Get ONMINE

Stay Ahead, Stay ONMINE

The second wave of AI coding is here

Ask people building generative AI what generative AI is good for right now—what they’re really fired up about—and many will tell you: coding.  “That’s something that’s been very exciting for developers,” Jared Kaplan, chief scientist at Anthropic, told MIT Technology Review this month: “It’s really understanding what’s wrong with code, debugging it.” Copilot, a tool built on top of OpenAI’s large language models and launched by Microsoft-backed GitHub in 2022, is now used by millions of developers around the world. Millions more turn to general-purpose chatbots like Anthropic’s Claude, OpenAI’s ChatGPT, and Google DeepMind’s Gemini for everyday help. “Today, more than a quarter of all new code at Google is generated by AI, then reviewed and accepted by engineers,” Alphabet CEO Sundar Pichai claimed on an earnings call in October: “This helps our engineers do more and move faster.” Expect other tech companies to catch up, if they haven’t already. It’s not just the big beasts rolling out AI coding tools. A bunch of new startups have entered this buzzy market too. Newcomers such as Zencoder, Merly, Cosine, Tessl (valued at $750 million within months of being set up), and Poolside (valued at $3 billion before it even released a product) are all jostling for their slice of the pie. “It actually looks like developers are willing to pay for copilots,” says Nathan Benaich, an analyst at investment firm Air Street Capital: “And so code is one of the easiest ways to monetize AI.” Such companies promise to take generative coding assistants to the next level. Instead of providing developers with a kind of supercharged autocomplete, like most existing tools, this next generation can prototype, test, and debug code for you. The upshot is that developers could essentially turn into managers, who may spend more time reviewing and correcting code written by a model than writing it from scratch themselves.  But there’s more. Many of the people building generative coding assistants think that they could be a fast track to artificial general intelligence (AGI), the hypothetical superhuman technology that a number of top firms claim to have in their sights. “The first time we will see a massively economically valuable activity to have reached human-level capabilities will be in software development,” says Eiso Kant, CEO and cofounder of Poolside. (OpenAI has already boasted that its latest o3 model beat the company’s own chief scientist in a competitive coding challenge.) Welcome to the second wave of AI coding.  Correct code  Software engineers talk about two types of correctness. There’s the sense in which a program’s syntax (its grammar) is correct—meaning all the words, numbers, and mathematical operators are in the right place. This matters a lot more than grammatical correctness in natural language. Get one tiny thing wrong in thousands of lines of code and none of it will run. The first generation of coding assistants are now pretty good at producing code that’s correct in this sense. Trained on billions of pieces of code, they have assimilated the surface-level structures of many types of programs.   But there’s also the sense in which a program’s function is correct: Sure, it runs, but does it actually do what you wanted it to? It’s that second level of correctness that the new wave of generative coding assistants are aiming for—and this is what will really change the way software is made. “Large language models can write code that compiles, but they may not always write the program that you wanted,” says Alistair Pullen, a cofounder of Cosine. “To do that, you need to re-create the thought processes that a human coder would have gone through to get that end result.” The problem is that the data most coding assistants have been trained on—the billions of pieces of code taken from online repositories—doesn’t capture those thought processes. It represents a finished product, not what went into making it. “There’s a lot of code out there,” says Kant. “But that data doesn’t represent software development.” What Pullen, Kant, and others are finding is that to build a model that does a lot more than autocomplete—one that can come up with useful programs, test them, and fix bugs—you need to show it a lot more than just code. You need to show it how that code was put together.   In short, companies like Cosine and Poolside are building models that don’t just mimic what good code looks like—whether it works well or not—but mimic the process that produces such code in the first place. Get it right and the models will come up with far better code and far better bug fixes.  Breadcrumbs But you first need a data set that captures that process—the steps that a human developer might take when writing code. Think of these steps as a breadcrumb trail that a machine could follow to produce a similar piece of code itself. Part of that is working out what materials to draw from: Which sections of the existing codebase are needed for a given programming task? “Context is critical,” says Zencoder founder Andrew Filev. “The first generation of tools did a very poor job on the context, they would basically just look at your open tabs. But your repo [code repository] might have 5000 files and they’d miss most of it.” Zencoder has hired a bunch of search engine veterans to help it build a tool that can analyze large codebases and figure out what is and isn’t relevant. This detailed context reduces hallucinations and improves the quality of code that large language models can produce, says Filev: “We call it repo grokking.” Cosine also thinks context is key. But it draws on that context to create a new kind of data set. The company has asked dozens of coders to record what they were doing as they worked through hundreds of different programming tasks. “We asked them to write down everything,” says Pullen: “Why did you open that file? Why did you scroll halfway through? Why did you close it?” They also asked coders to annotate finished pieces of code, marking up sections that would have required knowledge of other pieces of code or specific documentation to write. Cosine then takes all that information and generates a large synthetic data set that maps the typical steps coders take, and the sources of information they draw on, to finished pieces of code. They use this data set to train a model to figure out what breadcrumb trail it might need to follow to produce a particular program, and then how to follow it.   Poolside, based in San Francisco, is also creating a synthetic data set that captures the process of coding, but it leans more on a technique called RLCE—reinforcement learning from code execution. (Cosine uses this too, but to a lesser degree.) RLCE is analogous to the technique used to make chatbots like ChatGPT slick conversationalists, known as RLHF—reinforcement learning from human feedback. With RLHF, a model is trained to produce text that’s more like the kind human testers say they favor. With RLCE, a model is trained to produce code that’s more like the kind that does what it is supposed to do when it is run (or executed).   Gaming the system Cosine and Poolside both say they are inspired by the approach DeepMind took with its game-playing model AlphaZero. AlphaZero was given the steps it could take—the moves in a game—and then left to play against itself over and over again, figuring out via trial and error what sequence of moves were winning moves and which were not.   “They let it explore moves at every possible turn, simulate as many games as you can throw compute at—that led all the way to beating Lee Sedol,” says Pengming Wang, a founding scientist at Poolside, referring to the Korean Go grandmaster that AlphaZero beat in 2016. Before Poolside, Wang worked at Google DeepMind on applications of AlphaZero beyond board games, including FunSearch, a version trained to solve advanced math problems. When that AlphaZero approach is applied to coding, the steps involved in producing a piece of code—the breadcrumbs—become the available moves in a game, and a correct program becomes winning that game. Left to play by itself, a model can improve far faster than a human could. “A human coder tries and fails one failure at a time,” says Kant. “Models can try things 100 times at once.” A key difference between Cosine and Poolside is that Cosine is using a custom version of GPT-4o provided by OpenAI, which makes it possible to train on a larger data set than the base model can cope with, but Poolside is building its own large language model from scratch. Poolside’s Kant thinks that training a model on code from the start will give better results than adapting an existing model that has sucked up not only billions of pieces of code but most of the internet. “I’m perfectly fine with our model forgetting about butterfly anatomy,” he says.   