, 3, 5, 7, 15 billion parameter model that can very well run on my CPU. And that's exactly where the innovation is happening. And if you think about, pick a vendor, Microsoft, Google, Amazon is very, very service centric, but pick any other vendor, essentially, they have an open source model.

32:45and those open source models are a lot smaller, 5 to 7 billion parameter model. Heck, even OpenAI has a GPT-OSS 20 billion parameter model. Grok, XAI just did their Grok model. It's a large model, but it's open source. All of these large commercial entities are realizing that open source is the way by which you can bring the largest developer mindshare. As quirky as Grok is, but now open sourcing it, you understand the quirkiness and then you can start participating and make it better. So I think in that sense, with AI, frankly, more open source is going to happen because it's not going to be less.

33:33Yes, people are going to be able to write more, generate more code. The thing that I worry about, though, is I'm not a coder, but I'm using wide coding tool. And before wide coding, I could have done maybe 400, 500 lines of code in a day. But with wide coding, now I can do 4 ,000 lines of code a day. What kind of technical debt that generates for future generation? So for example, I remember seeing a keynote from GitHub where they say, hey, we want to reach out to a billion developers. Sure, we want to reach out to a billion developers, but what does that mean? The classic example they said is, a grandma should be able to say on her phone, hey, take all my Instagram photos, organize them in certain order, store them somewhere.

34:23And she doesn't even say store, organize them in a certain order, et cetera, et cetera. Now, the idea was that is an NLU interface that you will parse that. You will, end of the day, you need software to organize all of that. So if you organize that source code and it's sitting in a GitHub repository, so the classic example is great where you get it work up and running for the first time. Who maintains that source code if things go wrong? And the models are not there. I mean, I'll give you an example. I was playing with the app. I've been working on an app actually and that app really allows you compare different protocols.

35:03So REST, gRPC, WebSocket, SSE, GraphQL. So there's a graph database and I make the query to the database and I compare them and I create a postman collection for it, for example. So I did all that. Now I created a GitHub Codespaces version of it where by single click you can have that environment deployed in your own environment. and then after a few days I realized oh you know what that code spaces environment is not working so I asked cursor that delete that code spaces button from the repo and the dude didn't understand it says oh you want me to delete the button so I'm gonna infer you want me to delete the dot code spaces and dot dev container definition from the repo it struck it out completely.

35:53I say, what the heck? I want you to delete the button only. Am I not explaining it? Are you not getting it? Are you answering it? So I think in that sense, the amount of frustration, the amount of technical debt it'll create, yes, it'll generate more source code, but the amount of debt it'll create is scary. I have a hard time disagreeing with that, especially with the latest round of releases. I think specifically chat GPT-5 has showed what I would call diminishing returns in large language model advancements. It seems like, you know, when Sam Alton posts a picture of the Death Star prior to the launch, we're expecting something big and it falls flat.

36:34It's like marginally better, whatever. Like they, they, they fix a few sycophanty things, which apparently people liked. They were mad that you took away my sycophant because I want more compliments, but it just wasn't much. And so that is leading me towards what I've been saying for a long time is like, what happens if we plateau in these models and the code that we're getting right now only gets marginally better over the next two to five years? There's going to be a lot of not so great code being produced. And a lot of people that are producing it and publishing it online are people who don't know what good code looks like.

37:12That's dangerous, right? It's dangerous. I was listening to the A16Z podcast this morning, and Martin Casado and all these folks, I think Aaron Levy from Box, they were talking about exactly that concept that OpenAI has been talking about AGI. And they were asking, what number do you put to it? When will AGI be possible? And GPT-3 and 4, we saw massive improvements in terms of significant improvement in terms of the LLMs. but GPT-5 now is only marginal improvement if at all we call it that way. Some folks have actually even expressed frustration with the users of GPT-5 in that sense. So I think if AGI was a thing to be delivered by 2027, this is not the kind of progress that we want to see over there.

37:58We want to see more aggressive progression to be able to deliver AGI by 2027. So I think in that sense, what does even AGI look like? Is it a marketing terminology? Is it washing, you know, AI washing your product? You know, I've heard Danilo and Dario Amidai talking about, yeah, EGI is a thing. You know, we should be getting ready for that in the next couple of years. But are the models really operating at that level? Because this cursor dude, which is using Cloud4 as a backend, deleting my dev spaces directory, I'm not giving control to you at all. Right. Yeah. You got to turn the auto button off.

