Jevons Intelligence: Why Agent Coders Will Turn Everyone Into Software Developers
17:20 Share

Today on the AI Daily Brief, Replit has released an agent app that not only can code your ideas from a prompt, but can do it from your phone. Before that in the headlines, robotics company Figure has dropped their partnership with OpenAI. The AI Daily Brief is a daily podcast and video about the most important news and discussions in artificial intelligence. To join the conversation, follow the Discord link in our show notes. Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around five minutes.

Apologies, as you can tell, I am finally succumbing to the child sickness part of the year. And unfortunately, given how much I talk, it always hits my voice first. Anyways, we persevere. And we start today with robotic startup Figure AI dropping their partnership with OpenAI in favor of using their own in-house models. Founder Brett Adcock posted, Today I made the decision to leave our collaboration agreement with OpenAI. Figure made a major breakthrough on fully end-to-end robot AI built entirely in-house. We're excited to show you in the next 30 days something no one has seen on a humanoid.

Interestingly, OpenAI has been a long-time investor in FIGURE as well as being a strategic partner. Last August, the two companies announced that the FIGURE O2 humanoid robot would use OpenAI models for language processing. As it turns out, though, FIGURE decided that general-purpose models are not well-suited for robots. Instead, they found a need to build more focus models optimized for power-specific hardware.

Adcock told TechCrunch, We found that to solve embodied AI at scale in the real world, you have to vertically integrate robot AI. We can't outsource AI for the same reason we can't outsource our hardware. Figure is building general-purpose humanoids suitable for home and industrial use, but their current focus is optimizing their robots for the factory floor.

FIGURE announced last January that BMW had begun to deploy their robots in their South Carolina plant, but the scope of the rollout was not disclosed. There's also the question of whether OpenAI will compete directly with FIGURE in developing robots. The company has staffed up a small robotics division and filed a trademark application earlier this week to cover humanoid robots. Certainly no guarantee that they're bringing a product to market, but pretty good confirmation that they're at least thinking about it.

Based, Beth Jazo says, all robotics company will have to become foundation model companies if they want a real data flywheel moat. Probably the right move. As for the major breakthrough that Adcock referenced, it could be something to do with a new contract signed by the company.

Last week, Adcock announced that Figure had signed their second major client, claiming it to be one of the biggest US companies. He said that the new contract, combined with the BMW deal, could allow Figure to ship 100,000 robots over the next four years. And while that part caught the headlines, Adcock also commented, "...last week we successfully began running an end-to-end neural network on the new client's use case. Learning the use case with AI is the only path here as heuristics would be impossible to write. And every time I see these policies running, it feels like pure magic."

Others believe that breakthroughs in open source AI have made open AI look like an extremely expensive option and that might be what's driving this behavior. 10Storens Mark O'Connor wrote, In hindsight, this is an obvious consequence of open source R1, probably other such cases. Jonathan Mayorka commented, Robots are coming sooner than you think. I think this decision from the founder of Figure correlates with the advancement from DeepSeek. It being open source, a reasoning model, and the fact that you can run it on a PC locally with no internet, my mind went immediately to robots.

Speaking of controversial Chinese companies, TikTok parent company ByteDance have released a new generative video model. Called OmniHuman1, the model is tuned for image-to-video workflows, specifically for use in animating humans. The model is capable of animating both the face and the body, which has been a struggle in the past. The generated videos include lip sync and gestures that match speech or music. In their research paper, ByteDance researchers said...

End-to-end human animation has undergone notable advancements in recent years. However, existing methods still struggle to scale up as large general video generation models, limiting their potential in real applications. The reverse is also true. When OpenAI released Sora late last year, many found the image-to-video feature extremely disappointing. ByteDance said they trained OmniHuman on 18,700 hours of human video data. They used a novel training strategy of coupling text, audio, and body movement data together to cut down on data wastage.

Alex Volkov, the host of the Thursday AI pod, wrote, ByteDance is on fire. OmniHuman 1 is the craziest one-image-to-human avatar model so far.

