/
RSS Feed
Ever since Andrej Karpathy first tweeted it, “vibe coding” has been on every software developer’s mind. Join Ben Lorica and Steve Yegge to find out what vibe coding means, especially in a professional context. Going beyond the current memes, what will the future of software development look like when we have multiple agents? And how do you prepare for it? Don’t push back against AI now; lean into it.
Check out other episodes of this podcast on the O’Reilly learning platform.
About the Generative AI in the Real World podcast: In 2023, ChatGPT put AI on everyone’s agenda. In 2025, the challenge will be turning those agendas into reality. In Generative AI in the Real World, Ben Lorica interviews leaders who are building with AI. Learn from their experience to help put AI to work in your enterprise.
Points of Interest
- 0:00: Introduction to Steve Yegge, Evangelist at Sourcegraph
- 0:36: Let’s start with CHOP. What do you mean by “chat-oriented programming,” and how does it change the role of a software developer?
- 1:02: Andrej Karpathy has come up with a more accessible packaging: “vibe coding.” Gene Kim and I are going with the flow in our book, which is also about agentic programming.
- 2:02: The industry has the widest distribution of understanding that I’ve ever seen. We’ve got people saying, “You ought to stop using AI”; we’ve got people refusing to use AI; we’ve got people spread out in what they’re using.
- 3:03: Vibe coding started off as “it’s easy.” But people misinterpreted Karpathy’s tweet to mean that the LLM is ready to write all the code. That’s led to production incidents, “no vibe coding,” and a debate over whether you can turn your brain off.
- 3:35: Google decided to adopt vibe coding because you can do it as a grownup, as an engineer. You don’t have to accept whatever AI gives you. If you’re doing a weekend project or a prototype, you don’t have to look carefully at the output. But if you’re doing production coding, you have to demand excellence of your LLM. You have to demand that it produces code to a professional standard. That’s what Google does now.
- 4:38: Vibe coding means using AI. Agents like Claude Code are pretty much the same.
- 4:58: There’s traditional AI-assisted coding (completions); with vibe coding, the trust in AI is higher. The developer becomes a high-level orchestrator instead of writing code line by line.
- 5:37: Trust is a huge dimension. It’s the number one thing that is keeping the industry from rocketing forward on adoption. With chat programming, even though it’s been eclipsed by agent programming, you get the LLM to do the work—but you have to validate it yourself. You’re nudging it over and over again. Many senior engineers don’t try hard enough. You wouldn’t boot an intern to the curb for failing the first time.
- 7:18: AI doesn’t work right the first time. You can’t trust anything. You have to validate and verify. This is what people have to get over.
- 7:53: You’re still accountable for the code. You own the code. But people are struggling with the new role, which is being a team lead. This is even more true with coding agents like Claude Code. You’re more productive, but you’re not a programmer any more.
- 8:51: For people to make the transition to vibe coding, what are some of the core skillsets people will have to embrace?
- 9:07: Prompt engineering is a separate discipline from CHOP or vibe coding. Prompt engineering is static prompting. It’s for embedding AI in an application. CHAT programming is dynamic; lots of throwaway prompts that are only used once.
- 10:13: Engineers should know all the skills of AI. With the AI Engineering book by Chip Huyen, that’s what engineers need to know. That is the skills you need to put AI in applications, even if you’re not doing product development.
- 11:15: Or put the book into a RAG system.
- 12:00: Vibe coding is another skill to learn. Learn it; don’t push back on it. Learn how it works, learn how to push it. Claude Code isn’t even an IDE. The form factor is terrible right now. But if you try it and see how powerful agentic coding is, you’ll be shocked. The agent does all the stuff you used to have to tell it to do.
- 13:57: You’ll say “Here’s a Jira ticket; fix it for me.” First it will find the ticket; it will evaluate your codebase using the same tools you do; then it will come up with an execution plan. It’s nuts what they are doing. We all knew this was coming, but nobody knew it would be here now.
- 15:18: There’s the UX of the tool and the underlying models. The models are converging in capability, and a lot of that has to do with the fact that programming lends itself to AI. In your experience, are you using different models?
- 16:16: Yes, I use different models. Models aren’t all equal. Often, getting a second set of eyes is a great way to validate or verify. For example, if you’re working in a language you’re not familiar with. You can’t trust a model. Get a second opinion from another model.
- 17:20: With that said, Claude is the best at coding, even though the benchmarks aren’t really reflecting it. The models all seem to be getting their specializations.
- 18:10: But there’s a commoditization going on here. The models are all getting better. At some point, they will all be good enough. You don’t need a fancy model to brute force problems. Models may not matter much in a year.