38:33That's for sure. When you request that change, don't auto that one. definitely approve that one by manual submission. It's such a mixed bag because sometimes they do it so right and you're amazed. And then other times they're just like, are you a child? You know, obviously there's no cognition there. So they're just, but the way that I think about it, it's like, seriously, dude, you just deleted the entire Codespaces file? Like just stupid. That's just stupid. Well, and I have to sometimes, like one of the most frustrating things that I don't like about it is, he said okay you can commit to the repo but i don't want you to push to the repo i will let you know when to push because if other developers are using it i don't want them to have a half-baked experience and again and again and again i have to remind cursor please do not push without my explicit concept okay i got it and then a couple of hours later is pushing back to the repo back to how i do things right yeah i said dude are you listening to me at all it's kind of ironic because all these years computers just do exactly what we tell them you know like they're deterministic they they're calculators they just calculate things and they do exactly as they're told and we've always wanted them to be able to do more and to intuit and do all the stuff that humans can do and now we've kind of gotten a little bit of taste of that but it's also brought in all of our flaws of just like just not doing what you're told like we're used to people not doing what they're told but we can rely on computers to at least do it if i tell you to do the wrong thing you're going to go do the wrong thing and that's my fault but at least you did as you're told now it's just like right maybe maybe i'll do what i'm told and it's like oh gosh what have we created you're right it makes you wonder that see because i was reading an article and they were saying it's not about prompt engineer prompt engineering these days it's about context engineering So it made me wonder that when I give that command to Cursor that, hey, you know what?

40:32Delete that button. Should I have said that do not delete the directory? I did not. Understood that you will do that for me automatically. So I think in that sense, it does make you wonder how much of context should I give it? And even if I'm telling you not to push to GitHub repo, what made you change your mind that you're still doing it? And then to talk about psychofancy, he says, oh, I apologize. I should not have done this because you asked me to do that earlier and follow the command. So don't ask me again and again. So I think in that sense, there is definitely immaturity in terms of what my son was, my younger son was building an application and he was deploying to AWS.

41:16And all of a sudden, the entire DynamoDB database was nuked by cursor. And he was like, what happened? I said, well, I'm glad we set up the whole daily backup. So you have a backup you can restore from that. So things like that, very, very mature right now. You know, you gave us a laundry list of the ones you're using. You mentioned Cloud Code. You mentioned some of the obvious ones. Cursor, you mentioned Cursor, but you haven't gone back to other ones you're using. I'm a fan of Augment Code, one of our sponsors as well, but big, big fan of their tool. I think it's arguably one of the better tools to use out there.

41:55I like AMP code a lot as well, but I haven't had this challenge where it randomly deletes things. And I think it's because you mentioned context engineering and Jared, you may remember this way back. I think way back is like in this year. So forgive me by saying way back, but just a few months ago, I said, you know what it is? It's document driven development. and this spec-driven development. But I was like, I like DDD better. It sounds cooler. And so I was like, document-driven development is the way to go. You have to document what you're going to build and then give it that clarity is the spec, is what you've designed.

42:35And our friends in the Python language community, they have PEPs. I think it stands for Project Enhancement Proposals, I believe, or is it Python Enhancement Proposals? Python Enhancement Proposals, yeah. So this idea of PEPs. And so what I've done in my work playing with this is I took that model of PEPs and I've borrowed it and replaced it. And it's project enhancement proposals. And so I take I've taken the document driven development as a as an operating system. And I've implemented something I'm calling agent flow. And the way all this flows together is a way for these agents to draft PEPs.

43:18implement that work successfully, document what I'm calling knowledge-based articles or KBs, bugs maybe even, or even updating back to that PEP, or even builder logs, which are like stories of what they do to build. And this entire workflow, I'm calling agent flow. It's groundbreaking, I think. I think this context engineering is the way to go. PEP engineering, however you want to call it, but document-driven development, giving them a full spec of a feature and not like build the whole app but more like make this thing better in these ways and it's very clear there's phases in there there's whatever and it's doing all that work all i'm telling it is the rough idea and all i'm playing with really is like little cli tools like a granola cli uh don't don't get upset granola i uh reverse engineered not me but somebody else uh the agent reverse engineered the unofficial granola api just so i can extract markdown files from granola granola ai it's so cool but this idea of like document-driven development and agent flow to me has been just uh impressive and full of results like positive good results that's pretty awesome yeah i think my flow has been similar.

44:36For example, I don't do discussion in Cursor. All of my discussion is with ChatGPT. So, hey, I'm formulating an idea because I don't know at what point of time Cursor will trigger. Oh, let me start implementing the repo. No, don't implement the repo right now. I'll let you know when I need to implement the repo. Stop. Don't do anything. Yeah, exactly. Right? So what I do is I discuss with ChatGPT that here's what I would like to do. What about this? What about this? What about this? so I craft my prompt from ChatGPT because ChatGPT can maintain the session. So as I'm doing back and forth, then I say, all right, now craft a prompt.

