If you have been crafting long, step-by-step prompts to get better results from AI, you might actually be making things harder.
According to Anthropic's popular AI chatbot Claude Code creator Boris Cherny, the "common mistake" users make is trying to give lengthy prompts to micromanage AI, instead of trusting it to figure out the best way to complete a task.
Speaking at a recent Y Combinator event, Cherny said many Claude users overload the AI with highly detailed instructions, leaving little room for the model to use its reasoning capabilities. "I think a really common mistake that I see is people are using Claude Code, they're using Claude, and they just give way over-specific instructions," he said.
He explained that users often dictate every step of the process instead of simply describing the desired outcome. "They're like, 'I want you to do this, but I want you to do it in this way, this way, this way. You must do like one, then two, then three, then four.'"
According to Cherny, that approach may have made sense several months ago, but the latest generation of AI models has become far more capable of independently determining how to solve complex problems. He also encouraged users to assign AI "harder tasks" tasks than they believe it can handle, arguing that each new generation of models significantly expands what is possible.
To explain his point, Cherny revealed that Claude recently spent 11 days rewriting an entire production codebase from one programming language to another. According to him, the same project would have taken human engineers more than a year to complete.
ALSO READ: Indian Pharma Steps Up AI Push in Drug Discovery, But Investment Gap With Global Peers Persists
'Treat AI As Your Co-Worker'
Speaking about learning prompt engineering, Cherny argued that the future is actually about giving AI models more autonomy to accomplish tasks on their own. The Anthropic executive stated that, "The verification I think is probably the single most important thing that people do not get right." He explained that users should provide AI with ways to test and validate its output instead of relying on increasingly elaborate prompts.
"I think people tend to overthink it a little bit. I think people tend to overengineer," Cherny said. He further suggested that people should treat AI "like you would a coworker," as "the level of intelligence that it's at now." This will help in better AI performance, rather than trying to force AI to follow the exact process a human would use.
However, Cherny is not the only person speaking against giving AI models overly specific prompts. Earlier last year, Google Brain co-founder Andrew Ng shared a similar idea in a post on X (formerly Twitter), in which he called giving AI models specific prompts as "lazy prompting".
ALSO READ: AI Bubble Burst Can 'Sharply Weaken' Global Economy, Singapore Central Bank Warns
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.