Algorithms + Data Structures = Programs - Episode 276: Mini AI / Cursor Update (+ Running)

Episode Date: March 6, 2026

In this episode, Conor and Bryce give a mini AI & Cursor update, as well as chat about running and more.Link to Episode 276 on WebsiteDiscuss this episode, leave a comment, or ask a question (on GitHub)SocialsADSP: The Podcast: TwitterConor Hoekstra: LinkTree / BioBryce Adelstein Lelbach: TwitterShow NotesDate Recorded: 2026-02-26Date Released: 2026-03-06Cursor NVIDIA ArticleIntro Song InfoMiss You by Sarah Jansen https://soundcloud.com/sarahjansenmusicCreative Commons — Attribution 3.0 Unported — CC BY 3.0Free Download / Stream: http://bit.ly/l-miss-youMusic promoted by Audio Library https://youtu.be/iYYxnasvfx8

I work more now because it is intoxicating what you were able to do with this stuff, which is kind of like, I've heard on podcast being like, wasn't this supposed to free us from work? And it's like, this is not work. I am Harry Potter with the elder, the elder one. Welcome to ADSP the podcast episode 276 recorded on February 26, 2006. My name is Connor. And today with my co-host, Bryce, we give a mini update on our AI and cursor usage, as well as chatabot running and more. Oh, I bought running shoes, by the way.

I did buy running shoes. It is better. You were correct that I needed to own running shoes that you're going to run. This guy saying his knees hurt while he's running or whatever. And I'm like, well, what kind of running shoes are you running? He's like, oh, do I need running shoes? I'm just like, yes, you need running. I guess maybe I'm being, what's the word, elitist about running knowledge.

Maybe it's not a well-known fact. If you are having knee troubles, shin troubles, and you're trying to get into running, especially if you are above the age, we'll just call it 30. I'm not saying anything about people over 30, just when you're young, it doesn't really matter. The older you get, the more important it is you need a decent pair of riding shoes. The more cushion, the better for protecting joints and stuff. And if you've had a pair for like four years, you probably need a new pair. I get a new pair every like 600 kilometers.

I won't say how frequently that is, but it's pretty frequent because you want to be able to exercise and walk for a long time. We may not be thinking about it in our 30s. But at some point when we get to 60s, 70s, 80s, you're going to start to have joint problems. And the better the shoes, the longer your joints are going to work for. So we did not end up doing the 5K on Valentine's Day because Ramona was very sick and it was very cold. And I could not make the sale ongoing through with it. But we are going to sign up for some other 5K. But I stopped running for a little bit after the first snowstorm in New York.

and it's been hard for me to get back into it. Part of the problem is it's been too cold for Ramona to come running with me. And I am not as motivated to run when my girlfriend is not there to embarrass me by running a lot faster than me. So I keep telling her, you need to come back and start running with me again because it will be better for me if, like, you are there to force me to keep going. You just got to run enough until you get addicted to it. I have a race on Sunday called the Chili Half Marathon, which I've run twice. before and it's usually been right above freezing you know zero degrees for Canadians 32 for Americans I should say Canadians and the rest of the world and this Sunday it's

supposed to be minus 12 feels like minus 18 which is got to be like potentially below zero for Fahrenheit folks and I'm very unexcited about it but you sign up for it you got to do it only one time in my race have I ever missed a race and it's because I was definitely ill because my nephew Franklin gave me some virus which made it impossible to leave bed for like a couple days. One of the big factors in us canceling was a misperception about the timing that I only realized at the last minute, which is Ramona's been taking these acting and then these speech classes and she has one on Saturday.

And we signed up for the race and they said like, the starting time was like at 8 a.m. And I'm like, that's great. Like 8 a.m. 30, like a 5K will take like, you know, like 30. minutes, you know, tops for us. So then we'll be done and shall have time to get to our class at noon. And then I realize that they probably don't have all the people start at the same time. And we, the like very inexperienced runners are probably not starting in the first group, which is probably the like professional or very fast runners who have run a lot before. So we probably would not, it would like not have been possible for us to have done that anyways,

probably. Yeah, that makes sense. Well, I look forward to your future race, your future times. and I'm glad you got running shoes. I have a question for you. Hey, technical question, not related to our other stuff we have going on, which is, have you started playing around

with Claude Code at all yet? No. Although we can chat about this, I'm going to cut it into the next episode, folks, because there's a lot of, so here's the preview, Bryce, because I don't have time to record

for a couple hours today. Bryce and I are going to record, sometime next week and that's where we got a lot of stuff. We got to talk about parrot. We got to talk about a Raybox. I've been dying to talk about it. And also in general, we got to talk about the state. AI's back folks.

I mean, it never left. But it's, is it back? Sure, we'll call it. We'll say it's back. The topic of it is back on the podcast. It's better than ever. Yeah, I asked this question less as like a podcast question to lead into a discussion. More is like, I just need an answer to the question.

Now, so the answer is no. I have not tried Claude Code yet. still I'm still cursor-pilled. So, but with cursor, the thing I'm increasingly finding with cursor is it's been a force multiplier for me, but I think it could be a lot more of a force multiplier if I didn't have to approve the commands. And so at Nvidia, they said, I found it an email. And also, too, we can actually talk about this.

I mean, so there was an article published by cursor that said 30,000 people at Vindivida are daily active users of Cursor. But I was, I hit my peak rating of 168, which put me in the, I don't know, whatever, top percentage. And I've been really loving Cursor. And that is the main thing, though, because it's a corporate, you know, version. It has a lot of safety restrictions and a lot of that it requires us to, you know, basically approve any shell command.

But they changed it recently so that you can now do allow all. So they changed it back in January saying that they were. rolling out an auto, you know, approve for a list of things, I have not been able to get that to work. So I thought that there were three modes. I thought the modes were, I thought the modes in our version were that there was approve every command. There was allow, there was like auto-approved with an allow list where you can enter in, like,

commands that are okay. And then there was like just always auto-approved like full YOLO mode. Yeah, we don't have the YOLO mode. I thought that we had it. No, I think you're right, actually. No, they're rolling out the allow list and the allow list is going to contain most of what you need over time. So here's my thing. I increasingly find, because like what I tell it to start and then, you know, I usually try to go in context to do some other task.

I have not yet set it up to be doing like multiple chats at a time or sorry. Cursor or clog code. With cursor. Sometimes with cursor. Cursor, it's just by default. You don't have to set anything up. No, no, no, but like, like, I would need to, like, set up, like, get work trees or something. If I want to have, like, if I want to do two tasks that might touch the same files,

and I spin that up in, like, two separate chats. Oh, you want, you want multiple agents working on different tasks, not multiple agents working on a single task to get you the best response as fast as possible. No, because, like, the responses that I do get are generally, like, good enough. Like, I don't need more power for each individual request. You're not pushing it. You're not pushing it hard enough, Bryce. I guess maybe I do need it.

I guess, no, maybe that's not true because if it did take, if I got better responses or if it took less time, that would accelerate me too. But it almost certainly the thing that's increasingly getting me is the like command approval. And part of it's because one of the things I've been doing a lot has been working on our like educational content and on like the basically the DevOps setup for that, which I think is a little bit more like command oriented. It's not just like purely writing code. It's like write some code and then like go and test something. So I really want to be able to give it a task and then like let it go and do its own thing. What I've heard is that Claude Code has some sort of sandboxing mode

where it will go and run its thing in a sandbox and like as a sandbox it doesn't have network access so it has like limited damage that it can do. I'm scared of the allow list, I guess. Because I think even with the allowless, I would feel uncomfortable, like, working on the production code base on a machine where, like, I could mess things up. I thought maybe about just, like, setting up a containerized environment and then in the containerized environment running a cursor where, like, it doesn't have access to anything that can touch the corporate net. But yeah, I was wondering if you tried it yet. I think I'm going to try to set it up like sometime next week because I've got a big backlog of stuff I've got to do. I think if I can have it go off and do its things on its own and then come back to me

with like a full PR I can look at, I think that'll be better. The other thing, and I'm going to have to show you this next week once I get it fully set up, but I've come to the determination that I think that it would be very challenging me for me to go work at anywhere else it's not in video because I'm increasingly starting to believe that Navidia has the best access to AI tools to use for our day-to-day work. Yeah, I think this was maybe not true three or six months ago, but that precursor blog post that went out, like I think that's indicative about Nvidia using cursor. I think that's indicative of like a change here.

And I think there's some pretty interesting tooling that we're like now dog fooding and playing around with. And like, it's funny to me, like, how now it's become like a job, like, a requirement for me in any job I would go to would be that I would have to be able to have access to, like, AI coding tools. And, you know, you and I, I think, have never really worried much about the budget for our token usage. And I think that that's in part because we're at a big company and at, specifically, in Vidia, where our business is the token generation. But I think that other people you know, who are not here who are paying for themselves for their tokens, you know, are a lot more constrained by that. Yeah. Well, we have to wrap this up. And this is technically already not airing till

next week. But yeah, I've got so much to say. Yes, I agree. I mean, I'm not going anywhere. I love working for Nvidia. Obviously, it's an amazing company to work for right now. If there are, if I were to leave, there's only two companies I would consider leaving to. One's Anthropic and one's any sphere, the company that makes cursor. For basically the same reasons that you said, like, I am now, I'm trying to completely re-architect the way that I think about everything. Yeah, me too. And this, we'll talk about this more like next week, but literally, and Shima is to think

for this, my wife, because I've been building all this stuff, like Array box, dashboards behind Array box. I have another website and like I'm writing a book that's built on top of Array box. I'm building a whole other platform built on top of array box. And then I have a whole other things that I'm going to build. And I was whining about Descript and these AI podcast editing tools. And I was just like they're all just garbage. And I was whining about it to her.

And she hears me like nonstop ad nauseum saying how amazing these tools are. And she sees what I'm doing with it. And she's just like, she understands why like I work more now because it is intoxicated. what you were able to do with this stuff, which is kind of like, I've heard on podcast being like, wasn't this supposed to free us from work? And it's like, this is not work.

I am, I am Harry Potter with the elder, the elder wand. I don't know. Anyways, there's some wand. It's like the ultimate wand. That's what it feels like. And I was wanting about this stuff.

And she's like, why don't you just go build your own podcast editor? Because like I want to be in the world. Like I spend so many hours a week on like content creation of just like removing ums, cleaning up audio tracks, like, you know, all this stuff.

and like AI will be able to do it all at one point. I still want creative control of when I insert like tracks from some, you know, a movie or song or whatever or some, you know, advertisement. And I was just like, well, it's a pretty tall task to like replace audacity, the features that I want from it. But then I was thinking and I was like, well, I do, why, why can't I do it? Five hours later, I have this like minimal thing. And now, like, I've realized that I have complaints about everything,

like the podcast players that I use. Like, like, and, Anyway, so I just realized, like, I have the ability to build the world, and I'm just going to go build it. Literally, it's just the time in the day that I need in order, and people like, why would you go create your own podcast player? I spend more time editing and listening to podcasts than any person I know, and I have a very, very good, like, intuition for, like, good design and, like, we always whine about. But most importantly, like, you spend a lot of time doing the editing, so why not automate it? Exactly. And the other thing is, even just consuming podcasts, like, there are so many complaints.

that I have and like, you know, super simple ideas. First of all, no, no ads at all in the podcast player. I don't want to make money off of this thing. I wanted this for myself, so it's a better experience. Second thing, so that's, it's no ads, but what if we could do better? What if I could use AI to skip, automatically skip the ads for me? In the daily, it's 30 seconds at the beginning, 60 seconds in the middle, sometimes, you know, 15 seconds at the end before the other things you need to know. In other podcasts, it's like, you know, eight minutes at the beginning of the podcast but 10 seconds. If this takes off, you're going to get canceled by all the podcast creators who are going to be like,

hey, you take it my revenue. You know what? We've got ad blockers on Chrome. And, you know, I'm already skipping the ads myself manually. Yeah. Just like, and that's the thing is like, I know. I know. That's why I know the daily has 30 seconds.

I add at the beginning, 60 in the middle. And like, I have my podcast player set up. You can customize like the skip ahead 30, skip back 10. Like, like, anyways. And also, too, like, there's just, in the podcast player that I use CastBox, it's so awful. It's got all these community features. It's got all these, like, top podcasts. And the thing is, I want to switch to Spotify.

They don't have a queue. What podcast player in their... Yeah, it's like, I use podcast addict and it's like, okay, because I'm on Android. And it's like, okay, but not... Yeah, like, a lot of the same problems I have to. None of them are good. And Shima was saying that even I was playing around with her Apple podcast because she's on iPhone. And she was like, oh, I've got so many complaints.

