Claude Code creator Boris Cherny on what not to miss while prompting Claude

Boris Cherny, creator of Anthropic's Claude Code, has shared a simple piece of advice for users aiming to improve their interactions with AI models. In a post shared on X (formerly known as Twitter), Cherny suggested speaking to Claude in the same way one would converse with a coworker, eliminating the need for overly structured and detailed prompts.

He expressed surprise that people were surprised by this approach, as he believes modern AI models can figure out how to achieve a goal without excessive scaffolding or prescriptive instructions.

Cherny contrasted this current approach with the early days of Claude's Sonnet 3.5 models, where the specific wording of prompts was crucial. Today, he argues, the focus should shift from "how" to "what," emphasizing clarity around three key aspects:

1. The desired task for the AI model

2. The amount of effort the user expects the AI to invest

3. The verification method Claude should use to confirm it accomplished the task correctly

This aligns with Cherny's earlier public discussions, where he cautioned against over-specifying every step and instead recommended clearly defining the desired outcome. He advocates for users to describe the goal and success criteria rather than dictating a rigid sequence of actions.

To illustrate his point, Cherny shared an example prompt he employs with Claude. He asked the AI to create an interactive artifact based on a podcast episode discussing Home Depot's acquisition strategy. The prompt required Claude to produce a visually rich, interactive experience highlighting key business lessons. Unlike traditional prompts, this example focused on the end goal, desired outcome, and quality bar, rather than specifying a detailed process.

This approach exemplifies Cherny's prompting philosophy and demonstrates the power of clarity and focus when working with AI models.