Cosine claims that its generative coding assistant, called Genie, tops the leaderboard on SWE-Bench, a standard set of tests for coding models. Poolside is still building its model but claims that what it has so far already matches the performance of GitHub’s Copilot. “I personally have a very strong belief that large language models will get us all the way to being as capable as a software developer,” says Kant. Not everyone takes that view, however. Illogical LLMs To Justin Gottschlich, the CEO and founder of Merly, large language models are the wrong tool for the job—period. He invokes his dog: “No amount of training for my dog will ever get him to be able to code, it just won’t happen,” he says. “He can do all kinds of other things, but he’s just incapable of that deep level of cognition.”   Having worked on code generation for more than a decade, Gottschlich has a similar sticking point with large language models. Programming requires the ability to work through logical puzzles with unwavering precision. No matter how well large language models may learn to mimic what human programmers do, at their core they are still essentially statistical slot machines, he says: “I can’t train an illogical system to become logical.” Instead of training a large language model to generate code by feeding it lots of examples, Merly does not show its system human-written code at all. That’s because to really build a model that can generate code, Gottschlich argues, you need to work at the level of the underlying logic that code represents, not the code itself. Merly’s system is therefore trained on an intermediate representation—something like the machine-readable notation that most programming languages get translated into before they are run. Gottschlich won’t say exactly what this looks like or how the process works. But he throws out an analogy: There’s this idea in mathematics that the only numbers that have to exist are prime numbers, because you can calculate all other numbers using just the primes. “Take that concept and apply it to code,” he says. Not only does this approach get straight to the logic of programming; it’s also fast, because millions of lines of code are reduced to a few thousand lines of intermediate language before the system analyzes them. Shifting mindsets What you think of these rival approaches may depend on what you want generative coding assistants to be.   In November, Cosine banned its engineers from using tools other than its own products. It is now seeing the impact of Genie on its own engineers, who often find themselves watching the tool as it comes up with code for them. “You now give the model the outcome you would like, and it goes ahead and worries about the implementation for you,” says Yang Li, another Cosine cofounder. Pullen admits that it can be baffling, requiring a switch of mindset. “We have engineers doing multiple tasks at once, flitting between windows,” he says. “While Genie is running code in one, they might be prompting it to do something else in another.” These tools also make it possible to protype multiple versions of a system at once. Say you’re developing software that needs a payment system built in. You can get a coding assistant to simultaneously try out several different options—Stripe, Mango, Checkout—instead of having to code them by hand one at a time. Genie can be left to fix bugs around the clock. Most software teams use bug-reporting tools that let people upload descriptions of errors they have encountered. Genie can read these descriptions and come up with fixes. Then a human just needs to review them before updating the code base. No single human understands the trillions of lines of code in today’s biggest software systems, says Li, “and as more and more software gets written by other software, the amount of code will only get bigger.” This will make coding assistants that maintain that code for us essential. “The bottleneck will become how fast humans can review the machine-generated code,” says Li. How do Cosine’s engineers feel about all this? According to Pullen, at least, just fine. “If I give you a hard problem, you’re still going to think about how you want to describe that problem to the model,” he says. “Instead of writing the code, you have to write it in natural language. But there’s still a lot of thinking that goes into that, so you’re not really taking the joy of engineering away. The itch is still scratched.” Some may adapt faster than others. Cosine likes to invite potential hires to spend a few days coding with its team. A couple of months ago it asked one such candidate to build a widget that would let employees share cool bits of software they were working on to social media.  The task wasn’t straightforward, requiring working knowledge of multiple sections of Cosine’s millions of lines of code. But the candidate got it done in a matter of hours. “This person who had never seen our code base turned up on Monday and by Tuesday afternoon he’d shipped something,” says Li. “We thought it would take him all week.” (They hired him.) But there’s another angle too. Many companies will use this technology to cut down on the number of programmers they hire. Li thinks we will soon see tiers of software engineers. At one end there will be elite developers with million-dollar salaries who can diagnose problems when the AI goes wrong. At the other end, smaller teams of 10 to 20 people will do a job that once required hundreds of coders. “It will be like how ATMs transformed banking,” says Li. “Anything you want to do will be determined by compute and not head count,” he says. “I think it’s generally accepted that the era of adding another few thousand engineers to your organization is over.” Warp drives Indeed, for Gottschlich, machines that can code better than humans are going to be essential. For him, that’s the only way we will build the vast, complex software systems that he thinks we will eventually need. Like many in Silicon Valley, he anticipates a future in which humans move to other planets. That’s only going to be possible if we get AI to build the software required, he says: “Merly’s real goal is to get us to Mars.” Gottschlich prefers to talk about “machine programming” rather than “coding assistants,” because he thinks that term frames the problem the wrong way. “I don’t think that these systems should be assisting humans—I think humans should be assisting them,” he says. “They can move at the speed of AI. Why restrict their potential?” “There’s this cartoon called The Flintstones where they have these cars, but they only move when the drivers use their feet,” says Gottschlich. “This is sort of how I feel most people are doing AI for software systems.” “But what Merly’s building is, essentially, spaceships,” he adds. He’s not joking. “And I don’t think spaceships should be powered by humans on a bicycle. Spaceships should be powered by a warp engine.” If that sounds wild—it is. But there’s a serious point to be made about what the people building this technology think the end goal really is. Gottschlich is not an outlier with his galaxy-brained take. Despite their focus on products that developers will want to use today, most of these companies have their sights on a far bigger payoff. Visit Cosine’s website and the company introduces itself as a “Human Reasoning Lab.” It sees coding as just the first step toward a more general-purpose model that can mimic human problem-solving in a number of domains. Poolside has similar goals: The company states upfront that it is building AGI. “Code is a way of formalizing reasoning,” says Kant. Wang invokes agents. Imagine a system that can spin up its own software to do any task on the fly, he says. “If you get to a point where your agent can really solve any computational task that you want through the means of software—that is a display of AGI, essentially.” Down here on Earth, such systems may remain a pipe dream. And yet software engineering is changing faster than many at the cutting edge expected.  “We’re not at a point where everything’s just done by machines, but we’re definitely stepping away from the usual role of a software engineer,” says Cosine’s Pullen. “We’re seeing the sparks of that new workflow—what it means to be a software engineer going into the future.”