Lastly today, it is earnings season, and DeepSeek continues to cast a shadow. Google reported $96.5 billion in revenue and a 12% annualized growth rate for Q4, which was its lowest for the year. Net income came in at $26.6 billion, up 28.3% from a year earlier. The stock plunged by 7% in after-hours trading in response to the cooling numbers. More interesting than that, though, the discussions on the earnings call centered around DeepSeek and how the company was responding to the perceived threat of Chinese competition in AI.

Referring to DeepSeek CEO Sundar Pichai said, "First of all, you know, I think it's a tremendous team. I think they've done very, very good work." However, he added that Gemini's Flash models are "some of the most efficient models out there," including comparing to DeepSeek's V3 and R1. More generally, Pichai said that the collapse in price for serving AI models would be a good thing for everyone, stating, "Part of the reason we're so excited about the AI opportunity is we know we can drive extraordinary use cases because the cost of actually using it is going to keep coming down, which will make more use cases feasible."

That's the opportunity space. It's as big as it comes, and that's why you're seeing us invest to meet that moment. Pichai said the company would spend $75 billion on capital expenditure this year, up almost 50% from 2024.

Investor Lawrence McDonald put forward one of the most common takes from Wall Street at the moment, posting, Google CapEx forecasts $75 billion versus $58 billion previous guidance. This is an all-you-can-eat AI arms race buffet. Former cash earnings cows are becoming highly capital-intensive businesses, all trying to outgun the next guy. Fiber optic cable 1998 to 2001, shale oil 2010 to 2014, repeat.

However, there is a big divergence between that attitude and those actually building AI companies. With Aaron Levy, the CEO of Box, writing, Google expects to spend $75 billion on CapEx this year. That's up about 50% from last year. Turns out we're only at the start of the AI super cycle.

That, I think, is a perfect place to leave this Headlines episode. Up next, the main episode. Today's episode is brought to you by Vanta. Trust isn't just earned, it's demanded. Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in.

Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC 2 and ISO 27001. Centralized security workflows complete questionnaires up to 5x faster and proactively manage vendor risk. Vanta can help you start or scale up your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building your company.

Join over 9,000 global companies like Atlassian, Quora, and Factory who use Vantage to manage risk and prove security in real time.

For a limited time, this audience gets $1,000 off Vanta at vanta.com slash nlw. That's v-a-n-t-a dot com slash nlw for $1,000 off. If there is one thing that's clear about AI in 2025, it's that the agents are coming. Vertical agents by industry horizontal agent platforms.

Agents per function. If you are running a large enterprise, you will be experimenting with agents next year. And given how new this is, all of us are going to be back in pilot mode.

That's why Superintelligent is offering a new product for the beginning of this year. It's an agent readiness and opportunity audit. Over the course of a couple quick weeks, we dig in with your team to understand what type of agents make sense for you to test, what type of infrastructure support you need to be ready, and to ultimately come away with a set of actionable recommendations that get you prepared to figure out how agents can transform your business.

If you are interested in the agent readiness and opportunity audit, reach out directly to me, nlw at bsuper.ai. Put the word agent in the subject line so I know what you're talking about. And let's have you be a leader in the most dynamic part of the AI market. Hello, AI Daily Brief listeners. Taking a quick break to share some very interesting findings from KPMG's latest AI Quarterly Pulse Survey.

Did you know that 67% of business leaders expect AI to fundamentally transform their businesses within the next two years? And yet, it's not all smooth sailing. The biggest challenges that they face include things like data quality, risk management, and employee adoption. KPMG is at the forefront of helping organizations navigate these hurdles. They're not just talking about AI, they're leading the charge with practical solutions and real-world applications.

For instance, over half of the organizations surveyed are exploring AI agents to handle tasks like administrative duties and call center operations. So if you're looking to stay ahead in the AI game, keep an eye on KPMG. They're not just a part of the conversation, they're helping shape it. Learn more about how KPMG is driving AI innovation at kpmg.com slash US.

