According to Boris Cherny, the developer of the Claude Code tool at Anthropic, the era of detailed template instructions and “prompt engineering” is gradually becoming a thing of the past thanks to the rapid development of artificial intelligence algorithms. Instead of multi-level prompts, users should switch to natural communication with models.
According to Boris Cherny, the developer of the Claude Code tool at Anthropic, the era of detailed template instructions and “prompt engineering” is gradually becoming a thing of the past thanks to the rapid development of artificial intelligence algorithms. Instead of multi-level prompts, users should switch to natural communication with models.
He explained this in a post on the social network X (Twitter). According to him, with the improvement of models, the need for deep refinement or excessive structuring of text queries has significantly decreased.
“Talk to Claude as you would talk to a colleague. There is no secret to the prompts. For most tasks, there is no need for overly structured or directive language,” Cherny noted.
Instead of complex algorithms with caveats, the engineer advises focusing on just three basic elements of the query:
what exactly the model should do;
how much effort she has to make;
how the algorithm should verify the correctness of the work performed.
As an example, Cherny gave his own casual prompt for analyzing a podcast episode, where he asked Claude to use more tokens and make the visualization better than the world’s leading publications. In response to users' comments about his phrase "maybe add watercolor," the developer emphasized that the latest model Opus 5.5 already has enough "taste" to make stylistic decisions.
Boris Cherny previously said that the popular term vibecoding has become too frivolous for an industry that generates billions of dollars. He believes that a more serious alternative is needed.