Ask people building generative AI what generative AI is good for right now—what they’re really fired up about—and many will tell you: coding. 

“That’s something that’s been very exciting for developers,” Jared Kaplan, chief scientist at Anthropic, told MIT Technology Review this month: “It’s really understanding what’s wrong with code, debugging it.”

Copilot, a tool built on top of OpenAI’s large language models and launched by Microsoft-backed GitHub in 2022, is now used by millions of developers around the world. Millions more turn to general-purpose chatbots like Anthropic’s Claude, OpenAI’s ChatGPT, and Google DeepMind’s Gemini for everyday help.

“Today, more than a quarter of all new code at Google is generated by AI, then reviewed and accepted by engineers,” Alphabet CEO Sundar Pichai claimed on an earnings call in October: “This helps our engineers do more and move faster.” Expect other tech companies to catch up, if they haven’t already.

It’s not just the big beasts rolling out AI coding tools. A bunch of new startups have entered this buzzy market too. Newcomers such as Zencoder, Merly, Cosine, Tessl (valued at $750 million within months of being set up), and Poolside (valued at $3 billion before it even released a product) are all jostling for their slice of the pie. “It actually looks like developers are willing to pay for copilots,” says Nathan Benaich, an analyst at investment firm Air Street Capital: “And so code is one of the easiest ways to monetize AI.”

Such companies promise to take generative coding assistants to the next level. Instead of providing developers with a kind of supercharged autocomplete, like most existing tools, this next generation can prototype, test, and debug code for you. The upshot is that developers could essentially turn into managers, who may spend more time reviewing and correcting code written by a model than writing it from scratch themselves. 

But there’s more. Many of the people building generative coding assistants think that they could be a fast track to artificial general intelligence (AGI), the hypothetical superhuman technology that a number of top firms claim to have in their sights.

“The first time we will see a massively economically valuable activity to have reached human-level capabilities will be in software development,” says Eiso Kant, CEO and cofounder of Poolside. (OpenAI has already boasted that its latest o3 model beat the company’s own chief scientist in a competitive coding challenge.)

Welcome to the second wave of AI coding. 

Correct code 

Software engineers talk about two types of correctness. There’s the sense in which a program’s syntax (its grammar) is correct—meaning all the words, numbers, and mathematical operators are in the right place. This matters a lot more than grammatical correctness in natural language. Get one tiny thing wrong in thousands of lines of code and none of it will run.

The first generation of coding assistants are now pretty good at producing code that’s correct in this sense. Trained on billions of pieces of code, they have assimilated the surface-level structures of many types of programs.  

But there’s also the sense in which a program’s function is correct: Sure, it runs, but does it actually do what you wanted it to? It’s that second level of correctness that the new wave of generative coding assistants are aiming for—and this is what will really change the way software is made.

“Large language models can write code that compiles, but they may not always write the program that you wanted,” says Alistair Pullen, a cofounder of Cosine. “To do that, you need to re-create the thought processes that a human coder would have gone through to get that end result.”

The problem is that the data most coding assistants have been trained on—the billions of pieces of code taken from online repositories—doesn’t capture those thought processes. It represents a finished product, not what went into making it. “There’s a lot of code out there,” says Kant. “But that data doesn’t represent software development.”

What Pullen, Kant, and others are finding is that to build a model that does a lot more than autocomplete—one that can come up with useful programs, test them, and fix bugs—you need to show it a lot more than just code. You need to show it how that code was put together.  

In short, companies like Cosine and Poolside are building models that don’t just mimic what good code looks like—whether it works well or not—but mimic the process that produces such code in the first place. Get it right and the models will come up with far better code and far better bug fixes. 

Breadcrumbs

But you first need a data set that captures that process—the steps that a human developer might take when writing code. Think of these steps as a breadcrumb trail that a machine could follow to produce a similar piece of code itself.

Part of that is working out what materials to draw from: Which sections of the existing codebase are needed for a given programming task? “Context is critical,” says Zencoder founder Andrew Filev. “The first generation of tools did a very poor job on the context, they would basically just look at your open tabs. But your repo [code repository] might have 5000 files and they’d miss most of it.”

Zencoder has hired a bunch of search engine veterans to help it build a tool that can analyze large codebases and figure out what is and isn’t relevant. This detailed context reduces hallucinations and improves the quality of code that large language models can produce, says Filev: “We call it repo grokking.”

Cosine also thinks context is key. But it draws on that context to create a new kind of data set. The company has asked dozens of coders to record what they were doing as they worked through hundreds of different programming tasks. “We asked them to write down everything,” says Pullen: “Why did you open that file? Why did you scroll halfway through? Why did you close it?” They also asked coders to annotate finished pieces of code, marking up sections that would have required knowledge of other pieces of code or specific documentation to write.

Cosine then takes all that information and generates a large synthetic data set that maps the typical steps coders take, and the sources of information they draw on, to finished pieces of code. They use this data set to train a model to figure out what breadcrumb trail it might need to follow to produce a particular program, and then how to follow it.  

Poolside, based in San Francisco, is also creating a synthetic data set that captures the process of coding, but it leans more on a technique called RLCE—reinforcement learning from code execution. (Cosine uses this too, but to a lesser degree.)

RLCE is analogous to the technique used to make chatbots like ChatGPT slick conversationalists, known as RLHF—reinforcement learning from human feedback. With RLHF, a model is trained to produce text that’s more like the kind human testers say they favor. With RLCE, a model is trained to produce code that’s more like the kind that does what it is supposed to do when it is run (or executed).  

Gaming the system

Cosine and Poolside both say they are inspired by the approach DeepMind took with its game-playing model AlphaZero. AlphaZero was given the steps it could take—the moves in a game—and then left to play against itself over and over again, figuring out via trial and error what sequence of moves were winning moves and which were not.  

