📷 Image: Times of India (via NewsData)
Technology
Claude Code Creator Warns Against Over‑Instruction in AI Prompting
✍️ Times of India
🗓 30 Jul 2026, 10:40 AM
👁 2
Boris Cherny, the mind behind Anthropic’s Claude Code, cautioned users at a Y Combinator event that giving step‑by‑step instructions to the model is the most common mistake. He urged a shift to task description, guardrails and success criteria, and introduced a new framework to gauge enterprise AI ROI beyond token usage.
Boris Cherny, the creator of Anthropic’s Claude Code, addressed a gathering at Y Combinator, highlighting a frequent pitfall in AI prompting: over‑instruction. "When users hand the model a detailed, step‑by‑step order, they limit its creative potential," Cherny said.
He advised users to shift from rigid instructions to a broader task description, establishing clear guardrails and defining what success looks like. This approach, he explained, allows Claude to interpret the intent more flexibly while staying within safe boundaries.
In addition to prompting guidance, Cherny unveiled a framework for measuring enterprise AI returns that goes beyond token‑burn dashboards. He argued that usage metrics track activity, not value, and that organizations should focus on outcomes such as efficiency gains, cost savings, and business impact.
The speaker’s remarks were well received by the audience, many of whom are exploring Claude for commercial applications. The emphasis on intent‑driven prompting and outcome‑based ROI is expected to influence how enterprises deploy and evaluate AI tools in the coming months.
He advised users to shift from rigid instructions to a broader task description, establishing clear guardrails and defining what success looks like. This approach, he explained, allows Claude to interpret the intent more flexibly while staying within safe boundaries.
In addition to prompting guidance, Cherny unveiled a framework for measuring enterprise AI returns that goes beyond token‑burn dashboards. He argued that usage metrics track activity, not value, and that organizations should focus on outcomes such as efficiency gains, cost savings, and business impact.
The speaker’s remarks were well received by the audience, many of whom are exploring Claude for commercial applications. The emphasis on intent‑driven prompting and outcome‑based ROI is expected to influence how enterprises deploy and evaluate AI tools in the coming months.