Claude Code creator Boris Cherny flags risks of AI writing 100% of code, calls it ‘problematic’

Boris Cherny, creator of Anthropic’s coding tool Claude Code, has cautioned that fully automating software development using artificial intelligence is proving to be “problematic” for companies, even as AI adoption accelerates across engineering teams.Speaking at a fireside chat hosted by Scale AI, Cherny said organisations are increasingly shifting focus toward return on investment (ROI) when deploying AI tools, particularly as usage costs rise. His comments come amid growing concerns from industry leaders about whether spending on AI is translating into measurable business value.Cherny was responding to broader industry debates, including remarks from Uber COO Andrew Macdonald, who recently questioned whether escalating AI investments are delivering enough tangible consumer-facing outcomes. According to Cherny, evaluating AI purely through cost or output is insufficient.“ROI is absolutely the right framing, you spend something on it and you get something back,” he said.While AI-generated code has significantly improved developer productivity, Cherny acknowledged that allowing AI systems to write the entire codebase introduces new bottlenecks. He said measuring progress by the percentage of AI-written code is no longer meaningful, as many engineers already rely heavily on such tools.“Once you get it to this point… the bottleneck is going to be good ideas,” Cherny noted, highlighting a shift in software development from execution to innovation and problem definition.He emphasised that as AI takes over repetitive coding tasks, the focus for organisations must move toward idea generation, product thinking and strategic problem-solving, areas where human contribution remains critical.Cherny also flagged the economic constraints of large-scale AI deployment, pointing out that “tokens”—the units of computation used by AI systems—carry real costs even for providers like Anthropic. “Every token we use is a token we do not give to a customer, so there’s an opportunity cost,” he said, underscoring the need for careful resource management.At the same time, he cautioned against prematurely limiting AI usage, arguing that experimentation is essential to unlocking innovation. Restricting usage too early could prevent teams from discovering higher-value applications of the technology.Cherny also pointed to a shift in how developers interact with AI systems. He said the industry is moving toward “loop engineering,” where AI agents autonomously generate and refine prompts, reducing the need for manual input.In this model, developers interact with higher-level systems that coordinate tasks across multiple AI agents. However, he noted that this approach can increase costs, especially when several agents and sub-agents operate simultaneously.Cherny’s comments reflect a broader recalibration in the tech industry, where companies are balancing productivity gains from AI with rising costs, operational challenges and the need to maintain human oversight in critical workflows.Read more: Anthropic launches ‘Claude Tag’ for team-based AI collaboration on Slack