“They let it explore moves at every possible turn, simulate as many games as you can throw compute at—that led all the way to beating Lee Sedol,” says Pengming Wang, a founding scientist at Poolside, referring to the Korean Go grandmaster that AlphaZero beat in 2016. Before Poolside, Wang worked at Google DeepMind on applications of AlphaZero beyond board games, including FunSearch, a version trained to solve advanced math problems.

When that AlphaZero approach is applied to coding, the steps involved in producing a piece of code—the breadcrumbs—become the available moves in a game, and a correct program becomes winning that game. Left to play by itself, a model can improve far faster than a human could. “A human coder tries and fails one failure at a time,” says Kant. “Models can try things 100 times at once.”

A key difference between Cosine and Poolside is that Cosine is using a custom version of GPT-4o provided by OpenAI, which makes it possible to train on a larger data set than the base model can cope with, but Poolside is building its own large language model from scratch.

Poolside’s Kant thinks that training a model on code from the start will give better results than adapting an existing model that has sucked up not only billions of pieces of code but most of the internet. “I’m perfectly fine with our model forgetting about butterfly anatomy,” he says.  

Cosine claims that its generative coding assistant, called Genie, tops the leaderboard on SWE-Bench, a standard set of tests for coding models. Poolside is still building its model but claims that what it has so far already matches the performance of GitHub’s Copilot.

“I personally have a very strong belief that large language models will get us all the way to being as capable as a software developer,” says Kant.

Not everyone takes that view, however.

Illogical LLMs

To Justin Gottschlich, the CEO and founder of Merly, large language models are the wrong tool for the job—period. He invokes his dog: “No amount of training for my dog will ever get him to be able to code, it just won’t happen,” he says. “He can do all kinds of other things, but he’s just incapable of that deep level of cognition.”  

Having worked on code generation for more than a decade, Gottschlich has a similar sticking point with large language models. Programming requires the ability to work through logical puzzles with unwavering precision. No matter how well large language models may learn to mimic what human programmers do, at their core they are still essentially statistical slot machines, he says: “I can’t train an illogical system to become logical.”

Instead of training a large language model to generate code by feeding it lots of examples, Merly does not show its system human-written code at all. That’s because to really build a model that can generate code, Gottschlich argues, you need to work at the level of the underlying logic that code represents, not the code itself. Merly’s system is therefore trained on an intermediate representation—something like the machine-readable notation that most programming languages get translated into before they are run.

Gottschlich won’t say exactly what this looks like or how the process works. But he throws out an analogy: There’s this idea in mathematics that the only numbers that have to exist are prime numbers, because you can calculate all other numbers using just the primes. “Take that concept and apply it to code,” he says.

Not only does this approach get straight to the logic of programming; it’s also fast, because millions of lines of code are reduced to a few thousand lines of intermediate language before the system analyzes them.

Shifting mindsets

What you think of these rival approaches may depend on what you want generative coding assistants to be.  

In November, Cosine banned its engineers from using tools other than its own products. It is now seeing the impact of Genie on its own engineers, who often find themselves watching the tool as it comes up with code for them. “You now give the model the outcome you would like, and it goes ahead and worries about the implementation for you,” says Yang Li, another Cosine cofounder.

Pullen admits that it can be baffling, requiring a switch of mindset. “We have engineers doing multiple tasks at once, flitting between windows,” he says. “While Genie is running code in one, they might be prompting it to do something else in another.”

These tools also make it possible to protype multiple versions of a system at once. Say you’re developing software that needs a payment system built in. You can get a coding assistant to simultaneously try out several different options—Stripe, Mango, Checkout—instead of having to code them by hand one at a time.

Genie can be left to fix bugs around the clock. Most software teams use bug-reporting tools that let people upload descriptions of errors they have encountered. Genie can read these descriptions and come up with fixes. Then a human just needs to review them before updating the code base.

No single human understands the trillions of lines of code in today’s biggest software systems, says Li, “and as more and more software gets written by other software, the amount of code will only get bigger.”

This will make coding assistants that maintain that code for us essential. “The bottleneck will become how fast humans can review the machine-generated code,” says Li.

How do Cosine’s engineers feel about all this? According to Pullen, at least, just fine. “If I give you a hard problem, you’re still going to think about how you want to describe that problem to the model,” he says. “Instead of writing the code, you have to write it in natural language. But there’s still a lot of thinking that goes into that, so you’re not really taking the joy of engineering away. The itch is still scratched.”

Some may adapt faster than others. Cosine likes to invite potential hires to spend a few days coding with its team. A couple of months ago it asked one such candidate to build a widget that would let employees share cool bits of software they were working on to social media. 

The task wasn’t straightforward, requiring working knowledge of multiple sections of Cosine’s millions of lines of code. But the candidate got it done in a matter of hours. “This person who had never seen our code base turned up on Monday and by Tuesday afternoon he’d shipped something,” says Li. “We thought it would take him all week.” (They hired him.)

But there’s another angle too. Many companies will use this technology to cut down on the number of programmers they hire. Li thinks we will soon see tiers of software engineers. At one end there will be elite developers with million-dollar salaries who can diagnose problems when the AI goes wrong. At the other end, smaller teams of 10 to 20 people will do a job that once required hundreds of coders. “It will be like how ATMs transformed banking,” says Li.

“Anything you want to do will be determined by compute and not head count,” he says. “I think it’s generally accepted that the era of adding another few thousand engineers to your organization is over.”

Warp drives

Indeed, for Gottschlich, machines that can code better than humans are going to be essential. For him, that’s the only way we will build the vast, complex software systems that he thinks we will eventually need. Like many in Silicon Valley, he anticipates a future in which humans move to other planets. That’s only going to be possible if we get AI to build the software required, he says: “Merly’s real goal is to get us to Mars.”

Gottschlich prefers to talk about “machine programming” rather than “coding assistants,” because he thinks that term frames the problem the wrong way. “I don’t think that these systems should be assisting humans—I think humans should be assisting them,” he says. “They can move at the speed of AI. Why restrict their potential?”

“There’s this cartoon called The Flintstones where they have these cars, but they only move when the drivers use their feet,” says Gottschlich. “This is sort of how I feel most people are doing AI for software systems.”

“But what Merly’s building is, essentially, spaceships,” he adds. He’s not joking. “And I don’t think spaceships should be powered by humans on a bicycle. Spaceships should be powered by a warp engine.”

