- 2 hours ago
Category
🗞
NewsTranscript
00:00This is a story about a double-edged sword.
00:02At this point, AI is promised to do just about everything,
00:06but it turns out that it's best at creating the very thing that it is made of, code.
00:11It was once considered a high-skill task,
00:14but coding is now accessible to just about anyone with the help of generative AI.
00:19On the one hand, this unlocks possibilities for creating a wide range of products and businesses
00:23that otherwise might never have seen the light of day.
00:26On the other hand, it's made the future less certain for those who write code professionally.
00:31Our colleague Ed Ludlow brings us the story of what happens when we lean in and let AI do the
00:37work.
00:43In Upper Tract, West Virginia, Jamie Grove owns a boutique warehouse
00:47helping clients ship out anything from dinosaur bones to board games.
00:51Last year, he built software that automates shipping out packages with help from AI.
00:58Where's the order at that you're looking at there?
01:00There's absolutely no way I could have done any of this without AI.
01:04For the initial run, to take some spreadsheets that we had, that we were working with,
01:08to get an actual workable sample ready to go, took me less than a day.
01:13And we were live, and we were using it, right here in the warehouse.
01:17In Oakland, California, Cynthia Chen wished there was an app
01:21that collects pictures of different breeds of dogs.
01:24And then it was when I learned about vibe coding online
01:27is when I thought maybe I should try it myself.
01:30The first time you see, like, this little pop-up and it says build succeeded,
01:34and then you can see the app pop up and, like, it's actually real.
01:38That was, like, the sort of magical moment where I was like,
01:41oh my gosh, this is crazy, I can actually build things myself.
01:46The term vibe coding was coined by Andrej Karpathy,
01:50a founding member of OpenAI,
01:52to describe a process of computer programming
01:54akin to having a conversation with a robot.
01:58Here's what it looks like in practice.
01:59Say I want to create a website that visualizes and animates a data set.
02:03I would use a generative AI tool like Claude or Codex,
02:07or in this case, Gemini,
02:08and tell it exactly what I want using plain English.
02:12AI then writes the code for me.
02:13And as coding has gotten easier,
02:15it's led to the creation of more code than ever before.
02:19Activity on GitHub, the platform used for storing and sharing code,
02:22has seen a massive increase in activity,
02:24surging in early 2025 when AI coding pilots became popular.
02:31And although vibe coding has helped small businesses
02:34and hobbyists create their own software,
02:37it's the professionals who are leading the way.
02:39So what goes on in this building is we have a mix of folks working on Google Cloud.
02:44At Google, when we talk about autonomy and agentic engineering,
02:48we have systems that allow you to kind of have almost a virtual software engineer.
02:53Hold on.
02:54If all these brilliant engineers are using AI in this way,
03:00what is it that they're doing all day long in beautiful buildings like this one?
03:04Excellent question.
03:08Adi Osmani is the director of Google Cloud AI.
03:11He oversees teams of engineers currently building the next generation of AI tools for businesses.
03:17So if you are vibe coding, you're pretty much just giving into the vibes.
03:21You don't necessarily have a clear, full idea of your vision.
03:24You're just working with the LLM.
03:26You're trying to get somewhere with it.
03:28If you're engineering, that's where you have to apply rigor to it.
03:32You have to have this clear set of requirements.
03:35You are testing.
03:36And whether you are a startup or whether you're in a big enterprise right now,
03:40the role of the software engineer is going to be evolving to one
03:44where you are increasingly a little bit more of a manager.
03:47You're going to have effectively like a virtual team of agents that you're responsible for.
03:52And you have to own the outcomes.
03:54It doesn't matter how many agents you're running.
03:56You're responsible for the output.
03:56You're responsible for the output.
03:58Exactly.
03:59And so you need to decide like, how am I evaluating quality?
04:04How much time am I going to spend evaluating quality?
04:07Because there are some people who very much enjoy YOLO.
04:09Like, okay, well, the agents ran overnight.
04:11Looks good.
04:12Kind of runs.
04:13I'm just going to deploy it.
04:14But if you're building any kind of serious software,
04:16you still need to have some idea of like, what is the quality bar?
04:20What are my quality gates?
04:21How am I making sure this is actually going to meet the needs of my users in a consistent way?
04:29For the engineers at Google's headquarters here in Sunnyvale, California,
04:32Osmani says AI isn't just making their lives easier.
04:35It's making them better at coding.
04:38The extension of that question, which we pose largely by investors,
04:41is how do we measure the productivity gains of that engineer or that team?
04:48I remember in the earlier days of AI, you know, org leaders would look at things like,
04:53oh, hey, well, how many lines of code are being generated, right?
04:56Which is not in any way a good proxy for productivity.
05:00But these days, I think that people use a mix of different kinds of metrics.
05:04You try to use qualitative and quantitative.
05:06Generally speaking, there's a big productivity boost.
05:09In the earlier days of AI, I would have said, you know, that boost is 10 to 15%.
05:12These days, it's anywhere from 30 to 50%.
05:15And I see that number only continuing to go up.
05:19At MIT, Frank Nagel and a team of researchers surveyed over 187,000 software developers
05:27who are using GitHub Copilot, a generative AI tool for coding.
05:31And they found that workers are more productive because what they spend time on has changed.
05:37I think one of the big things that we often think about with AI is that it's just going to
05:40enhance productivity, right?
05:42It's going to make us faster doing whatever it is we do.
05:46But that's just really just scratching the surface.
05:48You have 100% of your time that you allocate to work.
05:51How did that break down along these dimensions of what we called core work, actual coding,
05:56versus more project management type of work?
