At a recent two-day conference in London, Anthropic urged developers to allow their AI tool, Claude, to independently write and verify code. The event, dubbed "Code with Claude," revealed that nearly half of the participants admitted to submitting pull requests composed entirely by Claude without even reviewing the generated code. This highlights a significant shift in routine practices among developers, as the company aims to automate coding processes to the extent that AI can self-check and rectify its work.
Jeremy Hadfield, an engineer at Anthropic, engaged the audience by asking how many had submitted such pull requests, prompting nervous laughter as most hands were raised in acknowledgment. The shift towards AI-assisted coding is a clear indication of how tools like Claude Code, alongside competitors such as OpenAI's Codex and similar offerings from Google and Microsoft, are transforming software development.
Hadfield noted that much of Anthropic's software is now being produced by Claude, particularly following recent updates that have enhanced the tool's capabilities. The latest versions, Claude 4.6 and Claude 4.7, have made the AI more efficient, allowing developers to delegate routine tasks to it.
Boris Cherny, the creator of Claude Code, expressed ambitious goals for the tool, envisioning a future where Claude generates its own requests and handles debugging autonomously. "The key principle is to not interfere with Claude. We like to say, 'Let him finish the job,'" said Ravi Trivedi, another Anthropic engineer.
A new feature called "dreaming" was also introduced, where AI agents autonomously document their tasks, facilitating Claude's understanding of specific codebases over time. Companies such as Spotify and Delivery Hero have already adopted Claude Code into their development processes.
Despite the enthusiasm at the conference, concerns linger outside the event. Developers on platforms like Reddit and Hacker News are voicing frustrations that AI tools, pushed by management for productivity, often lead to an increased volume of code that requires review. Some claim their programming skills have declined as they rely more on generated code, raising questions about the safety and quality of AI-produced software.
Katelyn Lesse, the technical lead for Claude, acknowledged these concerns but reiterated that established coding practices remain crucial. She noted that some technical managers at Anthropic are feeling overwhelmed by the amount of code being generated. Currently, Claude is estimated to write code at the level of a mid-level engineer, with experienced professionals still necessary for complex system designs.
As the competition intensifies in the AI space, Claude's evolving capabilities may reshape the landscape of software development. This shift could streamline processes but might also challenge traditional roles within tech teams, prompting a reevaluation of skill requirements and responsibilities in the industry.
Informational material. 18+.