If that sounds wild—it is. But there’s a serious point to be made about what the people building this technology think the end goal really is.

Gottschlich is not an outlier with his galaxy-brained take. Despite their focus on products that developers will want to use today, most of these companies have their sights on a far bigger payoff. Visit Cosine’s website and the company introduces itself as a “Human Reasoning Lab.” It sees coding as just the first step toward a more general-purpose model that can mimic human problem-solving in a number of domains.

Poolside has similar goals: The company states upfront that it is building AGI. “Code is a way of formalizing reasoning,” says Kant.

Wang invokes agents. Imagine a system that can spin up its own software to do any task on the fly, he says. “If you get to a point where your agent can really solve any computational task that you want through the means of software—that is a display of AGI, essentially.”

Down here on Earth, such systems may remain a pipe dream. And yet software engineering is changing faster than many at the cutting edge expected. 

“We’re not at a point where everything’s just done by machines, but we’re definitely stepping away from the usual role of a software engineer,” says Cosine’s Pullen. “We’re seeing the sparks of that new workflow—what it means to be a software engineer going into the future.”

Shape
Shape
Stay Ahead

Explore More Insights

More Articles

Stay ahead with more perspectives on cutting-edge power, infrastructure, energy,  bitcoin and AI solutions. Explore these articles to uncover strategies and insights shaping the future of industries.

Shape

How network diversity protects utility operations in an evolving landscape

From redundancy to resilience: The next level of network diversity Despite billions invested in grid modernization and widespread adoption of dual-carrier approaches, many utilities still find their communications infrastructure vulnerable to disruption. It’s a paradox: utilities have engineered redundancy across power generation and distribution, yet communications—the digital backbone of the modern grid—often lacks the comprehensive diversity needed to truly eliminate single points of failure.  True grid resilience now depends on comprehensive network diversity that enables consistent quality of service, operational flexibility and long-term adaptability. The cost of connectivity blind spots As utilities expand advanced metering infrastructure (AMI), SCADA systems, distributed energy resource (DER) management and connected field operations, the strain on communications networks continues to grow. Relying on a single technology exposes utilities to significant risk—from natural disasters and carrier outages to rural coverage gaps and network congestion. The result: delayed outage restoration, compromised grid visibility and operational inefficiencies that undermine reliability and customer trust. The versatility imperative: Why utilities can’t just “upgrade” Utility communications can’t follow the consumer tech cycle. Equipment often remains in service for 20 to 30 years, even as wireless standards evolve from private radio networks to LTE, 4G, 5G and emerging spectrum such as Anterix Band 106. Each shift brings new performance expectations—but utilities can’t afford to rip and replace infrastructure every few years. Instead, they need adaptable, field-proven solutions that bridge technologies and spectrum generations, ensuring consistent, secure connectivity today while accommodating tomorrow’s innovations without interrupting critical operations. Three pillars of network diversity While many utilities have already adopted dual-carrier strategies, the expanding complexity of modern grid operations demands a renewed look at comprehensive network diversity. True resilience comes from a multi-path approach—one that enables seamless communication under any condition. Three key strategies define this new standard of reliability. A. Public-to-public redundancy Dual-radio routers

Read More »

Why conductor strength matters for grid reliability

As utilities work to strengthen and modernize America’s electric grid, they face growing mechanical and environmental challenges below and between the lines. Buried grounding networks, pole grounds and substation grids must all withstand decades of stress from soil movement, moisture, corrosion and fault current events. Each of these physical forces can compromise a system’s electrical integrity — making mechanical strength as vital as electrical performance in ensuring long-term reliability. In earlier decades, utility conductors were relatively short, stationary and installed in stable soil. Today’s infrastructure is different. Expansion into remote terrain, widespread undergrounding and the integration of renewable and distributed resources have multiplied the number of grounding paths and exposed more cable to movement, vibration and stress. These systems must remain reliable through decades of shifting soils, thermal cycling and fault events — all while supporting uninterrupted power delivery. When a grounding conductor fails, the results can be costly. Broken bonds or weakened terminations can increase ground resistance, trigger equipment faults or leave assets unprotected from lightning and surge events. Repairs often require excavation, downtime and new material — expenses that compound across large service territories. In short, mechanical failure doesn’t just compromise safety; it undermines reliability, budgets and public confidence. That’s why conductor strength has become a defining factor in grid resilience. Copper-Clad Steel (CCS) conductors are engineered to meet this demand. By metallurgically bonding a copper layer to a high-tensile-strength steel core, CCS combines the conductivity of copper with the durability of steel. The result is a grounding conductor that resists stretching, breakage and deformation while maintaining long-term electrical integrity. Unlike soft copper, which can elongate or fracture under mechanical strain, CCS retains its shape and strength even after repeated mechanical or thermal stress. That strength translates into reliability you can measure. Stronger conductors stay tight at terminations, maintain

Read More »

Russian Crude Output Rose Last Month

Russia’s crude oil production edged up in October, but remained below its OPEC+ quota as international pressure mounted on the country’s energy sector. Russia pumped an average 9.411 million barrels a day last month, people with knowledge of the data said, asking not to be identified discussing confidential information. While that’s 43,000 barrels a day higher than in September, it’s 70,000 a day below a quota that includes compensation cuts for previous overproduction, Bloomberg calculations show. Oil watchers are closely following Russian production data to assess the impact of sanctions — and Ukrainian drone strikes — against the country’s energy industry. The latest US penalties on the sector, which hit oil giants Rosneft PJSC and Lukoil PJSC, have already eroded crude exports as some refiners in India, China and Turkey prove less willing to take sanctioned barrels. Meanwhile, Ukrainian attacks have intensified, putting pressure on Russia’s crude-processing sector even as refinery owners rush to repair infrastructure.  If Moscow eventually finds itself unable to find buyers for oil from its sanctioned producers, and struggles to restore refining, it’ll be forced to halt output at some fields, risking damage to wells. The Energy Ministry didn’t immediately respond to a request for comment on the production data. Deputy Prime Minister Alexander Novak said last month that the nation has capacity to raise oil production further, but will do it gradually, according to Tass news service. Compensation Cuts Russia, historically one of the biggest laggards in complying with OPEC+ output agreements, has agreed to make additional cuts to compensate for previous overproduction. The monthly schedule for those curbs has been regularly revised, with the latest plan published earlier this month.  It shows that October was the last month when Russia had to make such cuts. Moscow’s pledge to reduce daily output by 10,000 barrels below a quota of 9.491

