Claude Code Creator: ROI Focus Is Fine, But Experiment Too

Boris Cherny, the Anthropic engineer who built Claude Code, is pushing back gently on the all-or-nothing debate around AI return on investment. Speaking to Business Insider, Cherny said companies demanding measurable results from their AI spending are making a reasonable call. At the same time, he warned that organizations locking down every AI initiative behind a strict ROI gate will end up shortchanging themselves in the long run.

The Case for Keeping Both Doors Open

Cherny's argument is straightforward: ROI pressure creates discipline, and discipline is not a bad thing when companies are deploying AI at scale. Boards and finance teams asking hard questions about what they are getting for their software budgets are doing exactly what they should. The problem emerges when that pressure becomes so intense that engineers and product teams never get the latitude to try things that do not have an obvious business case on day one. Cherny has previously spoken about how AI is reshaping the role of software engineers, a shift that itself grew from exactly the kind of exploratory work he is now advocating for.

The tension Cherny is describing is real and widely felt across the industry. Companies that rushed into AI pilots in 2023 and 2024 are now being asked by leadership to justify the spend. Many are tightening scope, cutting speculative projects, and funneling resources toward use cases with clear, near-term payoffs. That is a rational response to economic pressure. But Cherny's point is that some of the most useful AI applications were not obvious until someone had the freedom to stumble onto them. His own team at Anthropic manages tens of thousands of AI agents daily, a capability that emerged from sustained, iterative experimentation rather than a pre-approved project plan.

Companies are right to want ROI. But if every project needs a business case before anyone is allowed to touch it, you are going to miss the things that are actually interesting.Boris Cherny, Anthropic, via Business Insider

Where the Balance Sits in Practice

Cherny did not offer a precise formula for how much experimentation budget a company should protect. Instead, his advice leaned toward cultural and structural changes. Teams need psychological safety to report when something is not working, and they need time that is not fully accounted for in quarterly deliverables. That sounds soft, but it maps to how most significant internal tools get built. Claude Code itself started as an internal experiment before Anthropic recognized its broader potential and turned it into a product.

The competitive landscape makes this advice more urgent. Rivals are actively targeting Claude Code's developer market share, which means Anthropic cannot afford to stand still, and neither can the enterprises relying on these tools. Companies that restrict AI exploration too tightly may find themselves behind peers who let teams run looser experiments and found better workflows as a result. The ROI conversation is necessary. The experimentation conversation is equally necessary, and it tends to get crowded out when earnings pressure dominates.

Cherny's broader message is less about defending any single tool or approach and more about the mindset companies bring to AI adoption. Treating every AI initiative as a cost center to be optimized will produce incremental gains. Treating some portion of it as a research function, even a small one, is where larger shifts tend to originate. For developers and technology leaders trying to make the case internally, that distinction may be the most useful thing Cherny has said yet.