Boris Cherny explains why AI ROI matters more than token costs and how vibe coding is evolving into autonomous AI-driven development.
As businesses pour more money into AI, concerns about rising AI costs are growing louder. But according to Boris Cherny, the creator of Anthropic's Claude Code, companies may be looking at the issue the wrong way. The discussion around Boris Cherny on Vibe Coding has gained attention as he continues to challenge conventional thinking about how organizations measure the value of AI investments.
Speaking during a recent fireside chat hosted by Scale AI, Cherny argued that the real conversation should not be about how much AI costs, but about the value it creates. His comments come as organizations increasingly question whether growing AI token expenses are delivering meaningful business results.
AI tokens are the units used to measure how much text AI systems process. As companies rely more heavily on AI tools for coding, customer service, and business operations, token usage and costs can rise quickly.
Focusing on ROI Opens the Door to Innovation
Cherny believes return on investment (ROI) should be the primary metric when evaluating AI spending.
Rather than restricting employees from using AI due to budget concerns, he suggested giving teams room to experiment. According to him, some of the most valuable AI-driven ideas often emerge from unexpected places within an organization, not just from executives or technology teams.
He noted that innovative solutions can come from employees in finance, marketing, or other departments who identify practical ways to improve workflows using AI. Limiting experimentation too early could prevent businesses from discovering those opportunities.
Is AI Writing Too Much Code?
One of Cherny's most striking observations was that AI-generated code is becoming so common that traditional productivity measurements are losing relevance. In the past, companies tracked how much code AI produced. Today, many developers allow AI to generate nearly all of their code. As a result, the challenge is shifting.
The new bottleneck, Cherny says, is no longer writing software faster. It is generating better ideas. When engineers can produce large amounts of code with AI assistance, business success increasingly depends on creativity, problem-solving, and identifying valuable opportunities.
The Era of Prompting May Be Coming to an End
Cherny also believes software development is entering a new phase beyond traditional "vibe coding."
Instead of humans constantly writing prompts, he envisions "loop engineering," where AI agents create prompts for other AI agents, evaluate outcomes, and continue improving results with minimal human involvement. In this model, developers act more like managers. Their role becomes designing workflows, setting goals, and supervising networks of AI agents rather than directly writing code.
Industry experts such as Peter Steinberger, Addy Osmani, and Claire Vo have echoed similar views, suggesting future developers may spend more time orchestrating AI systems than programming manually.
Higher Autonomy Brings Higher Costs
The shift toward autonomous AI systems comes with a trade-off. Multiple AI agents working together can rapidly consume token budgets.
Cherny acknowledged the challenge and highlighted that Anthropic offers enterprise controls such as budget limits, usage monitoring, and per-seat spending controls. He also pointed out that tokens carry costs for AI providers as well, creating an incentive for efficient usage on both sides.
Looking Ahead
As Business Fortune observes, AI tools become more autonomous and businesses may move beyond measuring productivity by lines of code or prompt quality. The next competitive advantage could be how effectively companies combine human creativity with AI-driven execution. If Boris Cherny's vision proves correct, the future of software development will not be about writing more code. It will be about building smarter systems that generate ideas, solve problems, and continuously improve on their own.