Read More »

Oil Rises but Logs Second Weekly Loss

Oil rose on Friday but still notched a second weekly loss as the market continued to weigh the threat to output from sanctions on Russia against a looming oversupply. West Texas Intermediate futures rose around 0.5% to settle below $60 a barrel, but were still down for the week. Adding to fears of a glut, oil prices have also been buffeted by swings in equity markets this week. Meanwhile, the White House’s move to clamp down on the buying of Russian crude led oil trading giant Gunvor Group to withdraw an offer for the international assets of Lukoil PJSC. The fate of the assets, which include stakes in oil fields, refineries and gas stations, remains unclear. One possible exception to that crackdown could emerge soon: President Donald Trump signaled an openness to exempting Hungary from sanctions on Russian energy purchases as he hosted Prime Minister Viktor Orban, briefly pushing futures to intraday lows. The development appeared to allay shortage fears, given that Budapest imports over 90% of its crude from Moscow. Senior industry figures have warned the latest US curbs on Russia’s two largest oil companies are beginning to have an impact on the market, particularly in diesel, where prices have been surging in recent days, with time spreads for the fuel signaling supply pressure. At the same time, the US measures have come against a backdrop of oversupply that has weighed on key crude oil metrics. The spread between the nearest West Texas Intermediate futures closed at the weakest level since February on Thursday. “If the market flips to contango, we may see more bearish funds enter the crude space,” said Dennis Kissler, senior vice president for trading at BOK Financial said of the potential that longer-dated contracts trade at a premium to nearer-term ones. “Most traders remain surprised

Read More »

Gunvor Scraps Lukoil Deal

Commodity trader Gunvor Group has withdrawn its offer for the international assets of sanctioned Russian oil producer Lukoil PJSC after the US Treasury Department called it “the Kremlin’s puppet” and said the oil and gas trader would never get a license. Gunvor pushed back on the Treasury comment on social media, calling it “fundamentally misinformed and false.” The Geneva-based company said it would seek to correct a “clear misunderstanding” but that it would withdraw its bid for now. President Trump has been clear that the war must end immediately. As long as Putin continues the senseless killings, the Kremlin’s puppet, Gunvor, will never get a license to operate and profit. — Treasury Department (@USTreasury) November 6, 2025 The comment is a remarkable volte-face after a week in which Gunvor has been in talks with the US Office of Foreign Assets Control, part of the Treasury Department, and other bodies in charge of sanctions to help press its case for a deal that would have transformed it into an integrated oil producing and processing colossus. Gunvor swooped on the assets at the end of last month following the US blacklisting of Lukoil and fellow Russian oil giant Rosneft PJSC, and its exit may leave the door open to other suitors. Gunvor on Thursday also announced it had raised $2.81 billion in a credit facility financed by US arms of global banks. Like other major commodity traders, the firm funds the bulk of its trades of oil, gas and metals around the world with bank financing. For the trader, the comments are likely to revive questions about its connections in Moscow at a time when many oil industry participants are wary of any links to Russia.  The trader’s co-founder, Gennady Timchenko, is a friend of Russian President Vladimir Putin, and when the US imposed sanctions

Read More »

Ship With Russia Oil Makes Rare Move Offshore India

A tanker carrying crude from recently-sanctioned Rosneft PJSC has made a rare cargo transfer off Mumbai, as the Trump administration ramps up its scrutiny of India’s oil trade with Russia. But the unusual move has puzzled traders. The cargo was transferred from one blacklisted tanker to another sanctioned ship, meaning there’s been no attempt to hide its origin — typical of such a move — and the crude is still heading for an Indian port: Kochi in the south, rather than Mumbai on the west coast. India’s purchases of Russian oil have drawn the ire of President Donald Trump, and the US penalties on Rosneft along with Lukoil PJSC are expected to severely impact the trade. The market is keenly watching for disruptions to established flows before a grace period related to the sanctions ends later this month. “What we’re seeing now is this uncertainty in the market about what the sanctions risks are,” said Rachel Ziemba, an analyst at the Center for a New American Security in Washington. “The net result is more ship-to-ship transfers, more subterfuge, longer routes, more complicated transactions.” The Fortis took around 720,000 barrels of Russian Urals from Ailana on Tuesday near Mumbai, according to ship-tracking data compiled by Bloomberg, Kpler and Vortexa. The cargo was collected from the Baltic port of Ust-Luga before the US sanctioned Rosneft, and Ailana had idled in the area for nearly two weeks with no clear reason.  Ailana is on its way back to Russia, while Fortis is expected to arrive at Kochi early next week with the cargo, ship-tracking data shows. Both vessels have been sanctioned by the European Union and the UK. Fortis’ owner and manager — Vietnam-based Pacific Logistic & Maritime and North Star Ship Management — didn’t respond to emailed requests for comment. There are no contact details on maritime database

Read More »

Designing the AI Century: 7×24 Exchange Fall ’25 Charts the New Data Center Industrial Stack

SMRs and the AI Power Gap: Steve Fairfax Separates Promise from Physics If NVIDIA’s Sean Young made the case for AI factories, Steve Fairfax offered a sobering counterweight: even the smartest factories can’t run without power—and not just any power, but constant, high-availability, clean generation at a scale utilities are increasingly struggling to deliver. In his keynote “Small Modular Reactors for Data Centers,” Fairfax, president of Oresme and one of the data center industry’s most seasoned voices on reliability, walked through the long arc from nuclear fusion research to today’s resurgent interest in fission at modular scale. His presentation blended nuclear engineering history with pragmatic counsel for AI-era infrastructure leaders: SMRs are promising, but their road to reality is paved with physics, fuel, and policy—not PowerPoint. From Fusion Research to Data Center Reliability Fairfax began with his own story—a career that bridges nuclear reliability and data center engineering. As a young physicist and electrical engineer at MIT, he helped build the Alcator C-MOD fusion reactor, a 400-megawatt research facility that heated plasma to 100 million degrees with 3 million amps of current. The magnet system alone drew 265,000 amps at 1,400 volts, producing forces measured in millions of pounds. It was an extreme experiment in controlled power, and one that shaped his later philosophy: design for failure, test for truth, and assume nothing lasts forever. When the U.S. cooled on fusion power in the 1990s, Fairfax applied nuclear reliability methods to data center systems—quantifying uptime and redundancy with the same math used for reactor safety. By 1994, he was consulting for hyperscale pioneers still calling 10 MW “monstrous.” Today’s 400 MW campuses, he noted, are beginning to look a lot more like reactors in their energy intensity—and increasingly, in their regulatory scrutiny. Defining the Small Modular Reactor Fairfax defined SMRs

