Anthropic’s obsession with developers took a major leap last week with the announcement of Claude Code. While it impressed several users with its ability to run and preview code using the Artifacts feature, Claude Code goes one step further as an ‘agentic coding tool’ that operates directly within the terminal.
It is capable of fixing bugs across a code base, resolving merge conflicts, creating commits and pull requests, and answering questions about the architecture and logic.
Moreover, as revealed by the company’s chief product officer Mike Krieger, Anthropic’s approach is largely about “picking its bets” carefully. With Claude Code, they took a strategic step by first releasing it internally to boost its own team’s performance.
“After seeing it play out for a couple of months, we thought, ‘This is good.’ It’s not a solution for all coding problems, and doesn’t obviate the IDE (integrated development environment). But it is useful to us in enough cases that we want to see people use it out,” said Krieger, in a podcast episode with venture capitalist Harry Stebbings.
“Our product engineers love Claude Code,” he added, indicating that most of the work for these engineers lies across multiple layers of the product. Notably, it is in such scenarios that an agentic workflow is helpful.
Meanwhile, Emmanuel Ameisen, a research engineer at Anthropic, said, “Claude Code has been writing half of my code for the past few months.”
Similarly, several developers have praised the new tool. Victor Taelin, founder of Higher Order Company, revealed how he used Claude Code to optimise HVM3 (the company’s high-performance functional runtime for parallel computing), and achieved a speed boost of 51% on a single core of the Apple M4 processor.
He also revealed that Claude Code created a CUDA version for the same.
“This is serious,” said Taelin. “I just asked Claude Code to optimise the repo, and it did.”
Several other developers also shared their experience yielding impressive results in single shot prompting.
Pietro Schirano, founder of EverArt, highlighted how Claude Code created an entire ‘glass-like’ user interface design system in a single shot, with all the necessary components.
Notably, Claude Code also appears to be exceptionally fast. Developers have reported accomplishing their tasks with it in about the same amount of time it takes to do small household chores, like making coffee or unstacking the dishwasher.
However, if one is looking at the intersection between AI and coding, Cursor has to be taken into consideration. The AI coding agent recently reached $100 million in annual recurring revenue, and a growth rate of over 9,000% in 2024 meant that it became the fastest growing SaaS of all time.
A user on Reddit compared both Cursor and Claude Code. The review stated that Claude Code produces code of “very high quality”. “This thing blows Cursor out of the water. I can’t believe both use the same model when I see the difference in how Claude-3.7 behaves in Cursor and how it behaves in Claude Code,” the review added.
“I’ve had no functionality breaking mistakes, which happen every now and then with Cursor, where it just breaks something or large files are truncated,” it further stated.
It is ‘Insanely Expensive’
Anthropic has always found a sweet spot in the hearts of developers for most, if not all, of their products – from Computer Use to MCP, and now the Claude Code. However, there has also been a fair share of criticism.
For one, Claude Code is very expensive as the API pricing for Anthropic’s AI models is one of the highest out there.
The Claude 3.7 Sonnet costs $3 per million input tokens and a whopping $15 per million output tokens. One developer called it “insanely expensive” and said they could easily spend $50-$100 on it, while agreeing that it is better than Cursor.
Matt Popovich, an engineer at Forge, said that Claude Code costs as much as hiring a developer.
However, autonomous coding agents are expensive, considering Devin, the first to the market, costs a whopping $500 a month, but with no limit on the number of seats in an organisation. Another developer said that Claude Code costs him $28 a day, adding that it will end up costing the same as Devin.
asking claude code to change the text on a label pic.twitter.com/chY4GKKP67
— Tetrabay 3 💎🌁 (@TetraspaceWest) March 4, 2025
Besides, there are other problems associated with Claude Code. Multiple users on GitHub pointed out that running a command to automate Claude updates on the Ubuntu Server 24.02 messed with the system file ownership, locking users out of admin access. However, Anthropic did provide a solution to mitigate the issue.
Moreover, a few developers also reminded that AI coding agents still have a long way to go, and plenty of room to elevate their scope. Petr Baudis, CTO of Rossum, said on X that Claude Code struggled with certain real-world engineering tasks.
He found that the tool wrote several redundant and unreviewable code, and it cost $55 to do so.
“To be clear, looking at this from 2023, it’s absolutely mindblowing that AI can do all this. But it’s also simply not useful for actual engineering tasks where plain code-writing isn’t the bottleneck. Not even tests,” he added.
It would be unfair to single out Claude Code, however, as the situation is more or less similar with multiple autonomous coding platforms.