45:11Now you understand my requirement. I think I've communicated clearly. So give me the requirement. So I read it. I make sure that everything is documented correctly. Then I say, make this as a prompt for cursor. Then I give that prompt to cursor and then it says, boom, now you go generate the first shot of the repo. And usually I commit the first set of the repo and then I start tweaking it for developer experience, whatever that needs to be done. Then the other tool that might be worth looking at as you're talking about, Adam, is Amazon Kero that was launched a few weeks ago. And they are doing exactly what you're saying about spec-driven development or document-driven development where you can define the spec.

45:50I've heard good comments about it. I was not particularly impressed by myself. I didn't see much value of it as such, but maybe I haven't explored it enough, but that is definitely worth a look, Amazon Keto. Yeah, interesting. Yeah, I think we have a show coming up with somebody from the Kero team. There was a great post on X a couple days ago about this topic, and I found it funny. I'll share it here. Programming has quietly turned into a practice of making micro wishes to a genie. The art is making the wishes in the right order in just the right way to eventually get what you want. Right. And I think that's been, that's approximately true.

46:30Now, the quote tweet, as we used to call them back in the Twitter days, is even funnier. Programmers don't realize that this is exactly what their PM's relationship is to them. So we are now the PMs, you know. They used to be making micro wishes to their programmers trying to get what they want. And now we're doing it to our cloud code or our cursor. Yeah, I mean, I was reading about, like, I played a little bit with Replay. I played a little bit with Lovable, all of these different tools, just to kind of get an idea of the landscape that what are the tools that people are doing. I mean, and the more you play with them, I mean, if you think about, Lovable has reached$120 million ARR in seven months.

47:14Seven months, that's astounding. This$100 million ARR companies used to take three to five years, and at$100 million, they say, aha, I'm ready to go IPO. The landscape, the dynamics are very different. Cursor is almost a billion dollar ARR. Almost a billion dollar ARR. And I was listening to 20VC podcast. They were talking in that podcast that by end of next year, it'll potentially be$4 billion. But then this morning, they were also talking about on A16Z or I think different podcasts about the AI bubble. because what is cursor? Tomorrow, if say Anthropic says, we're not going to let you use Claude, cursor is going to fall flat because they're really relying upon the back end of the Claude.

48:08And all of these companies, Anthropic, OpenAI, they're all burning money. They're not cash flow positive yet. Yes, their valuation is$60 billion or$500 billion, whatever that number is, but they're all burning cash. So they're burning more GPUs today, so they're not earning money. And if Cursor is just like a VS Code, nicer interface on top of Cloud, why would Anthropic not build deeper features with Cloud Code? So it's TVD that Cursor continues to stay the market leader versus Cloud Code kind of picking it up. So I think all of this landscape is very dynamic. And then on the parallel side, you start seeing tools like Klein or RooCode.

48:51These are open source wide coding tools. These are becoming popular as well. If you look at Klein, there are about 2.5 million developers using it. If you look at RueCode, there are about 20 ,000 stars to the repo. And they brag about how you can use, they're not tied to a backend model. You can choose whatever model. You can even choose Open Router. So I played with this. I configured, I believe in RueCode, I configured Open Router. and in$5, literally saying, what,$4.53, I could create a full-blown app that could talk to my Strava and get me some details. I think you can actually start quantifying dollars and the lines of code that are being returned and you can start justifying these tools more and more.

49:37Yeah, five bucks get an app, kind of cool. I think that's kind of what I mean. Like, I think, I don't know about the open source side of that, but I definitely think there'd be more code. I wonder if the world will be invited to somehow solve little problems in their house. But I really feel like I'm going to eat my words on this one because, dig this, this is sort of a side tangent, but I was talking to my wife as you would when you're driving. And I think I was saying something like, I had a conversation with somebody and I said, I compare Windows and Linux. And she said, that person has no idea what Linux is.

50:14I'm like, they have to know what Linux is. And I was comparing them. I was just saying like Windows versus Linux in this one case. I'm like, and by the way, this is Linux. You know, like this thing is Linux. I was doing a comparable. I'll save you the story in the back story on that front. But she's like, they have no idea what Linux is. I'm like, babe, don't you think half the, like most of the world knows that the internet, the most of what we do in the world runs on Linux? She's like, no, absolutely not. And I'm like, but here's me, the naive, hopeful one. I'm like, no, that's gotta be true.

50:44Like, it's so true for me. It has to be that true for so many other people. She goes on Facebook and she posts the thing and says, she essentially re-asks my question. And there's so many hilarious comments. And like the internet has an operating system. Like you wouldn't believe the things, all that to come back to say five bucks, get an app. That seems kind of cool. I want that for the world. I want to empower people to create software. I think more so in like home labs or in home spaces where they can solve their own problems. but I don't know. I don't know if that's really gonna be a thing.

