Anthropic Co-Founder Says Prompt Engineering Is Dying, Meet ‘Loop Engineering’

Just months after arguing that traditional software engineering was becoming obsolete, Anthropic co-founder Boris Cherny is now making a different prediction: the era of manually writing prompts for AI may soon be coming to an end. Instead, the Claude-maker executive believes the future belongs to something called "loop engineering." Here's everything you need to know what loop engineering is, why AI leaders are talking about it and how it could change the way people work with AI systems.According to a recent Business Insider report, Cherny now believes that users should spend less time crafting detailed prompts and more time designing systems that allow AI agents to manage those prompts themselves."It's an agent that prompts Claude. I don't

write the prompt anymore. Claude writes the prompt and now I'm talking to that new Claude that is coordinating," Cherny explained.The idea revolves around creating recurring workflows, or "loops," that enable AI agents to operate with greater independence. Rather than feeding instructions one step at a time, users define a goal and allow the AI system to continuously refine its own instructions until the task is completed.Several prominent figures in the AI industry appear to share a similar view. As per the Business Insider report, OpenAI-linked developer Peter Steinberger recently argued that users should stop focusing on prompting coding agents and instead build loops that guide those agents automatically. Meanwhile, Google's Addy Osmani described loop engineering as a combination of automations, connectors, plugins, skills, worktrees and specialised sub-agents working together.Supporters of the concept say loops allow AI systems to behave less like tools and more like employees. