Anthropic gave Claude a new job reviewing code.
However, some developers say the new features could eliminate the role of senior engineers and inflate the amount charged for their tokens.
The AI startup announced in a blog post on Monday that Claude’s code review is designed to detect more complex issues and suggest fixes for bugs. This feature “optimizes for depth” and is therefore “more expensive than lightweight solutions like Claude Code GitHub Action,” Anthropic said.
Claude Code GitHub Action is an integration that allows developers to use Claude to review code on GitHub, a platform for storing and collaborating on code.
“Reviews are charged based on token usage and typically average between $15 and $25, but scale depending on the size and complexity of the PR,” the company added, referring to pull requests, a common workflow in which developers propose changes to a codebase.
Thariq Shihipar, an Anthropic engineer who works on Claude Code, said in Tuesday’s X that the feature “tends to use more compute and catch more difficult bugs.”
Schichper was responding to criticism from another developer, Daniel Avila, who goes by Daniel Sun online and claims to be the head of AI at fintech startup Hegenia.
Avila said in an X post on Tuesday that using code reviews “did not show any improvement.”
“Apart from not requiring triggers like GitHub Action and being able to configure it directly within Claude Desktop, there are no additional features at all,” Avila said.
“The only real difference is that GitHub Actions uses an API and Code Review does not,” he added.
Reactions among developers have been mixed, especially regarding price.
Gene PD Meyer, a software engineer and founder of a software startup, questioned the pricing in a post about X.
“Claude Code Review goes from $15 to $25 per PR?!” Meijer wrote under the handle @initjean.
“One review is basically the price of an unlimited subscription to most code review apps,” he added. “Did they forget to subsidize the token?”
Karan Kendre, another developer at a design startup, also wondered about the logic behind this feature.
“If Claude Code can get clean code working in his codebase, why can’t he just write it from scratch?” Kendle wrote in Tuesday’s X.
“Because they want to charge $15 to $25 per code review on top of the Claude code subscription,” he added.
A developer building an AI startup by the name @TukiFromKL wrote, “Claude just put a price tag on senior developers.”
“$15-25 per code review. AI agent that launches instantly the moment you open a PR,” the user wrote.
“People who protect their $200,000 salaries through gatekeeping code reviews are going to have a very bad quarter,” the user added.
AI coding tools are getting expensive
Anthropic is not the only AI company facing increased scrutiny over token costs.
Billionaire investor Chamath Palihapitiya said on Friday’s episode of the “All-In Podcast” that his software startup is reconsidering its use of Cursors due to skyrocketing expenses.
“Since November, our costs have more than tripled,” he said. “If you add up the huge inference costs we pay to AWS, the Cursor costs, and the Anthropic costs, we’re just spending millions of dollars.”
Palihapitiya said his startup, 8090, plans to spend about $10 million a year on AI costs.
“The problem is that costs are going up three times every three months,” he said. “My income is not like that.”
He cited cursors as a contributing factor to the company’s higher bills and said Claude Code is relatively cheaper.