51:17It might be a while till that gap closes. I want that too. And I've been on this personal software kick. You know this, Adam, I've been talking about, you know, a home cooked meals and like single use apps. And now we can just vibe code up our own little solution. It doesn't have to get published. It doesn't have to be for anybody but us. And I want that for everybody because you know the power of like just scratching your own itch and then moving on in life. Like that is so awesome. If you do it 10 times in a week, now all of a sudden you're saving all this time and effort. But then I saw somebody say the other day, which struck me as true, they said everyone's going to have their own vibe-coded app just like everyone's going to have their own 3D printer in their house.

51:55Because that never actually happened. Like 3D printing was supposed to be this revolutionary new technology. There'd be one in every house. You're going to print your own furniture, your own this, your fixes. It's like, it's cool. And a lot of people make really amazing things with it, but it never actually permeated. not yet at least mass market like not i don't even know of all the 150 people in my personal network that everybody has according to that one guy i don't like two people that have a 3d printer you know and they're super i just feel like this might be like that yeah is that there's they're super nerds or they're just tinkering or they're creating like they're creative types no offense but like key chains and like trinkets yeah that's fun cool stuff or like maybe they fix a a shelf or something but yeah it hasn't actually done what everyone said it was going to do and because it's technically complicated it's fault there's a lot of faults you can run into we have a 3d printer we bought a cheap one should have bought an expensive one because it's just been a headache and so it just sits in the closet you know it's like and i feel like vibe coded apps are going to be kind of like that where it's like yeah now those are easier to get going than a 3d printer, but maybe on a similar trajectory.

53:05I don't know. Well, and I think to add on to that, you and I live this world on a regular basis, but can I ask my, my wife is a tech, you know, she's really awesome. She is a director of technical programs at NetApp. So she's very technical in that sense, but she's not a programmer. She has a master's in computer science and undergrad in computer science as well, but she left programming a long time ago. But can I ask her that, hey, you know, why don't you try wipe coding? I say, wait, what? What am I doing here? Why do I need to do? So I think that's one friction point, frankly. And the second friction point is, okay, so I can't ask her, okay, download cursor and give it an idea on what you want to build.

53:46So you're right in that sense. You know, it's like, yeah, sure, I will play with it for 15 minutes because if you want me to, that's not something that I'm going to build an excitement of and carry on forward. You and I kind of live this thing on a daily basis. and my GitHub profile is super green because I'm chunking out lots of code, kind of giving it the experience that I want and tweaking the code accordingly. But that's our life. That's not everybody's life. So I agree in that sense. And I worry about it, that lots of code is going to be generated, but the amount of technical that it creates, think about five years out from now, how many repos are going to be dead because they were created five years ago and never maintained, and now CVEs, vulnerabilities, all of a sudden are skyrocketing.

54:32That's the case right now, even. Like you look up, I did this recently. I looked up DNS servers on GitHub. Like what's out there in open source that is like DNS resolver related? Because I told you this, Jerry, recently, I replaced PyHall accidentally, but it's been kind of fun. I'll share more another time. But there's so many results and there's some experiments. There's so many dead repos out there now before VibeCoding, before ChatGPT really helped everyone leg up and leverage AI to generate so many lines of code in a day from 400 to 4 ,000, as you had said before. But I think, yeah, it's going to get even more.

55:15I wonder, though, like technical debt, like if it's a random application that I open sourced on GitHub or I published to GitHub because that will become the thing. It won't really be like I open sourced it. They'll be like, well, I need to use GitHub because GitHub is essential in the, you know, the build stage and the CIS stage to get to production or whatever production is or live on the Internet. They may even say, you know, like, how do we get to technical debt if I'm just building one off things that sort of matter to me? It has to be adopted and absorbed and, you know, communities have to surround around it and stuff like that to become technical debt.

55:51How would it how do you think technical debt would result? So it's not about your personal apps that are going to cause technical debt, right? Now imagine if I'm working on Kubernetes, right? I take a look at the issue. I suck up the entire core base in cursor. And I say, I want to resolve this issue. And it generates a whole bunch of code, which I have no idea of how the code looks like. And I said, yeah, send a pull request. So that's the challenge. because now Kubernetes maintainers are required, if they want to approve the merge request, they are required to review that code, which probably is not going to be per the standard and all those things because you're not giving it enough context and it doesn't understand all of that.

56:35And so those are the places where it's perfectly fine to create your hobby apps, but it's going to 10x your improvement for sure. And I've been enjoying it. But is those projects where projects like PyTorch or Kubernetes or OpenJDK, where I start injecting this white-coded code and that causes a problem then. I think maintainers are going to be all over this. They already are. So just yesterday in ChangeLog News, I covered Mitchell Hashimoto's recent decision on Ghosty to require disclosure of all AI tooling contributions in every poll request for it to be considered in Ghosty. So if you submit a PR and you used AI tooling and you don't disclose that you did, It's like an insta-close, basically, and I'm sure Mitchell knows what that looks like.

57:23He must. Because he's just fed up with it, and he thinks that it's common courtesy now to say, hey, this was written 90 % by cloud code. At least then your maintainers know what they should expect as they go into code review. And so that's just one step that one guy has made, and I'm sure other maintainers will follow suit. And hopefully GitHub actually creates some sort of processes around this. I know there's the co-authors line in Git where you could like cosign that this was written by Jared plus cursor or whatever it is. But if we formalize these things, then we can have more clear, what do you call it, disclosure, I guess, of who actually did the coding.

58:07And that will help us to avoid some of this stuff, I think. because some people are actually submitting stuff that adds the feature they want but they don't know how it adds the feature they want right they're trying to be helpful and they want a feature and so they submit a pull request written not by them but it works and so it's like okay yeah that's where i think it comes back to it's still just too hard it's just still too hard to to be a developer or to develop i i guess debtless software like that's yeah the debt free scream the old dave ramsey debt free scream stuff to get you with software yeah it's it's being a developer or making software however you want to frame what that role is these days as it changes i think it's still just too hard it even though lovable may get you there quicker it doesn't it doesn't like once you get a customer if that's a commercial facing application or anything beyond a toy, you face serious challenges at scale or even at not scale.

59:09Like your sale could be a hundred couple, could be a couple hundred people. It could be whatever your little thing is, whatever version of scale you're talking about. But it's still just too hard. There's still too much awareness of what a developer does from a terminal to production, even terminals, like I just said, like open sourcing versus publishing to get up. that's not, you know, the same to everyone or production or CI or all these different things that happen in the build stage. That's not common knowledge. And it's still just too hard. I think for most people and Arun, you mentioned your wife and how technical she is, but back to what I think you were saying was that she doesn't have the patience because that's not what she's trying to accomplish.

59:55She'll play with it as a tool because you mentioned to go and play with it and see what she thinks. But I think a lot of people will lack patience to put together what we as developers have had to put together for so long. And I wonder when will it actually get easier? I think it's easier to generate code, but it's not easier to generate good debtless code. I think that's really the key, really. How do you... I'll give a classic example, right? I was working on a simple GraphQL backend so it loads the data from a text file and it creates a GraphQL visualization and my son came back from office yesterday he was asking me dad what library are you using I said I don't know Cursor picked up a library so the point is that you are getting to that point because Cursor gives you that ability oh I'll pick up a good library it'll look good if you're caring only about the developer experience you don't know if that is the right library.

1:00:56You don't know if Cursor is really checking that does that library has enough star forks maintainability or is that Cursor's preference? So, I mean, as a developer, you go through that mental process that I'm going to pick chart3.js because that is the top-notch library. Again, it goes back to context engineering. If I'm not giving the context, Cursor will think what is right for it and then it'll move on and you have no knowledge of how this is being implemented. And if you have no knowledge, that is scary. Putting that code into production is scary. Yeah. These LLMs are like brute force. They have to give you a result.

1:01:32They're designed to give you some version of happiness. Correct. I even tell us sometimes, like, don't yes-man me, okay? Don't yes-person me. Give me critical thinking. Defend your decision or, you know, explain why mine sucks or whatever it might be. But don't yes-person me into the next phase because your job is to give me a result. You're absolutely right. I'm so sorry. I won't yes person you. Yeah. No, and that's one of the most annoying things. Touche. If you give like a suggestion to say cursor, which is using clawed back end, you say, hey, this doesn't look right. You are absolutely right.

1:02:09Let me go fix this. I couldn't even think of this. I know. Because they don't actually think. They just autocomplete. Well, and that's the point, right? We call it AI. It's very artificial and not intelligent. Yeah, well, we're good with marketing terms around these parts. I don't want to defend it deeply here, but I do want to say I think that that's one thing I like differently about Augment Code, honestly. They have this, I don't even know what it's called, some sort of context engine. It's got something in it where I haven't seen quite that same behavior, where the yes person, yes man, whatever you want to say is in there.

1:02:44It's, it seems to be, it retains things I've told it even after a few prompts or, you know, kind of like back and forth later. Like I've been very surprised at how much it retains of the, the core goal. And I don't know if it's the peps or if it's the agent flow or if it's the DDD or whatever it may be, but something in this flow that I've been doing doesn't give me that result. It's actually been quite uncanny. I've been surprised at how well it seems to retain the context. And it's without me prompting it for the context. It retains it in some way. So I don't know if it's what I've been doing, the way I've been doing it, or it's AugmentCodes.

1:03:24I do know they do the RAG method of retrieval and stuff like that, but this context engine, I think it might have some different smarts with the way it looks at your code base and it retains certain contexts. I'm not really sure, but I haven't had quite that same experience. It's been a bit more joyful. It's probably some traditional software engineering they're doing in order to provide the model with that context and everything that it needs. A lot of the differentiation right now between these tools is how much software you write around the model in order to really put its best foot forward in a continuing fashion.