Welcome back to the AI Daily Brief. One of the most profound and important questions lurking underneath the advent of artificial intelligence is what it's likely to do to human labor. This is, of course, never far away from the discussion. AI increasingly knows more than almost anyone at any given field, and agents are actually finally coming online, able to take on some amount of tasks independently. And more than anything that exists today, the trajectory is just incredibly clear.

So what happens in a world where the cost of intelligence falls to zero? Will it involve the chaotic destruction of the economy and the replacement of all human labor with robot brains? Or is there another possibility?

For those looking for evidence, an interesting place to start is in the realm of software development. One of the great ironies of software engineering and AI is that these tools that this particular group have built seem like they might be most in line to displace that group. The product that launched the latest round of discussion is Replit's new phone app. If you're not familiar with Replit, they're a new kind of coding assistant aimed at non-developers. At least that's what they've become more recently.

The idea is to empower normal people to create apps and websites with a single prompt. And the product has now gotten to the point where embedded agents actually improve the user's prompt, create the code, debug, and test, and finally deliver a finished app within about 10 minutes. This week, they launched a version of this for phones. Balaji Srinivasan writes, Okay, this finally solves coding on the go. You have the idea, you tap out the tweet length description into Replit, you hit improve prompt, and then the AI agents start grinding away to make it.

Check back on a desktop a while later and you have a V1 prototype. The service at this stage is a little limited. Most of the examples are very simple and you're more likely to have success building an exercise tracker or a basic game than a more complex fully featured app. But this is still absolutely a step change in accessibility for programming.

Non-developers can now have a working prototype without knowing the first thing about coding. It's fast, cheap, and good enough to be very useful. And importantly, Replit is far from the only company tackling this particular space. John Rush, in fact, pointed to 24 AI coding agents and IDEs. Cursor, SoftGen, Windsurf, Rapify, Copilot, Lovable, Bolt, V0, Replit, MarsX, Claude, Amazon Q, Pear, Devon, GitHub Spark, IDX, WebDraw, Tempo, Klein, Continue, DataButton, Base44, Kodo, and Ader.

And whichever tool you pick, it seems pretty clear that these no- and low-code apps are starting to insert themselves earlier and earlier in the software design process. Paul Graham, the founder of Y Combinator, recently said that a CEO of a moderately big tech company had replaced Figma with Replit as their design tool, commenting, "'Replit is so good at generating apps that they just go straight to prototype now.'" Greg Eisenberg had a very similar example. He writes, "'I had to laugh at our product meeting last week. While we were discussing ideas for a new startup, one of my partners was quietly building it on vZero, Bolt, Replit, etc. right in front of us in real time.'"

By the time we finished describing features of What If, he had a working prototype on screen. Suddenly, we weren't having another product meeting we were debugging. No more imagine if we had conversations. Instead, it's look what happens when we. The gap between imagination and reality shrinks to minutes. Meetings become workshops. Ideas become experiments. Decisions become data-driven.

What's more, these services are getting major traction across the software industry. Swix posted, Today I learned that Cursor is the fastest growing SaaS in the history of SaaS, $1 to $100 million in 12 months with a wide and small customer base. And as you might expect, this has led to a million Xthreads from the Xthread boys, along the theme of this one from Deval Maquana that writes, RIP web developers, in 90 seconds AI can generate a fully functioning website. And Zinzeel writes, I think Replit just wiped out a jack load of developers. Mind-blowing.

Edward Frazier, a YC alum, even made an app with Replit to find out how long it would take for him to be replaced by Replit. After inserting a picture of his LinkedIn profile, the app said he would be fully replaced by 2028. Fekri writes, if anyone can build an app from their phone, what's our edge as founders? And this brings us to this question, nominally about software developers, but really about the world more broadly, of what happens when the cost of intelligence and the skills associated with it crash to near zero.

I have long said on this show that in a world where code is one-tenth as costly and difficult to create, we don't have one-tenth of the developers, we have ten times the amount of code. Indeed, my base case for all of artificial intelligence is that we will just live in a world of more. More software, more content, more everything.