The same thing that happens on CastBox is like after a certain number of days, if you haven't listened to it, it like removes it from your device. They separate downloaded episodes versus the, anyway, the point being there's a whole episode we have to have on like, you know, just being like an idea guy. Like I have had so many ideas for so many years that I just didn't have the time or the knowledge to like go and implement. Like I would never go create my own podcast player like app. But now like you could probably do that in a day on a weekend like the idea. person thing is why I think I need to prioritize getting like the sandboxing set up because I've had this big, I was just going through my to do list and I had this big backlog of like things that like I know I know the scope of the project. I know exactly how to describe what needs to be

done. And if like if a draft of it is presented to me, I can I can direct it to the finish line. I just like, but I have like 20 of these things. I can't if the way that I'm using AI is that I'm like monitoring, like, you know, monitoring it doing each one of those 20 tasks serially, it's not going to work. But if I can go and tell it, like, do these 20 things, come back to me and then spend a day reviewing those 20 things. It's a world that reward may reward people with ADHD. But anyways, I should let you go.

Yeah. And I got to, it's so sad, though, though, and that's the thing. I was supposed to stop recording 20 minutes ago. And then it's just, I'm so excited to talk about this stuff. I've recorded two different clips from two different podcasts. and one of them is from the Peter Steinberger, the guy that did OpenClaw, formerly Moldclaw, formerly Claude, a bot. And, oh my God, it's just, he's just like, you get, one of the quotes is a good driver can get results from any of these models. And then there was, I didn't clip it, but there was a guy, the guy that is doing ghosty Hashimoto, you know, he's, he was saying that, like, he was talking about exactly what you're saying is that he's very specific about which models he uses, you know, haiku librarian.

one other one. So for the simple tasks, he delegates there. And I do the same. Anyways, there's so much stuff to talk about. We'll leave it there. Maybe I will release this episode next week and then you'll be hearing us. Well, maybe we're, anyway, I'll figure it out. It's too much stuff to talk about. You got one final thing to say. Nothing for the podcast. Well, okay, I guess it is for the podcast, which is we've, there's two guest ideas that we've talked about in the past and that have come up. And I have on my to-do list to, like, think about them. And it's a Chris Leary, who is one of the main guys He came up, yeah.

We've already talked about him. He's J.F's co-host of TLB. Do we want to reach out to him and see about scheduling any time to record with him? Not yet. Because we're going to record a lot next week or the week after, which is going to be like three or four episodes of content. Schedule it after later in the year.

That's true. That's true. And the other one would be... I mean, if this is going to be on the podcast, I guess he's potentially going to find out because someone who knows him might let him know. Yeah, the other would be to have Emily Berger on. Yeah, we could get him on as well. What happened to Carpathie? Have Carpathie on? Didn't reply to my emails. Come on, Carpathie. The path to Carpathie. Let's go. Because that's what Bryce and I were talking offline. Who knows how much

of this makes it in, but about like the guests that I want to talk to. Yeah. And honestly, I was like, you know, I used to be programming, Mr. Programming Languages, Mr. Algorithms. I still love that stuff, but like, let's be honest, programming languages don't matter anymore. Ideas matter. And actually, I was talking to Michael, my boss. And he was like, no, ideas are a dime in a dozen. Insights are what matters. Insights are ideas worth keeping. And I love that quote.

I love that quote. I'm going to put it on a T-shirt. Insights? All right. That's a very Michael Garland quote. Yeah. But like Ilya Susskever, Andrei Carpathie, who are the other people that I mentioned? There was four people and I couldn't figure out a fifth that I wanted to talk to.

You know, Dario Amadee, he's too CEO-ish. I don't want to talk to Sam Altman. I don't. Oh, yeah. Demis Sassavis. He's definitely top three. Anyways, there's a couple others. If we could get those people on, but they're just, they're two, we're not a big enough.

We're top five percent podcast, but we're not a top point one, you know? I think that, like, I think we ought to reach out to one or two people and see if we can schedule something for later in the air, because otherwise we'll just talk about guests and not actually pull them on. So here's a, what do you call it, ADSP, ADSP promise? Is that the next time Bryce and I record after the next time, we'll have a. guest. Yeah. Might not be any of the people we just mentioned. You might not even be hearing this right now.

Be sure to check these show notes, either in your podcast app or at ADSP thepodcast. com for links to anything we mentioned in today's episode, as well as a link to a get-up discussion where you can leave thoughts, comments, and questions. Thanks for listening. We hope you enjoyed and have a great day. Low quality, high quantity. That is the tagline of our podcast. It's not the tagline.

Our tagline is chaos with sprinkles of information.