Anthropic Engineer Urges ROI Focus, Preserve AI Experimentation

Boris Cherny, an engineer on Anthropic's Claude Code team, spoke at a Scale AI fireside chat and discussed corporate approaches to AI spending, Business Insider reports. Cherny said "ROI is absolutely the right framing because you don't want to just think about cost because you kind of spend something on it and you get something back," Business Insider quotes. When asked by Jesse Chen, Meta's director of product management, about concerns raised by Uber COO Andrew Macdonald over token spending, Cherny advised allowing internal experimentation before enforcing strict cost controls, Business Insider reports. He stated, "The way to do this is give people tokens and give them safety to experiment so they feel like they can try stuff and they're not going to get penalized for it," Business Insider reports. Cherny also noted that Anthropic provides enterprise cost controls, including per-seat budgeting, Business Insider reports.

The article defines tokens as units of text used to measure model usage, a standard billing metric for large language models, Business Insider explains. Industry practitioners commonly balance token-based metering with business outcomes; technical controls that shift throttling or rate limits to backend systems can preserve developer and employee experimentation while limiting spend at scale. Companies building internal AI platforms often implement safe sandboxes, quota inheritance, and monitored cost aggregation to reconcile exploration with budget constraints.

Companies tightening AI budgets in response to headline token costs risk reducing low-cost, high-value experimentation from nonengineer staff, Business Insider's account suggests. Observed patterns in similar transitions show that grassroots experimentation frequently surfaces unexpected process improvements and product ideas; product managers and platform teams typically try to protect small exploratory budgets while centralising fiscal controls.

Indicators to follow include broader uptake of per-seat or departmental quota controls across enterprise AI vendors, adoption of backend cost throttling patterns in internal platforms, and whether firms publish metrics linking token spend to measurable ROI. For practitioners, tracking how vendor features map to governance needs will determine how easily organizations can balance experimentation and cost discipline.