AI Token Budgets: Companies Should Prioritize ROI and Employee Experimentation

Companies must balance AI ROI demands with employee experimentation, says Anthropic’s Boris Cherny

Anthropic employee Boris Cherny, creator of the Claude Code AI tool, has advised businesses to prioritize return on investment (ROI) in AI spending while ensuring employees retain access to tokens for experimentation. According to Cherny, who spoke at a recent Scale AI fireside chat, companies risk stifling innovation if they overreact to token cost concerns by restricting access upfront.

Cherny’s remarks follow Uber COO Andrew Macdonald’s public concerns about the rideshare giant’s AI spending failing to deliver sufficient returns. Tokens—units measuring AI usage such as prompts processed by large language models—are central to this debate, as their cost directly impacts enterprise budgets.

Why companies must allow AI experimentation despite ROI pressures
Cherny emphasized that companies should first enable employees across roles to experiment freely with AI tools. “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,” he said. Restricting access too early, he warned, could prevent breakthrough ideas from emerging from unexpected sources.

“Often, some of the most interesting ideas and the most innovative ways to improve processes and new product ideas are going to come from an accountant somewhere in the corner of the org or a marketing person that the CEO has never heard of,” Cherny noted. Once promising use cases are identified, costs can then be controlled through backend measures like per-seat budgeting, he added.

Anthropic offers multiple tools to help enterprises manage token expenses, including granular cost controls. However, Cherny acknowledged that AI providers like Anthropic have a vested interest in driving token usage, particularly as they approach potential IPOs. “Every token we use is a token we do not give to a customer, so there’s an opportunity cost,” he said, framing the challenge as an ROI issue for both companies and AI firms.

How Anthropic measures AI ROI—and why traditional metrics are outdated
Cherny highlighted that measuring AI’s impact has evolved beyond simple metrics like the percentage of code written by AI. As adoption grows, he said, companies should now assess how much AI accelerates engineers’ output and what bottlenecks remain.

“Once you get it to this point where engineers are just writing a lot of code, the bottleneck is going to be like good ideas,” Cherny explained. “So, how do you un-hobble that so that your company can generate ideas faster?”

His comments align with broader industry discussions. OpenAI CEO Sam Altman has also recently addressed companies’ concerns about AI spending ROI, signaling a growing focus on balancing cost with innovation.

What this means for enterprises adopting AI tools
Anthropic’s approach reflects a tension between controlling expenses and fostering creativity. While cost management remains critical, Cherny’s advice suggests that companies should avoid prematurely restricting access to AI tools. Instead, they should allow experimentation early, then implement controls once effective use cases are proven.

For enterprises, this means adopting a phased strategy: allocate tokens broadly at first, monitor internal adoption, and later refine spending based on measurable outcomes. The risk, Cherny warned, is that overly restrictive policies could suppress the very ideas that drive long-term value.

Key takeaways from Cherny’s guidance

• Prioritize experimentation: Allow employees across departments to test AI tools without immediate cost penalties.
• Control costs later: Implement budgeting tools like per-seat limits once successful use cases are identified.
• Reassess metrics: Move beyond basic adoption rates (e.g., code written by AI) to measure productivity gains and idea generation.
• Balance incentives: AI providers and companies must align on ROI, but experimentation should not be sacrificed for short-term savings.

Cherny’s perspective underscores a broader industry shift: AI’s true value may lie not just in efficiency gains but in unlocking creativity and problem-solving from unexpected sources within organizations.