Read More »

Top network and data center events 2025 & 2026

Denise Dubie is a senior editor at Network World with nearly 30 years of experience writing about the tech industry. Her coverage areas include AIOps, cybersecurity, networking careers, network management, observability, SASE, SD-WAN, and how AI transforms enterprise IT. A seasoned journalist and content creator, Denise writes breaking news and in-depth features, and she delivers practical advice for IT professionals while making complex technology accessible to all. Before returning to journalism, she held senior content marketing roles at CA Technologies, Berkshire Grey, and Cisco. Denise is a trusted voice in the world of enterprise IT and networking.

Read More »

Google’s cheaper, faster TPUs are here, while users of other AI processors face a supply crunch

Opportunities for the AI industry LLM vendors such as OpenAI and Anthropic, which still have relatively young code bases and are continuously evolving them, also have much to gain from the arrival of Ironwood for training their models, said Forrester vice president and principal analyst Charlie Dai. In fact, Anthropic has already agreed to procure 1 million TPUs for training and its models and using them for inferencing. Other, smaller vendors using Google’s TPUs for training models include Lightricks and Essential AI. Google has seen a steady increase in demand for its TPUs (which it also uses to run interna services), and is expected to buy $9.8 billion worth of TPUs from Broadcom this year, compared to $6.2 billion and $2.04 billion in 2024 and 2023 respectively, according to Harrowell. “This makes them the second-biggest AI chip program for cloud and enterprise data centers, just tailing Nvidia, with approximately 5% of the market. Nvidia owns about 78% of the market,” Harrowell said. The legacy problem While some analysts were optimistic about the prospects for TPUs in the enterprise, IDC research director Brandon Hoff said enterprises will most likely to stay away from Ironwood or TPUs in general because of their existing code base written for other platforms. “For enterprise customers who are writing their own inferencing, they will be tied into Nvidia’s software platform,” Hoff said, referring to CUDA, the software platform that runs on Nvidia GPUs. CUDA was released to the public in 2007, while the first version of TensorFlow has only been around since 2015.

Read More »

Cisco launches AI infrastructure, AI practitioner certifications

“This new certification focuses on artificial intelligence and machine learning workloads, helping technical professionals become AI-ready and successfully embed AI into their workflows,” said Pat Merat, vice president at Learn with Cisco, in a blog detailing the new AI Infrastructure Specialist certification. “The certification validates a candidate’s comprehensive knowledge in designing, implementing, operating, and troubleshooting AI solutions across Cisco infrastructure.” Separately, the AITECH certification is part of the Cisco AI Infrastructure track, which complements its existing networking, data center, and security certifications. Cisco says the AITECH cert training is intended for network engineers, system administrators, solution architects, and other IT professionals who want to learn how AI impacts enterprise infrastructure. The training curriculum covers topics such as: Utilizing AI for code generation, refactoring, and using modern AI-assisted coding workflows. Using generative AI for exploratory data analysis, data cleaning, transformation, and generating actionable insights. Designing and implementing multi-step AI-assisted workflows and understanding complex agentic systems for automation. Learning AI-powered requirements, evaluating customization approaches, considering deployment strategies, and designing robust AI workflows. Evaluating, fine-tuning, and deploying pre-trained AI models, and implementing Retrieval Augmented Generation (RAG) systems. Monitoring, maintaining, and optimizing AI-powered workflows, ensuring data integrity and security. AITECH certification candidates will learn how to use AI to enhance productivity, automate routine tasks, and support the development of new applications. The training program includes hands-on labs and simulations to demonstrate practical use cases for AI within Cisco and multi-vendor environments.

Read More »

Chip-to-Grid Gets Bought: Eaton, Vertiv, and Daikin Deals Imply a New Thermal Capital Cycle

This week delivered three telling acquisitions that mark a turning point for the global data center supply chain; and more specifically, for the high-density liquid cooling mega-play now unfolding across the power-thermal continuum. Eaton is acquiring Boyd Thermal for $9.5 billion from Goldman Sachs Asset Management. Vertiv is buying PurgeRite for about $1 billion from Milton Street Capital. And Daikin Applied has moved to acquire Chilldyne, one of the most proven negative-pressure direct-to-chip pioneers. On paper, they’re three distinct transactions. In reality, they’re chapters in the same story: the acceleration of strategic vertical integration around thermal infrastructure for AI-class compute. The Equity Layer: Private Capital Builds, Strategics Buy From an equity standpoint, these are classic handoff moments between private-equity construction and corporate consolidation. Goldman Sachs built Boyd Thermal into a global platform spanning cold plates, CDUs, and high-density liquid loop design, now sold to Eaton at an enterprise multiple north of 5× 2026E revenue. Milton Street Capital took PurgeRite from a specialist contractor in fluid flushing and commissioning into a nationwide services platform. And Daikin, long synonymous with chillers and air-side thermal, is crossing the liquid Rubicon by buying its way into the D2C ecosystem. Each deal crystallizes a simple fact: liquid cooling is no longer an adjunct; it’s core infrastructure. Private equity did its job scaling the parts. Strategic players are now paying up for the system. Eaton’s Bid: The Chip-to-Grid Thesis For Eaton, Boyd Thermal is the final missing piece in its “chip-to-grid” thesis. The company already owns the electrical side of the data center: UPS, busway, switchgear, and monitoring. Boyd plugs the thermal gap, allowing Eaton to market full rack-to-substation solutions for AI loads in the 50–100 kW+ range. It’s a statement acquisition that places Eaton squarely against Schneider Electric, Vertiv and ABB in the race to

Read More »

Space: The final frontier for data processing

There are, however, a couple of reasons why data centers in space are being considered. There are plenty of reports about how the increased amount of AI processing is affecting power consumption within data centers; the World Economic Forum has estimated that the power required to handle AI is increasing at a rate of between 26% and 36% annually. Therefore, it is not surprising that organizations are looking at other options. But an even more pressing reason for orbiting data centers is to handle the amount of data that is being produced by existing satellites, Judge said. “Essentially, satellites are gathering a lot more data than can be sent to earth, because downlinks are a bottleneck,” he noted. “With AI capacity in orbit, they could potentially analyze more of this data, extract more useful information, and send insights back to earth. My overall feeling is that any more data processing in space is going to be driven by space processing needs.” And China may already be ahead of the game. Last year, Guoxing Aerospace  launched 12 satellites, forming a space-based computing network dubbed the Three-Body Computing Constellation. When completed, it will contain 2,800 satellites, all handling the orchestration and processing of data, taking edge computing to a new dimension.