- 18:59: You say that one way to think of these projects is like a graph. The LLMs are great at handling the leaf nodes—subtasks. Where are we in terms of LLMs handling context across the whole graph?
- 19:30: It ain’t good right now. The world is missing MCP (Model Context Protocol) servers for everything. Without MCP, there’s no visibility; AI is missing context.
- 20:11: To clarify: Right now, you have to hook up all the connections manually.
- 20:20: Yes. AI can read your Git repo, but what about databases, logs, documentation? You want the AI to have all of that at some point. It’s invisible. If you want AI to help you, you have to show it your stuff. MCP is a protocol to tell AI how to get to the data. We’re all digging through a giant mountain. With coding agents, you have a full tunnel boring machine. It won’t get through the mountain in a day, but it will get stuck, [and] you’ll have to get it back on track. It’s not a panacaea. You have to come at it with the right mindset, but you will still be more productive.
- 22:21: There’s software development, testing, QA, security; one of the pushbacks must be that AI can introduce bugs and vulnerabilities. But I imagine there is also AI for those aspects: I imagine AI doing penetration testing. Are you following developments for these other areas?
- 23:08: It’s AI all the way down. The next skills developers will need to hone are humanities skills. You need to know how to communicate and coordinate, and all the things you don’t learn in engineering degrees. We’ll all be managing agents, and our job will be to be the shepherds of these things.
- 23:51: You wrote an interesting post about career development. What is career development in the age of AI? Even before that: What is computer science? Should you get a degree? And how do I progress?
- 24:27: This is one of the hardest questions in the world. It’s tough. But when assembly language went away, people freaked out. There was a lot of pearl clutching. But software jobs went up dramatically. That’s what’s happening with programming now.
- 25:28: It’s all computer science; you’ve just moved up the abstraction ladder. You are not so worried about languages; you are more worried about architecture and algorithms and network topologies and things you need to understand or else you’re flying blind.
- 26:12: You’ve hinted at agents. Where will we be with agents in 6 to 12 months?
- 26:30: First, I’ve come to understand that there are six overlapping waves of AI modality, of which five are active right now: traditional programming; code completions; chat programming; vibe coding; agents. Where does this take us? The sixth wave, which we don’t have yet, is agent clusters. If you sit down with a coding agent, aider chat or Claude Code, you realize that you want to run two at once. When it’s working, you’re not. So you want more. I managed to run two for a weekend. That gave me a terrifying glimpse of the future.
- 29:09: Each agent is roughly as productive as I am if I can keep its queue full. They cost about $10 per hour. If you can use five at once—if you can have a dashboard that manages multiple agents at once—now you’re spending $50 per hour to be five times as productive. And if you don’t do this, you’ll fall behind.
- 30:12: Most companies just finished their yearly budget planning, and it’s frozen. Coding agents are taking off, and that adds to a nontrivial percentage of the engineer’s salary which wasn’t budgeted for. Where’s the money going to come from? If you have deep pockets, you’re in good shape, but if you don’t have the capital to absorb the bump in token spend, you will have to cut somewhere. And there’s only one department to cut, and that’s the one that’s getting 5x as productive. That’s why I’m shouting that people better learn this stuff.
- 32:04: For our listeners, what’s your advice to make sure they’re not part of the layoff? There was a natural skill progression. Will companies cut junior developers? And how do I avoid that?
- 32:52: Now I’m seeing more senior developers struggle with this than junior developers. Junior developers have figured this out. They have been going to an AI and saying, “Make me a great programmer.” They’re buying the AI book; they’re reading papers. They’ll survive. If you’re still dabbling with coding assistants, it will be over for you. You have to evolve.
- 34:24: There’s a disconnect here too. A lot of the Y Combinator startups are generating most of their code by AI. But if you look on Hacker News, you see a lot of people pushing back.
- 35:10: It’s like somebody just threw a boulder in our pond. There’s denial; it’s wild. We’re writing a book, and every time we write something down it changes. But that reaffirms how much we need it. There’s been nothing like this in history.
- 37:12: It’s crazy the denial people are in. I got lucky; you need to generate your own luck. Memorize the AI Engineering book, and never write another line of code again.
- 37:45: Many of these services are relatively cheap, in terms of bang for your buck. Are you aware of tools for the people who don’t want to spend anything, are there tools that can run on Ollama?
- 38:17: Aider chat can use Ollama. That’s a cheap way. If you’ve never found this, just try ChatGPT or Claude. Give it some problems and start tinkering with it. It’s a low-friction way to get started.
On May 8, O’Reilly will be hosting Coding with AI: The End of Software Development as We Know It—a live virtual tech conference spotlighting how AI is already supercharging developers, boosting productivity, and providing real value to their organizations. Register for free here.