1:04:01And we see that in different ways, be better or worse. And that's really kind of the different, because other than that, there's not much that differentiates these things today. It's really hard, I think, as a product designer to, especially with the chatbots especially, is like how's Claude's chatbot UI or buttons or whatever, any different than chat GPTs really. There's not much there. They're just like their own wrappers around a model, which is why so many people have made money just putting better wrappers around the same model and getting people to adopt them, but there's not much, there's no moat there.

1:04:37And that's why I'm just laser focused on the quality because it was so bad for a while that I was like, what are you people talking about? I'm talking about the code gen quality up until Claude 4, where as listeners of this podcast know, I've been genuinely impressed. It's the first time where I've been like, this code doesn't suck every time. I could use this. Maybe this is better than what I would have written. Like I'm actually impressed. that being said there's so many things that make it not 100 like where are we on the AIs can write all the software you know spectrum I don't know maybe we're like in like the 94 they can do 94 % of what you need to have but the last 6 % is like everything that that's really what I've been watching like can they make that jump because until we don't have to look at the code anymore we absolutely have to look at the code.

1:05:27And if we don't know how to, then we're just in idiocracy and we're in serious trouble, like Arun's talking about, like serious trouble. Well, just to be clear, I'm still looking at the code deeply. I'm still reading the code. I'm actually learning more about different software as a result of that, which is great because it's like, well, here's in my own mental model. I know what I'm trying to accomplish with this reverse engineered API or creating this archive with 7z, which is something I've been doing. And I wrote this bash script a while ago that I wrote personally, and then ChatGPT helped me make it better, and then recently Cloud Code wowed my world and just totally improved it.

1:06:02And I was like, whoa. It's really just wild how you still have to look at the code. I'm still looking at it, and I'm learning more because my context of what I'm doing is personal. You can learn something more than you actually care about the thing. Like when you go through a training and you do a demo application, it's a to-do app or something like that, something trivial or very common. And you're like, you know what? I've seen this a thousand times. Sure, I have some context for a to-do app. But in my case, I'm solving a real problem, which is being able to better archive large directories.

1:06:33And I know what I want because I'm the user. I'm the one who has the problem. I know how I want it to work. And so I can care more deeply about the result it gives me. And so I can follow the code and desire to follow the code even more so. So I'm actually following it and learning more than anything. This visibility of magically prompting is not what I'm doing. It's very calculated prompting. Right. And then I'm also looking at the code as well, not just prompt, get, wish, production, boom, work, see you, buy, go, make millions and millions or billions and billions. You know, it's different. Yeah.

1:07:06Yeah. Well, you are the domain expert. So in that sense, you know, you know what needs to be done and you're intimately familiar with what needs to be done. So in that sense, the learning is always more. And personally, and I think Jared, you were talking about this. when the thinking mode is turned on, when it's saying, putting out the prompts that, hey, this is what I'm doing. Oh, no, nevermind. This is what needs to happen. Oh, nevermind. You know, I made the wrong fix. Let me back it up. That's where my learning happens. So that kind of helps me refine, validate, update my thinking that this is how I should have gone about problem solving essentially, and then validating that, okay, this is a code that is generated kind of thing.

1:07:44Well, where do we go from here, Arun? Where do we go from here? So I think I'm looking at lots of opportunities so far I'm exploring and I'm thinking about where to go. I'm really enjoying being a free agent and all the learning that needs to happen. But I'm also sensitive of the part that, hey, we are end of August almost. September is a major hiring season. But then once we get into October, November, December timeframe, then the hiring slows down. And so I'm thinking about hopefully I will find something that I'm really excited about. And then I'll be able to jump on that ship, essentially, and be able to drive those initiatives.

1:08:31If anybody cares about growing your developer communities massively, you know, tapping into it, I'm available. I'm available. Talk to me. Shoot me a simple mail, arun.gupta at Gmail. You can shoot me a mail. You can see the work that I've done. Are you only looking for full-time stuff or would you also advise or do like a consulting or like you're interested in that kind of stuff? Anything is open game at this point, really. So anything is an open game. And as a matter of fact, I talked to a couple of companies more recently where they're looking for an advisor and a consulting role. So I'm talking to them as well.

1:09:04So I think because I'm a free agent, I have the ability to do all of that. But in the meanwhile, if something solid comes up, then we have to kind of evaluate the opportunities accordingly. So yeah, anything is game at this point, really. Is there any, considering your history, your work history has been, like you said, you haven't looked for a job in basically forever. So it's, how do you even do it? That's not my question. It was more of just an outside thought as you hear me. But I'm thinking of the pressure potentially on you to your next role is because you've been at this company for this long with this kind of motive and this kind of direction and it's very clear.

1:09:44Do you feel pressure to that your next choice has to be, you know, the way you've been doing your career? Does it have to be this multi-year, big, big ceremonious thing? Or can it be a bit more like I think most ICs have been in the last five years to 10 years, which is like one or two years here and there and more jumpy and more more temporal, I would say. Is there a lot of pressure on your next choice? Not really. You know, I am, again, trying to keep a very open growth mindset. Depending upon what the opportunity comes up, I have not said no to any of the opportunities. As I said, you know, there was a two-person company that reached out to me to be the founding CEO, for example.

1:10:28So I'm exploring those. They're a very, very rough idea. They barely started working on this for the last couple of months. But exploring, it's like, oh, that's an area that I'm not super excited about. And I don't have a lot of core competency in that. CEO role sounds very exciting, but I need to really believe in the mission and the delivery of the mission and that sort of stuff. So to me, really, that alignment with my value, as I said, developer, open source and AI, those are the three core components that I'm looking at it. And of which developer and AI are sort of the more core components.

1:11:01Open source is naturally going to be part of it. But anything that aligns in that, whether it's an executive role, whether it's an IC role, whether it's an advisory role, whether it's a consulting role, I'm, again, keeping very open mindset here. Not trying to box myself, not trying to be encumbered by the past that, okay, my next role, I must be a VP as well. I mean, I was a director at Red Hat. From there, I went to VP at Couchbase, much smaller company. But then from there, I went to IC at Amazon. And from there, I went to manager at Apple and then VP at Intel. So I'm totally fine going back, being an IC at a bigger company because I could be very valuable in terms of defining that strategy where the execution is on somebody else.

1:11:59but I'm equally capable of rolling my sleeves up to help with the execution part of it. So if you have a large team, for example, that needs help, you know, not just the strategy part of it, but people who want to help getting started. So I can certainly do that. So I think, again, keeping options very open, my opportunities very open. Those are the things that I'm really good at. And that's what I'm looking at. I never really considered the fact that it's like, you know, when there's a hiring season, And similar to you, I just haven't had to look for a job, fortunately, in so long. So I don't know that there's cycles and hiring flows.

1:12:35I imagine there's a lot of folks who have had change, maybe even fellow colleagues from Intel in particular. And since, you know, when you made your list, you said devs, that was people, open source, that's software. And then AI, that's maybe the thing we're using to get to the software and the stuff. so speaking to the devs that may be in this transitional period in their life some of them may be still in the in the grief part of it or the denial part of it or just some spectrum of where you've been you've you're kind of come out the other side what is a strategy for someone looking at the remainder of this year in terms of i'm not hired i need to be hired how can i get hired and the ticking time clock that may be happening because i never considered that september when it happens in October, November, December, it sort of chills out and diminishes until maybe the new year.

1:13:26Yeah, I think I'm going to go back to the phrase that I made earlier, you know, action absorbs anxiety. So I would recommend, you know, because when you are in a job, then you don't need to, you need to prove yourself in the job itself, but you don't need to demonstrate your profile externally. Now, a few things that needs to be done, make sure you get your LinkedIn audit done very clearly because that's how recruiters reach out to you. Make sure your core competency is called out. Make sure your photo shoot, the photo on the top is a professional photo shoot. Make sure your about section is clearly called out.

1:14:01Make sure your work history is clearly called out. Not just the years, but the actual skills that you did, the exact work that you did. Have chat, GPT, cloud, whatever, review LinkedIn, your LinkedIn profile, edit it, and make sure you spend time on that. So that's one crisp part of it. The second part of it is you also need to kind of build that profile externally. So for the last several weeks, I've been blogging twice a week. Usually one article is about thought leadership and the one article is about technical leadership. So, for example, last week I talked about that thought leadership article on how to grow from zero to 100 million developers.

1:14:40And after this podcast, I'm going to record a video about the lab, about the app that I was talking about. I'm going to record that video and announce that GitHub repo essentially. And then I have another thought leadership article lined up for later this week. You know, basically, if you are running a dev rel, what your metrics should be. And that's the discussion that I've been having with the Developer Relations Foundation, DRF folks in LF. So I think in that sense, I'm just sharing sort of my blueprint. and frankly that keeps me sane that again solving that easiest problem then the next and then the next it allows me to show me progress and the worriness could differ based upon the stage of your life right you know i'm at that stage of the career where i'm not worried about that oh i do not have a shelter and things like that.

1:15:33Or my elder son is already working, so he's on his own already. Younger son is a senior in high school, so he's about to go to college. There is no financial constraints per se. So in that sense, I don't have a rush that even if I don't get something solid in the next few weeks, months as well, I'm not in a rush. No, I'll do a consulting gig and as an athlete i'm very comfortable with being uncomfortable physically mentally things like that so that's something you're going to kind of keep in your head but again build your external brand so that that makes you a more attractive talent everybody has access to chat gpd cloud etc but so everybody's writing a very impressive cover letter a very impressive resume what helps you stand out is networking with people.