05:58And what we found is that when coders started using these types of tools,
06:03they massively shift the amount of their time that they allocate to coding
06:08and they take away a whole lot of their allocated time from project management.
06:12And so part of the reason we think that this is happening is that if in the old days
06:17you were writing piece of code A and that piece of code was dependent on some other piece of code
06:22B,
06:22you had to wait for that other person, you had to interact with them
06:26and make sure that everything worked together, whereas now you can just write it all yourself.
06:32This productivity boost is particularly true for those who are writing and deploying brand new code,
06:38and especially those with no knowledge of code whatsoever, including creatives like Chen.
06:42This is Press Petals, the new app that I'm working on.
06:45And here is a press flower.
06:48So I'm going to go to the clod code that I have running,
06:52and then I'm simply going to describe what I wanted to do.
06:55I actually tried a whole bunch of tools in the beginning, and this was about a year ago,
06:59which I think is like, feels kind of like the stone ages of vibe coding.
07:03I was able to make the foundations of the app in maybe a month.
07:08And as like a totally non-technical person, I was actually able to like build a full stack app.
07:14So there's like front-end and back-end capabilities,
07:16and I was able to get it out on the app store just entirely myself.
07:23For business owners like Grove, his vibe-coded solution is helping his warehouse save on costs.
07:29We have three main coding solutions here that we've used AI for.
07:33One is to create batching, which is the important part in terms of taking orders that are different
07:39and getting them all together into groups that make that easy to pick.
07:43And then the other solution that we have is inventory tracking.
07:47So that's a standard warehouse feature, but our clients are also very different.
07:52And so trying to force a client into a single inventory tracking system is really difficult.
07:58So we actually, we've used AI to build a inventory tracking solution that allows them to be themselves, basically.
08:05So I have a pretty varied background.
08:07I started out as a programmer a long, long time ago, but even someone who is a very fast coder
08:15could not have built all these solutions impossible.
08:19If I were to do it with a team of programmers, I could have five programmers working on this full
08:23-time
08:23and still not deliver as many results.
08:26If I were to install a software system that does everything that we're doing now,
08:31let's just say without all the customizations and all the flexibility,
08:35we might be talking about an annual license of anywhere between $6,000 and $10,000 a year,
08:40scaling all the way up to maybe $30,000 to $50,000 a year,
08:43depending on how much volume we push through our warehouse.
08:46This doesn't have the flexibility that we would want.
08:49And it's expensive.
08:50Like for a small boutique warehouse, that's a big expense.
08:56Now, for about $20 a month, business owners like Grove are building their own software solutions.
09:02And whilst that's opened up new possibilities, there is a downside.
09:07Since 2022, employment for software engineers right out of college has fallen by nearly 20%.
09:13Nagel thinks companies are making a big mistake.
09:16I do think that one of the biggest risks of the whole thing
09:19is that people are going to get too focused on the short term and not think about the long term
09:24enough.
09:24First of all, if you don't hire any new people, who's going to run the company in 10 to 15
09:28years?
09:28But second of all, our research and others has shown that these junior people are actually the ones
09:34who are able to change their job and get the most out of using these tools.
09:38And so if we're not hiring them at the same rate we were before,
09:41then we're not going to be able to take advantage of that.
09:44And Chen agrees.
09:45I feel very strongly that this is like not replacing engineers.
09:52I think the more you use AI to build, the more you understand the space
09:56and kind of even know what you can and can't do.
09:59It's like technically I can make it, but an engineer probably could have made this
10:04in a much shorter timeline and probably with like much more robust code.
10:09From the business standpoint, I do think there's this opportunity
10:11where companies that have been thinking about things like reverse mentoring,
10:15where the younger folks can help the more experienced folks learn how to better use these types of tools,
10:21while the experienced folks are able to better help the younger folks understand how the industry works.
10:30If AI can write code, then what is the role of the software engineer?
10:34Osmani says today it's all about quality.
10:37If I want to build a robust engineering artifact, something that's going to last time,
10:43something I can ship to hundreds of millions of users or billions of users,
10:47there are a lot of things that it needs to factor in.
10:50And so what people in buildings like this are doing all day long are trying to make sure
10:53that the code is actually meeting that quality bar
10:56so that when we do ship something to you that happens to be using AI behind the scenes,
11:01you're actually getting what you want.
11:02That's the important thing for the users at the end of the day.
11:04They don't care if a human has been authoring it or AI has been authoring it.
11:07It doesn't help them get the job done in a reliable way.
11:10Is vibe coding a term that's therefore used?
11:13Is it banned within teams?
11:15Absolutely not.
11:16I think that vibe coding has a lot of value.
11:19Vibe coding is enabling people to go from idea to execution faster than ever.
11:23And it has completely changed how many teams approach prototyping.
11:27You know, so many times in the past, in Silicon Valley and lots of places,
11:31if you had an idea, you'd go through weeks or months of just discussing or debating,
11:36hey, can we afford to build this?
11:38Now you can just build it.
11:39And I think that it has a very concrete place in our language now.
11:44It's just important that you understand that vibe coding a thing does not necessarily mean
11:49that you have a production-ready artifact that's going to be battle-hardened.
11:55In Silicon Valley, the test of AI's progress is true autonomy.
11:59To start a project in the evening and wake up in the morning to code written autonomously by an AI
12:05agent.
12:06A dream for senior software developers and perhaps a nightmare for entry-level computer engineers
12:12who would have once done that work themselves.
12:14But for us, non-coding mere mortals, the moment could be ripe to do what mankind is best at.
12:21Building and creating.
Comments