transcript.txt

0:00So Andre Karpathi, he's the guy who literally helped build modern AI, who was at OpenAI, who got Tesla autopilot
0:077 secondsworking, and who coined the term vibe coding. He just told us if you're still building apps the way you were last year, he's got bad news for us. So he
0:1515 secondsjust gave a brilliant talk. I've spent hours breaking it down to help you understand how it may affect what you build as a developer, founder, or
0:2323 secondssoftware company. Thanks to monday.com for sponsoring this video. More on them later. In the next few minutes, I'm going to walk you through what he actually said about 2026, what it means
0:3232 secondsfor what you should be building and how you should be building, and the four frameworks that I think every AI builder needs to have in their head right now.
0:3939 secondsI'm Rob from Switch Dimension, and if you're trying to build with AI in 2026 and beyond, this might be the most important video you watch this month.
0:4646 secondsSo, stick with me. This kind of a brain is the single most powerful thing I've built in the last couple of months.
0:5151 secondsHighly recommend you do something similar yourself.
0:5757 secondsSo one of the first things that stood out to me was Karpathy's point about the December inflection. So what Karpathi is saying is round about December the

Andrej Karpathy
Andrej Karpathy is a Slovak-Canadian AI researcher, who co-founded and formerly worked at OpenAI, where he specialized in deep learning and computer vision. He served as the director of artificial intelligence and Autopilot Vision at Tesla. In 2024 he founded Eureka Labs, an AI education platform, whose first course will be LLM101n: Let's Build A Storyteller.
...more
1:061 minute, 6 secondsmodel's output started to actually just work. He stopped correcting it. He started just trusting the model and he basically went on this vibe coding
1:141 minute, 14 secondsbender. He said he has a ton of side projects that he's built up. And we all have that problem. Multiple different projects we're working on the same time
1:211 minute, 21 secondsbecause now we can build anything. So essentially if you haven't sat down in the last 60 days and seriously tried to build something end to end with uh claw
1:301 minute, 30 secondscodeex cursor in agent mode you are really flying blind according to Karpathy. So go do that this weekend.
1:371 minute, 37 secondsSeriously build something. So I think Karpathy's biggest takeaway in this talk was the idea of software 3.0 and we've
1:441 minute, 44 secondsheard that so many different times before but this is a slightly different angle I don't think many have seen. So Karpathy's breakdown of the software
1:521 minute, 52 secondsevolution is software 1.0 is handwritten rules. So basically writing out code.
1:581 minute, 58 secondsSoftware 2.0 was training neural networks via large data sets. And software 3.0 is where the large language model itself becomes the programmable
2:062 minutes, 6 secondscomputer. The interpreter and your code basically is the prompt and the context window is your lever. So that's Karpathy's highle take on it. How does this affect us practically as builders?

Andrej Karpathy
Andrej Karpathy is a Slovak-Canadian AI researcher, who co-founded and formerly worked at OpenAI, where he specialized in deep learning and computer vision. He served as the director of artificial intelligence and Autopilot Vision at Tesla. In 2024 he founded Eureka Labs, an AI education platform, whose first course will be LLM101n: Let's Build A Storyteller.
...more
2:162 minutes, 16 secondsSo, in traditional SAS, this is Levercast, an app I built. It uses AI to help you post to all your social media platforms. So, I would just click
2:242 minutes, 24 secondscreate, drop in my thoughts, generate a post with an existing style guide, and then it takes the pain out of the first
2:312 minutes, 31 secondsdraft for my social media posts across the various different channels that I work on. So, that's a traditional SAS app with a little bit of AI sprinkled in
2:382 minutes, 38 secondson top. So, here's the big shift and here's what's changing. We're going to be increasingly using our agents to do more work for us. We're going to be living in claw code, codec, cursor,
2:482 minutes, 48 secondswhatever is your agent of choice. In this new world, what people will increasingly be doing is basically just telling their AI agent to carry out a
2:562 minutes, 56 secondstask. So instead of me going and logging into Levercast and going through all the workflow, I might just say, "Hey, I've got an idea for a post. I could drop it
3:023 minutes, 2 secondsin and say, "Hey, use the Levercast MCP to carry this out." Or not even. I could actually use a skill that's pre-built in that will actually do all of this for
3:103 minutes, 10 secondsme. I go off to do something else or I have this working on a schedule and then bang, all my posts are published. So besides that being a cheap plug for my
3:193 minutes, 19 secondsown software, really what we're getting at here is a change in how we need to think about building things. As a builder, you really need to think about
3:263 minutes, 26 secondsthe current capabilities of multimodal models and even think ahead a few months to what's coming next and how that might affect the software you're building or
3:353 minutes, 35 secondsyou've already built. Do you need to have an MCP, an API, or some kind of agentic workflow in front of it so that an agent can discover it quickly? So,

Andrej Karpathy
Andrej Karpathy is a Slovak-Canadian AI researcher, who co-founded and formerly worked at OpenAI, where he specialized in deep learning and computer vision. He served as the director of artificial intelligence and Autopilot Vision at Tesla. In 2024 he founded Eureka Labs, an AI education platform, whose first course will be LLM101n: Let's Build A Storyteller.
...more
3:433 minutes, 43 secondsKarpathy's example of this was back last year, he built this quick app that basically allowed you to turn your menus
3:513 minutes, 51 secondsinto magic. It's the idea that you could take any menu and looking at this menu here, you might not know exactly what the food is or what it might look like.
3:593 minutes, 59 secondsBy dumping in this menu into his app, it basically was able to create AI representations of what the food was and what it looked like. So last year, this
4:074 minutes, 7 secondswas an entire application that he built that somebody could use. This year, he could solve the same problem with just chachi BT or Nano Banana. So you can see
4:164 minutes, 16 secondsthe entire menu gen workflow replaced by me just dropping in that original menu.
4:214 minutes, 21 secondsAnd here you can see an overlay of the dishes on the starters. So his big point here is the menu genen app that he built
4:294 minutes, 29 secondsa year ago doesn't really need to exist anymore. And here's why that's really uncomfortable. A huge percentage of the
4:374 minutes, 37 secondsapps people are building right now shouldn't exist either. They're basically orchestrating things the model
4:434 minutes, 43 secondscan already do natively. That's only just appeared in the last couple of months. They're software 1.0 plumbing wrapped around what should just be a single software 3.0 zero prompt. Anyway,
4:554 minutes, 55 secondsyou get the point. So, here's the test.
4:584 minutes, 58 secondsTake what you're building and ask, could I do this with a single multimodal prompt and the right tool calls or an MCP or two? If the answer is yes, you're
5:075 minutes, 7 secondsbuilding plumbing that's about to get eaten by the next model release, stop or pivot. For you Americans or Canadians who love hockey, we don't even have ice
5:165 minutes, 16 secondsin Ireland. This is where I insert a quote about you're skating to where the puck is going to be, not where it is right now. So, if you watch this channel
5:235 minutes, 23 secondsat all, you're going to be no stranger to vibe coding apps. And I wanted to talk today about monday.com's Vibe because basically I think it's
5:315 minutes, 31 secondsdifferent. It's a natural language app builder baked directly into the monday.com platform. So, yeah, of course there's tons of different app builders
5:405 minutes, 40 secondsout there, but I think what's really different about this, you're actually building your application on top of your monday.com infrastructure and context.
5:485 minutes, 48 secondsAnd as we talk about in this video, the context and the understanding is really important. You've got your workflows,
5:545 minutes, 54 secondsyour intakes, your OKRs, everything that's important to your company is managed in there. But often there's this last mile problem where you want to
6:026 minutes, 2 secondssolve something like, you know, I wish I had the perfect form for pulling that in or uh we need a clean page that represents our OKRs. So, with
6:116 minutes, 11 secondsmonday.com's new vibe, you can use natural language to create bespoke interfaces with inputs and outputs that write back to your monday.com data like
6:206 minutes, 20 secondsa sales forecasting app, a campaign health tracker, a time tracker app. So,
6:246 minutes, 24 secondsit's pretty much living in Monday where you and your staff works. This really is a super tool. Viewers can get started for free. Check out the link in the
6:326 minutes, 32 secondsdescription down below. So, already Carpath has given us some real gold.
6:366 minutes, 36 secondsLet's move on to his next point. And this is all about verifiability and building a moat as we go forward. So
6:436 minutes, 43 secondsKaparthi went on to explain why a lot of frontier models fall down and how there's an advantage in that. The reason models are so good at code is because
6:526 minutes, 52 secondsit's deterministic and it's verifiable and that's great feedback for the model.
6:566 minutes, 56 secondsUnfortunately, a lot of the world isn't as verifiable as code. So, Karpathy suggests what are the domains that the large frontier labs are not focusing on

Andrej Karpathy
Andrej Karpathy is a Slovak-Canadian AI researcher, who co-founded and formerly worked at OpenAI, where he specialized in deep learning and computer vision. He served as the director of artificial intelligence and Autopilot Vision at Tesla. In 2024 he founded Eureka Labs, an AI education platform, whose first course will be LLM101n: Let's Build A Storyteller.
...more
7:057 minutes, 5 secondsright now that have some level of verifiability. Financial trading, supply chain and routting optimization,
7:127 minutes, 12 secondscontinuous integration and migration agents, data cleaning, labeling, you all have domain expertise in some niche part
7:207 minutes, 20 secondsof any workflow or company. what are the things that you can build in that space that take advantage of the things Karpathy is talking about in this
7:297 minutes, 29 secondsconversation. So thankfully in this talk he basically retires the term vibe coding and replaces it. So vibe coding
7:377 minutes, 37 secondsis great. It raised the floor. Pretty much anyone can build now. We all needed that. It pretty much democratizes building. But he says what professionals are doing now is agentic engineering.
7:467 minutes, 46 secondsAnd that's the kind of stuff I teach a lot in my channel and on my course. This means things like using specs, plans,
7:527 minutes, 52 secondsmanaging your context window, doing proper review, making sure that we have unit level smoke tests to end tests and
8:008 minutesblockers in place in continuous integration so that we're not pushing bad code. Carpathy says he's seen people who are getting good at this go 10x
8:088 minutes, 8 secondsfaster. If you're on X at all, you'll see that, oh, people running 10 to 20 agents at the same time. Personally, I kind of think that's nonsense. I'm
8:168 minutes, 16 secondsreally good at this. I've been doing it for two years and the most I can keep in my head at any time is maybe three to four agents running at the same time. I
8:258 minutes, 25 secondsgive them a little bit of work. I check what they're working on. I have to review the code. These are production databases, production systems I'm working on. I'm not just going to push up whatever the agent was working on.
8:358 minutes, 35 secondsSo, if you're feeling behind and you're working with agents, but you don't feel like you're working with 10 or building entire uh empires in a single prompt,
8:428 minutes, 42 secondsthen don't worry. But the thing is we really need to think about how we get to that point because what Karpathi is talking about is we need to build again
8:508 minutes, 50 secondsfor where that puck is going. So we get to that point where not unlike in December when things just worked. I
8:588 minutes, 58 secondsthink we'll get to a point where our our agentic harnesses just work and we can run 10 to 20 agents at the same time if
9:059 minutes, 5 secondswe choose. We're not there yet, but we're on that path. By the way, if you're interested in building Agentic harnesses and production systems like
9:139 minutes, 13 secondsthat, so you can build apps really quickly, you can join my course and community at switchdimension.com.
9:199 minutes, 19 secondsIt's currently closed to new members right now, but I have a new course dropping in the coming weeks. If you want to be part of that cohort, just
9:269 minutes, 26 secondsdrop your name and email into the weight list. Okay, so let's wrap this all up into the four things you should build.
9:339 minutes, 33 secondsSo the first thing he suggests you build are tools that enhance your understanding, not just your speed. So here's what I do myself and here's a
9:419 minutes, 41 secondsreally easy way for you to get started with this. Here I've got clawed co-work, but it doesn't matter what tool you use.
9:479 minutes, 47 secondsI could do this in clawed code. You could do this in cursor. You can do it wherever you want. Essentially what you need is a folder and an initial prompt.
9:549 minutes, 54 secondsSo that prompt to your agent is where you tell it about your company, your app, your domain space, your life,
9:599 minutes, 59 secondswhatever it else it is, and ask it to create a set of strategy documents around that application, business,
10:0710 minutes, 7 secondswhatever the domain space is. So in my case, I have a company called Switch Dimension that helps people learn how to build with AI. And I got it to produce a
10:1410 minutes, 14 secondswhole lot of strategy documents around that through various conversations. And they're just simply stored as markdown in a folder. So my big problem is I want
10:2210 minutes, 22 secondsto build everything and take every opportunity. I just have a quick conversation with my strategy agent which now has an understanding of where
10:3010 minutes, 30 secondsI'm going and it keeps me on track. It gives me focus. It says you're going in the wrong direction if you work on that. Here's what you should be focusing on.
10:3610 minutes, 36 secondsHere's the opportunities. And every time I ask it to generate a doc or create new content, it does that with real context
10:4410 minutes, 44 secondsand understanding of my switch dimension world. This kind of a brain is the single most powerful thing I've built in the last couple of months. Highly
10:5110 minutes, 51 secondsrecommend you do something similar yourself. So the next thing you should build is agent first infrastructure.
10:5710 minutes, 57 secondsEverything we've built is pretty much built for humans. That's documents, dashboards, install flows, DNS settings.
11:0311 minutes, 3 secondsThe entire internet is built really for humans. So the real win in this next generation is stripping away all of the
11:1011 minutes, 10 secondshuman UI. Would an agent know how to use this directly without any kind of human translation in between? We're seeing this on websites in e-commerce where
11:1911 minutes, 19 secondsLLM.txt files are there so that when agents arrive on a website, they can quickly figure out how to use it rather than reading through all the emotional
11:2711 minutes, 27 secondsmarketing that's directed at humans. It wants to just know how the API works,
11:3111 minutes, 31 secondshow can work with your product, is it trustworthy as fast as possible. Build for that. Number three, we talked about verifiable domain capabilities. So, a
11:4011 minutes, 40 secondslot of the big labs are covering the large domain areas, but they're not going to have the time or be specifically focused on reinforcement learning for all these little subniches.
11:5111 minutes, 51 secondsThere are millions of them in every part of our life and business. Can you build a reinforcement learning environment around this? Fine-tune it and own that
12:0012 minutescapability. So, don't dismiss this as an opportunity for you. We can now build anything. That's exciting. Take your handbrake off and go and do it. And
12:0912 minutes, 9 secondsnumber four, the big one, let's build apps that only exist now because of software 3.0. So that is not a faster
12:1812 minutes, 18 secondsspreadsheet, a faster uh UI interface on top of a workflow. We need completely new things like that large language
12:2512 minutes, 25 secondsmodel knowledge base carpath heath was talking about that literally we couldn't exist because there was no code that could actually do it. There's so many
12:3212 minutes, 32 secondsnew things we can build now that we're still thinking about apps and the old way of doing things. There's a whole new approach now with these reasoning models
12:4012 minutes, 40 secondsthat we can apply. Let's be honest, AI will change the jobs market. It will change how we work. If we think positively about what we can build now,
12:4812 minutes, 48 secondsthe reframing of data, the compilation across all these modalities. We've got large language models now that actually push and progress other areas of science
12:5512 minutes, 55 secondsand medicine. We have a really exciting time ahead of us. On a personal note, I went through my own side project folder this week and I killed at least three project after watching Karpath's talk.
13:0713 minutes, 7 secondsBoth of them failed the menu gen test and the software 3.0 test. They were basically software 1.0 plumbing for
13:1413 minutes, 14 secondsthings the next model release is probably going to do natively. This of course is painful, but it's better just for me to kill them now than to shift
13:2113 minutes, 21 secondsthem and watch them die in 3 months time. But saying that, there are one or two apps that I am doubling down on that hits all four of the criteria up above.
13:3013 minutes, 30 secondsI'll be sharing more on that soon. So,
13:3213 minutes, 32 secondshit me up in the comments and I'm asking this seriously. What is the app you've built that probably shouldn't exist anymore? I want to see how many of us
13:3913 minutes, 39 secondsare in the same boat. No judgment. I'm in it, too. If this was useful,
13:4313 minutes, 43 secondssubscribe. I'm doing a follow-up on building agent first infrastructure that goes way deep into the practical side.
13:4913 minutes, 49 secondsAnd if you want the actual playbooks I use for building with agents, go and check out the switch dimension loller and community. They are linked below.
13:5613 minutes, 56 secondsI've also linked to Karpathy's full talk down below. I'm supposed to link one of my own videos, so you go there and watch one of those next, but I highly recommend checking it out. And I've also

Andrej Karpathy
Andrej Karpathy is a Slovak-Canadian AI researcher, who co-founded and formerly worked at OpenAI, where he specialized in deep learning and computer vision. He served as the director of artificial intelligence and Autopilot Vision at Tesla. In 2024 he founded Eureka Labs, an AI education platform, whose first course will be LLM101n: Let's Build A Storyteller.
...more
14:0414 minutes, 4 secondsgot a ton of other videos on this topic on the channel.

Sync to video time