Read More »

Microsoft will invest $80B in AI data centers in fiscal 2025

And Microsoft isn’t the only one that is ramping up its investments into AI-enabled data centers. Rival cloud service providers are all investing in either upgrading or opening new data centers to capture a larger chunk of business from developers and users of large language models (LLMs).  In a report published in October 2024, Bloomberg Intelligence estimated that demand for generative AI would push Microsoft, AWS, Google, Oracle, Meta, and Apple would between them devote $200 billion to capex in 2025, up from $110 billion in 2023. Microsoft is one of the biggest spenders, followed closely by Google and AWS, Bloomberg Intelligence said. Its estimate of Microsoft’s capital spending on AI, at $62.4 billion for calendar 2025, is lower than Smith’s claim that the company will invest $80 billion in the fiscal year to June 30, 2025. Both figures, though, are way higher than Microsoft’s 2020 capital expenditure of “just” $17.6 billion. The majority of the increased spending is tied to cloud services and the expansion of AI infrastructure needed to provide compute capacity for OpenAI workloads. Separately, last October Amazon CEO Andy Jassy said his company planned total capex spend of $75 billion in 2024 and even more in 2025, with much of it going to AWS, its cloud computing division.

Read More »

John Deere unveils more autonomous farm machines to address skill labor shortage

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Self-driving tractors might be the path to self-driving cars. John Deere has revealed a new line of autonomous machines and tech across agriculture, construction and commercial landscaping. The Moline, Illinois-based John Deere has been in business for 187 years, yet it’s been a regular as a non-tech company showing off technology at the big tech trade show in Las Vegas and is back at CES 2025 with more autonomous tractors and other vehicles. This is not something we usually cover, but John Deere has a lot of data that is interesting in the big picture of tech. The message from the company is that there aren’t enough skilled farm laborers to do the work that its customers need. It’s been a challenge for most of the last two decades, said Jahmy Hindman, CTO at John Deere, in a briefing. Much of the tech will come this fall and after that. He noted that the average farmer in the U.S. is over 58 and works 12 to 18 hours a day to grow food for us. And he said the American Farm Bureau Federation estimates there are roughly 2.4 million farm jobs that need to be filled annually; and the agricultural work force continues to shrink. (This is my hint to the anti-immigration crowd). John Deere’s autonomous 9RX Tractor. Farmers can oversee it using an app. While each of these industries experiences their own set of challenges, a commonality across all is skilled labor availability. In construction, about 80% percent of contractors struggle to find skilled labor. And in commercial landscaping, 86% of landscaping business owners can’t find labor to fill open positions, he said. “They have to figure out how to do

Read More »

2025 playbook for enterprise AI success, from agents to evals

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More 2025 is poised to be a pivotal year for enterprise AI. The past year has seen rapid innovation, and this year will see the same. This has made it more critical than ever to revisit your AI strategy to stay competitive and create value for your customers. From scaling AI agents to optimizing costs, here are the five critical areas enterprises should prioritize for their AI strategy this year. 1. Agents: the next generation of automation AI agents are no longer theoretical. In 2025, they’re indispensable tools for enterprises looking to streamline operations and enhance customer interactions. Unlike traditional software, agents powered by large language models (LLMs) can make nuanced decisions, navigate complex multi-step tasks, and integrate seamlessly with tools and APIs. At the start of 2024, agents were not ready for prime time, making frustrating mistakes like hallucinating URLs. They started getting better as frontier large language models themselves improved. “Let me put it this way,” said Sam Witteveen, cofounder of Red Dragon, a company that develops agents for companies, and that recently reviewed the 48 agents it built last year. “Interestingly, the ones that we built at the start of the year, a lot of those worked way better at the end of the year just because the models got better.” Witteveen shared this in the video podcast we filmed to discuss these five big trends in detail. Models are getting better and hallucinating less, and they’re also being trained to do agentic tasks. Another feature that the model providers are researching is a way to use the LLM as a judge, and as models get cheaper (something we’ll cover below), companies can use three or more models to

Read More »

OpenAI’s red teaming innovations define new essentials for security leaders in the AI era

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More OpenAI has taken a more aggressive approach to red teaming than its AI competitors, demonstrating its security teams’ advanced capabilities in two areas: multi-step reinforcement and external red teaming. OpenAI recently released two papers that set a new competitive standard for improving the quality, reliability and safety of AI models in these two techniques and more. The first paper, “OpenAI’s Approach to External Red Teaming for AI Models and Systems,” reports that specialized teams outside the company have proven effective in uncovering vulnerabilities that might otherwise have made it into a released model because in-house testing techniques may have missed them. In the second paper, “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning,” OpenAI introduces an automated framework that relies on iterative reinforcement learning to generate a broad spectrum of novel, wide-ranging attacks. Going all-in on red teaming pays practical, competitive dividends It’s encouraging to see competitive intensity in red teaming growing among AI companies. When Anthropic released its AI red team guidelines in June of last year, it joined AI providers including Google, Microsoft, Nvidia, OpenAI, and even the U.S.’s National Institute of Standards and Technology (NIST), which all had released red teaming frameworks. Investing heavily in red teaming yields tangible benefits for security leaders in any organization. OpenAI’s paper on external red teaming provides a detailed analysis of how the company strives to create specialized external teams that include cybersecurity and subject matter experts. The goal is to see if knowledgeable external teams can defeat models’ security perimeters and find gaps in their security, biases and controls that prompt-based testing couldn’t find. What makes OpenAI’s recent papers noteworthy is how well they define using human-in-the-middle

Read More »
More Articles