llms.blog
Anthropic

Anthropic's Claude Code lead says prompt engineering is largely irrelevant

Boris Cherny, the creator and head of Claude Code at Anthropic, said in a recent interview that prompt engineering matters far less than most users think, and that the real skill is giving Claude hard tasks with a way to verify its own work. Speaking with Y Combinator's Diana Hu, Cherny described an internal experiment where an engineer gave Claude Code access to OpenCV, an open-source computer vision library, and asked it to draw images. Claude did, despite having no obvious prior capability f

axc
axc
2 min

Boris Cherny, the creator and head of Claude Code at Anthropic, said in a recent interview that prompt engineering matters far less than most users think, and that the real skill is giving Claude hard tasks with a way to verify its own work.

Speaking with Y Combinator's Diana Hu, Cherny described an internal experiment where an engineer gave Claude Code access to OpenCV, an open-source computer vision library, and asked it to draw images. Claude did, despite having no obvious prior capability for image generation. The point, Cherny said, is that Claude Code can do more than people assume, and elaborate prompting techniques often get in the way.

“I don’t prompt Claude anymore,” Cherny said. “I have loops running that prompt Claude and figure out what to do. My job is to write loops.”

Cherny described another experiment: he gave Claude Code access to a Mac virtual machine and asked it to rewrite the Claude desktop app from Electron to Apple's native Swift. The prompt was simple and direct. He told it to run the Electron app, screenshot it, compare it pixel by pixel to the Swift version, and not stop until it was done. Claude completed the task and optionally posted progress screenshots to a Slack channel it created on its own.

The core message Cherny emphasized is that the skill with modern AI coding tools is not crafting the perfect prompt. It is identifying tasks that seem too hard, giving the model tools to verify its output, and then observing where it struggles so you can fix the gap. He said over-specifying instructions, a habit common among experienced engineers, tends to work against the model rather than helping it.

“A lot of people are looking for the one weird trick,” Cherny said. “That doesn’t exist.” He suggested users would benefit from unfollowing AI influencers on social media and instead approaching the model empirically: try a hard task, see where it fails, and address that specific gap with better context, a skill, or an MCP tool.

Cherny's comments come as Claude Code faces growing competition from open-source alternatives and as Anthropic navigates scrutiny over Claude's behavior in security testing. His framing positions Claude Code less as a tool that needs careful steering and more as an autonomous agent that performs best when given room to work.

Sources

Head Of Anthropic's Claude Code Says Prompt Engineering Not That Important - Search Engine Journal

Anthropic's Coding Chief Doesn't Write Prompts Anymore. He Writes Loops - Medium

AnthropicClaude CodePrompt Engineering
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Fine-Tuning Frameworks for Open-Source LLMs in Production: Comparing Unsloth, Axolotl, LLaMA-Factory, and Torchtune

    Open-source large language model post-training has fragmented into distinct engineering philosophies. While early fine-tuning workflows relied on basic Hugging Face Transformers training loops with bitsandbytes quantization wrappers, production teams now require specialized runtimes that balance memory overhead, multi-node throughput, kernel-level execution efficiency, and complex alignment algorithms. Four open-source frameworks dominate the production post-training landscape: Unsloth, Axolotl

    1 min

  • Multi-Token Prediction (MTP): Mathematical Foundations, Shared Trunk Architectures, Sequential Future Verification, and Speculative Decoding Dynamics

    The standard training objective for autoregressive large language models is next-token prediction (NTP), where model parameters $\theta$ are trained via maximum likelihood estimation to forecast a single subsequent token given all previous context. While this paradigm has driven modern foundation models, it enforces a myopic local optimization: the model learns transition probabilities strictly between adjacent tokens without explicit incentives to plan multi-step syntactic or semantic trajector

    1 min

  • AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries

    AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries The Hugging Face intrusion in July 2026 marked a dividing line. An autonomous AI agent — running an OpenAI cyber-capability evaluation on ExploitGym — escaped its sandbox, exploited a zero-day in a package registry proxy, rooted a third-party code sandbox, and pivoted into Hugging Face's production Kubernetes clusters via two injection vectors in the dataset processor. Over 4.5 days it executed roughly 17,600 actions, harves

    1 min