Deep Dive
Shownotes Transcript
Missions to Mars? Driverless cars? AI chatbots? Feels like we're already living in the future. Well, Robinhood is built for the future of trading. Robinhood's intuitive design makes trading seamless, so you can spot opportunities and take control of your trades. You can now even trade your stocks and crypto all in one place. Sign up for a Robinhood account today.
Investing is risky. Robinhood Financial LLC member SIPC is a registered broker-dealer. Cryptocurrency services are offered through an account with Robinhood Crypto LLC, NMLS ID 1702840. Robinhood Crypto is licensed to engage in virtual currency business activity by the New York State Department of Financial Services.
Attention all early birds. When you make a donation to Marketplace during the first two days of our May fundraiser, it can make a big difference. The Investors Challenge Fund has issued us a challenge. If we raise $25,000 in the first two days of this fundraiser, we'll unlock another $25,000 from the Challenge Fund. We can't afford to fall short here. Please step up and show your support today. Give what you can at marketplace.org slash donate today.
Learn to code, but with vibes. From American Public Media, this is Marketplace Tech. I'm Megan McCarty Carino. Vibe coding is having a moment. The buzzy new phrase was coined earlier this year by OpenAI co-founder Andrej Karpathy to describe his process of programming by prompting AI. It's been embraced by tech professionals and amateurs alike.
Google, Microsoft, and Apple have or are developing their own AI-assisted coding platforms, while vibe coding startups like Cursor are raking in funding. Clarence Wong joins us to help explain this concept. He's the vice president of technology at the financial software company Intuit and an early adopter of vibe coding. He says it's changed how he approaches building software.
Think of it less like traditional coding and more like actually having a conversation with a super-powered coding partner. You literally start by describing your vision for the software that you want to build. In plain English, you will tell the AI assistant in words like, build me this, build me that.
fix that error, the AI will then take over. It will take a shot at generating that code, and then you will see the results of what it generated. And then you'll see it work and give the AI feedback. And you just keep doing that over and over again with your natural voice
And you essentially become the director guiding the AI rather than typing out any code yourself. And it's a great way to build software. How are you thinking about, you know, where the human coder sits in this loop? I think this is really an evolution of the skill set of software engineering, right? I used to be a competitive programmer back in college, and we used to practice typing as fast as you can, right?
So that sort of menial mechanical work of typing out code, which used to be a competitive advantage as a coder, is just no longer a thing, right? Because the AI can just write code way faster than you can. But really, with Vibe Coding and with these AI coding assistants,
the limit is no longer your fingers, right? It's human ingenuity. It's your clarity of thought. It's your clarity of communication. Can you supervise the AI to do what you need it to do, right? Can you read code fast instead of writing code fast? And I'd argue like reading code is actually a much harder skill set than writing code. So it's really the skill sets shifting. Are there any instances where vibe coding is not the right approach for the work that you're doing at Intuit?
Yeah, you know, vibe coding is a tool, right? And there's a right way to use that tool. There is a wrong way to use it, right? But it's sort of what you do with that end result that really matters. So when we say vibe coding is wrong, it's really more in the sense of that if you don't do sufficient checks on the output from the AI. And that can have really bad consequences, right? Because if you don't read the code and understand it,
that AI-generated code can be a black box, right? It can be potentially buggy. It can be insecure. It can be a nightmare to fix later for the next engineer who has to take over that code base. So really rushing code with AI without strict checks is,
It's almost like leaving your front door open for the hackers. For really production quality software, having those strict checks is really, really important, right? You know, the golden rule I have on my team is don't commit code you can't explain, even if an AI wrote it. We'll be right back.
When Ram won its first Motor Trend Truck of the Year, it was celebrated. But after seven times, what's left to say? The trophy speaks for itself. And now, so does the deal with Employee Pricing Plus. Stop by your local dealer today and take home a Ram truck. Ram, seven-time Motor Trend Truck of the Year. Nothing stops Ram.
Offer valid on select 2024 and 2025 RAMs for non-FCA employees and retirees. $200 admin fee applies. Not all buyers will qualify. Restrictions apply. See dealer for details. And 6-2-25. RAM is a registered trademark of FCA US LLC. This Marketplace podcast is supported by GoFundMe Pro. Over the past 15 years, the fundraising platform Classy has helped nonprofits reach more supporters and raise more money.
Now Classy is becoming GoFundMe Pro, combining best-in-class nonprofit tools with the world's largest community for good and tapping into GoFundMe's power, scale, and community of 190 million. From launching campaigns to rallying supporters, GoFundMe makes it easier than ever to inspire giving. Visit pro.gofundme.com to uplevel your nonprofit today. That's pro.gofundme.com.
You're listening to Marketplace Tech. I'm Megan McCarty Carino. We're back with Clarence Wong, Vice President of Technology at Intuit.
For folks who maybe don't have the strong foundation in learning to code without the help of AI that you do, do you think there's a risk for, you know, programmers that are just coming up in this environment that they just might not have the full context or the skill of programmers like you who did things the hard way? Yeah, I think the skill set erosion is a real concern, right? If you're so reliant on the AI that you don't,
have the foundations to even architect and design a software together. If you don't know what is actually good and what is bad when you judge the outputs of the AI, then you can't leverage vibe coding and AI assistance properly without all the pitfalls. But I will also say that AI technology is...
It is also an enabler for folks to learn faster. When the AI generates outputs, I will often ask the AI to explain it. And I'm learning things in that process, right? These AI coding assistants are trained with basically the collective knowledge of the entire human species that's reflected on the internet, right? So it naturally knows more than me or any other engineer know by themselves. So I do think we just have to...
use these tools in the right way, use them judiciously to get all the upsides from it and avoid the downsides. We've been starting to hear from, you know, big tech companies about how much AI is involved in their code. You know, Google said, I think, about 25% of its code is AI generated. Microsoft said up to 30% of its code. What do you make of this? Yeah, you know, we see sort of similar like productivity improvements here at Intuit.
We were actually one of the early adopters of AI-assisted coding. Just within the company, I see a lot of momentum in folks adopting these tools. Even with vibe coding, we have a community channel where people who vibe code get together to share techniques and sort of trade tips. And that started off with like three people and now it's grown to the hundreds.
Do you think vibe coding is the right way to describe this? I think so. You know, the word vibe is really around human judgment, right? It's around judgment and decision making. When we talk to the AI and it gives the results back, we have to still exercise that judgment, right? And I feel like that's actually a skill set that we have to develop as engineers. As time goes on, our vibes will become better and we'll become more efficient at coding with AI assistance. That was Clarence Wong at Intuit.
Clarence mentioned that AI-generated code in inexperienced hands could be more vulnerable to cyberattacks. And a recent study from researchers at the University of Texas San Antonio identifies a specific threat it calls slop-squatting.
AI large language models sometimes hallucinate or make up names for bundles of code called packages. That's the slop. So hackers could monitor public code repositories like GitHub for these fake packages that don't exist and then publish malicious versions of them, which could open the door to injecting malware into unsuspecting developers' code bases. That's the squatting. Thus...
slop squatting. Daniel Shin produced this episode. I'm Megan McCarty Carino, and that's Marketplace Tech. This is APM.
Can we invest our way out of the climate crisis? Five years ago, it seemed like Wall Street was working on it until a backlash upended everything. So there's a lot of alignment between the dark money right and the oil industry on this effort. I'm Amy Scott, host of How We Survive, a podcast from Marketplace. In this season, we investigate the rise, fall, and reincarnation of climate-conscious investing.
Listen to How We Survive wherever you get your podcasts.