1:16:24So that's sort of where I've been spending having a structure to your day that for the first couple of hours, I'm going to scram through LinkedIn, comment on people's articles so that you build that, reestablish that relationship. On LinkedIn, there's a simple thing that you can say, who viewed my profile? So start engaging with those people. If they're viewing your profile, if there is an interest, maybe look at what their posts have been, start commenting on those. So some simple tips, but having that structure for the first couple of hours, I'm going to scan through LinkedIn for the next couple of hours.

1:16:54I'm going to refine my resume, whatever LinkedIn profile needs to be updated. And then afternoon is going to be really digging deep into a technology and then blogging about it. So I think if you have that kind of a structure, because otherwise if you are sitting empty, all sorts of weird thoughts come into your head that at least to me, it gives more peace and structure. Yeah. Idle hands kind of thing. Yeah. Get action. take some action y 'all I like that I like that action absorbs anxiety good job and this is not my quote I heard it from the podcast I like the idea it's solid it's a solid idea I do like that a lot what else is left on the set what else have we not asked you about I know we wanted to I reached out to you on LinkedIn a friend I was like oh my gosh how you doing we should talk You're like, let's pod.

1:17:49I'm like, of course, let's get you on the podcast. What else is left unsaid that we haven't covered that you might want to cover? No, I think we have covered most of the things here. I think my advice to people is be patient to yourself. Be self-compassionate. You know, oftentimes when you apply for a role, it's a multi-week cycle. Recruiters don't respond back. Sometimes you don't even get to a recruiter. you may be submitting application few applications every day, every week sometimes you don't get respond don't let that define you don't let that bring you down just keep chugging at it in a very relentless manner keep chugging at it each one of you that is looking for a job at this point of time has done something wonderful just focus on that part of it, don't let the negative energy come around you at all just pull the cord on it right away because if you're not going to believe in yourself, that's going to show up in your conversation.

1:18:52If you believe in yourself, if you have the confidence that, no, I've done this, yeah, I've done this at scale, and this is how I've done this at scale, imagine what that interview conversation might look like. Mock it. ChatGPT can help you mock those interviews. So I think start getting yourself ready so when the moment comes, then you are ready. Yeah, so I think that's the key part. I would comment but otherwise yeah I mean it's a road ahead the other statement that I love from Bible I'm not a Bible person but it says I really believe in it just because I don't see it doesn't mean the road has not been carved for me yet yeah and the moment you think about that oh you know what I don't know where am I going to go it's not a straight road ever so So hang in there, have faith, have belief in God or whatever you believe in.

1:19:50There is a road carved out for you already. You just don't sue it. Yeah. You know, something that you said there, if I can just add one to what you're saying, I think that's such a good idea to like mock up an interview. One thing I love a lot about AI is the ability to sandbox, I guess, iteration, like try things, like have it do lots of scenarios. And I don't know how it does it, but that's a great example of like, you can mock up like, hey, give me tough questions I will potentially have if I go for this interview, like how and practice with it. That's that to me is a pretty interesting thing.

1:20:27I didn't think about that because you were suggesting like how to clean up your LinkedIn profile, which is great, but to actually train, train with it. I think that's kind of cool. Yeah, because the common saying is I'm not worried about 10 ,000 things, you know. I'm worried about one thing that you have done 10 ,000 times because you're going to be so good at it. So train yourself well. Just figure out where do you fail? What are your weakness? What is your touch points? What triggers you? And how do you keep control under pressure? So ask those tough questions. Let ChatGPT help you. Let Claude help you.

1:21:02Ask those tough questions and think in your mind. You know, just prepare that pitch. You know, have those examples ready. that in the previous work, on in three years ago, this is what I did. So have those scenarios ready, practice it. That goes a long way. And really, that's the only way to go forward for me. Good deal. It's been good, Arun. Thank you for coming back on the pod. Good seeing you. Well, thank you very much for having me. I really enjoyed this discussion and very fulfilling. Thank you. ChangeLog++ members are in for a treat this week because after the official show, I asked Arun for his honest take on the U.S.

1:21:41government buying a 10 % stake on Intel, and he started with this. Lots of votes. Yes. That's coming right up unless you aren't in on it. Now's the time. Directly support our work. Ditch the ads. Hear what Arun thinks about Intel now that he doesn't work there anymore, and get other awesome bonuses. Sign up today at changelog.com slash plus plus because it's better. It is better. It's been better for years. Thanks again to our partners at Fly.io and to our sponsors of this episode, depot.dev and authzero.com slash AI. Next week on the pod, news on Monday, Jim Mesnick from XORuby on Wednesday, and Christian Rocha from Charm CLI on Friday.

1:22:24Have yourself a great weekend. Share the show with a friend or three who might dig it. And let's talk again real soon.

1:22:51Game on.