Anthropic just released Claude Sonnet 4.5, and the company is billing it as nothing less than the best coding model in the world.
This new AI model can tackle complex, multi-step engineering tasks, from building entire applications to managing databases. In one stunning demo, it generated 11,000 lines of code to create a Slack-style chat app, only stopping when the job was complete.
Anthropic claims that, in practice, the model can maintain focus for more than 30 hours on a single complex task.
But while the coding prowess is impressive, the real story lies in what this signals about the future of AI development, and which markets the AI labs are really after.
To break down what this release means, I talked it through with SmarterX and Marketing AI Institute founder and CEO Paul Roetzer on Episode 172 of The Artificial Intelligence Show.
First, it’s important to understand how Anthropic’s models are structured. Haiku is their smallest model, Sonnet is the mid-tier, and Opus is the largest and most powerful. But with this new release, something interesting happened: the mid-tier Sonnet 4.5 is now outperforming their top-tier Opus model.
According to Roetzer, this reveals a new pattern in the industry. An AI lab will perform a massive, expensive training run to create a frontier model like Opus. Then, just three to six months later, they can release a more efficient, affordable model like Sonnet that—through fine-tuning and reinforcement learning—is actually smarter than its predecessor.
“This is what's going to happen every three to six months,” Roetzer says. “Basically, you do a massive training run, then can do a much more affordable, efficient model like Sonnet and make it smarter than the big run they just did.”
And for anyone thinking AI development is about to hit a wall, the researchers on the front lines have a different message.
“He's like, we're not seeing it,” says Roetzer, referencing comments on a recent podcast from Anthropic AI researcher Sholto Douglas. “There's nothing we're seeing that tells us there's any wall whatsoever, that these things aren't going to just keep getting smarter and more generally capable.”
Anthropic’s intense focus on building an AI model that codes better than any other in the world isn’t an accident. Roetzer explains that it’s a twofold strategy.
First, the company believes the fastest path to more powerful AI is by automating the work of AI researchers themselves.
“This is their main North Star at the moment is: automate AI research,” he says. “Because then we can compound it.”
Second, it’s about the money. The software market is vast, and Anthropic sees a clear path to revenue by creating agents that can build software for a slice of that market, which Andreessen Horowitz general partner Alex Rampell recently estimated on a podcast at $300 billion annually.
“They see it as ‘Well if we can build coding agents that can build software, then we can go get a piece of that $300 billion annual market of software,’” says Roetzer.
But, while a $300 billion annual SaaS market is an attractive prize, Roetzer cautions that it’s just the tip of the iceberg. In the same podcast, Rampell said the market for human labor in the US alone is $13 trillion.
Follow the money: This is a simple acknowledgment of the economic forces at play. When you look at the billions of dollars VCs are pouring into AI labs, it becomes clear that the ultimate target isn’t just software—it’s labor.
“It is pure economics and pure capitalism, and I don't think it's even a debatable thing,” says Roetzer. “If you just zoom out and you just look at those numbers, there's no way people don't build to replace human labor.”
Anthropic’s Claude Sonnet 4.5 is a remarkable technical achievement, pushing the boundaries of what AI can do in the complex world of software engineering. Its ability to work coherently for over 30 hours is a massive leap forward for AI agents.
But more importantly, it’s another clear signal of the industry's trajectory. We’re in a rapid, repeating cycle where models get smarter every few months, driven by the relentless power of scale. And while the immediate applications are in coding and software, the ultimate economic destination is far larger.
The race to automate AI research and capture the software market is just a stepping stone toward the multi-trillion-dollar prize of automating knowledge work itself.