The rise of AI-generated code is prompting a fundamental shift in software development, as demonstrated at Anthropic's recent two-day event, Code with Claude. Held in London, the gathering revealed that developers are becoming more comfortable assigning programming tasks to AI systems like Claude. Engineers reported that nearly half of attendees had submitted pull requests entirely composed by Claude, often without reviewing the code. This trend indicates a growing acceptance of AI tools in coding workflows, even as concerns about oversight and security remain.
Much of the discussion at Code with Claude centered on the capabilities of Anthropic's latest iterations of Claude, particularly versions 4.6 and 4.7. These updates have significantly enhanced Claude’s coding abilities, leading many developers to allow the AI to manage code generation independently. Jeremy Hadfield, an Anthropic engineer, noted, "Most software at Anthropic is now written by Claude," underscoring the profound impact AI is having on traditional coding practices.
Anthropic's ambition is clear: to push automation to new heights. The company aims for Claude to self-correct and improve its output autonomously, rather than merely generating code for human developers to refine. Boris Cherny, who leads Claude Code, articulated this vision during his keynote, stating that the new default approach should be Claude prompting itself rather than relying on human prompts. This could eventually enable Claude to handle error corrections independently, lessening the burden on human engineers.
The Role of 'Dreaming' in AI Coding
One of the standout features presented at the event was Claude Code's new capability, known as "dreaming." This mechanism allows Claude to take notes and learn from past coding tasks, enhancing its effectiveness on subsequent projects. Ravi Trivedi explained that the concept revolves around enabling Claude to gather insights from its own previous work, thereby improving the overall coding process and efficiency. This self-learning aspect could make Claude increasingly proficient at managing various coding tasks, ultimately shaping a future where AI-generated code may dominate software development.
Despite the enthusiasm surrounding Claude Code, the event was not without skeptics. Outside the conference, many developers voiced concerns that the rapid adoption of AI in coding could lead to increased vulnerabilities in software. Critics argue that relying on AI-generated code without thorough human oversight could create serious security risks. A user on Hacker News pointed out that those most confident in generated code are often the ones who do not scrutinize it.
In response to these concerns, Anthropic's leadership emphasized the continued importance of traditional software development practices. Katelyn Lesse, the engineering lead for Claude, acknowledged that maintaining these best practices is essential, even as automation becomes more prevalent. She noted that while Claude may currently perform at a level comparable to a mid-level engineer, expert oversight is still necessary for complex problem-solving and system design.
Balancing Automation and Human Oversight
The rapid pace at which AI tools are evolving presents a unique challenge for technical management. Lesse highlighted that some managers at Anthropic have struggled to keep up with the volume of code produced by their teams, which has surged with the introduction of AI tools. This situation raises questions about balancing productivity gains with the need for careful oversight in software development.
Angela Jiang, the product lead for Claude, echoed this sentiment, asserting that while the goal is for Claude to become increasingly autonomous, human expertise will always be needed to navigate complex engineering challenges. The ultimate vision for Claude is to achieve a level of self-sufficiency where it can effectively build and improve upon its own coding capabilities.
As AI tools like Claude Code continue to advance, software development is likely to undergo significant changes. While integrating AI can lead to greater efficiency and innovative coding practices, it also requires careful examination of the implications for security and developer skills. The balance between human oversight and AI autonomy will ultimately shape the future trajectory of software development in an increasingly automated era.
The stories that move AI & crypto markets — before the market reacts.
Free. 7am ET. Five stories. 62,400 readers.