This will, I think, inevitably change what people spend their time doing. Coursera founder Andrew Ng writes, writing software, especially prototypes, is becoming cheaper. This will lead to increased demand for people who can decide what to build. Basically, he's talking about the rise of product managers.

In their recent request for startups, Y Combinator wrote, Will agents kill the job of software developer? No, we'll need more human software engineers in the future because software is going to run almost everything. These humans won't write much code directly. Instead, they'll manage teams of agents that build software for them. What's happening then is that the barriers to entry for development are being structurally lowered. And while I do believe that this is going to be the most dramatic example that we've ever seen, we have seen this happen over and over again in software engineering.

In the late 70s, the release of the Apple II led to an explosion in hobbyist coders creating games in their bedrooms. As 3D engines became cheaper in the mid-2000s, we saw a proliferation of indie games. Mobile apps have become easier and easier to create over the years, and the numbers keep climbing. And this is to say nothing of the proliferation of user-generated content, as the cost of that type of production came down as well.

In short, none of the tools that previously made developing software easier to create was met by reduced demand for developers. They simply caused more software to be published and consumed. This, to me, feels like the apotheosis of that trend. Will Preble writes, "...cannot overstate how awesome Replit is. Idea to MVP in minutes or hours for virtually nothing. We are entering a new era of software development that is accessible to anyone with a phone and an idea."

In other words, to use the framing that's been everyone's favorite for the last week or so, although in a very different context, we are in the midst of a Javon's paradox for intelligence. Javon's paradox is the idea that when technology makes something more efficient, like using less fuel or energy, it can actually lead to more overall consumption, not less.

For example, if new technology makes cars more fuel-efficient, driving becomes cheaper. Instead of using less fuel overall, people might drive more, leading to an increase in total fuel consumption. The paradox shows that efficiency improvements don't always lead to conservation. In many cases, they encourage even more use of the resource. And that, I think, fairly perfectly describes what happens when we enter an age of intelligence that's too cheap to meter.

And I will point out that when it comes to coding specifically, the impact will not be limited to more people doing what software developers do now. It's going to be regular people who aren't software developers doing totally new things with software that wouldn't have been done before. When was the last time you saw a web application or a mobile app as the centerpiece of a major marketing campaign? Pretty rarely, right? Strong bet that as creating code becomes available in the marketer's toolkit, it becomes every bit as regular a choice as a YouTube video or a blog post.

So of all the barriers to entry crater, does this mean that all the developers and entrepreneurs have no future? First of all, it's important not to underestimate just how long change takes in practice. Mike at NicheDown on Twitter writes, AI coding tools like Cursor and Replit will make the world awash with software products, but I don't think it happens overnight. Most people are not persistent, any roadblocks at all, and they quit. Honestly, just having to set up a database and get API keys is a challenge for some, even though it's clicking and copy-paste.

These tools will improve, but right now it takes a lot of patience and troubleshooting to get things working. And the last mile of getting a great code output on text alone is a hard challenge to solve. Even beyond that, being able to create something is not the same as being able to create something great.

If it was, we'd have a lot more very successful YouTube channels, rather than just a tiny, tiny fraction of content creators out there being able to build a big audience from it. And for that person who was asking what advantage startup founders have, this gets back to that classic trope of everyone having a friend who said, oh, I had that idea first. If only I had done it, I'd be a billionaire. In 2018, Sam Altman was talking at Y Combinator, and he pointed out that startup founders, the good ones, don't just have an idea. They never stop having ideas.

He said, you can give a founder an idea and they can start a company. The problem is they need to come up with new ideas for the company basically every week. You have to come up with crazy new ideas and big changes all the time. We are at the very, very dawn of understanding what work and creation look like in the future. The only thing that is clear right now is that more people are going to have a totally expanded capacity to take the things that are in their brain and make them real in the real world.

It's very, very hard for me to see that as anything but just incredible potential. That's going to do it for today's AI Daily Brief. Appreciate you listening or watching, as always. And until next time, peace.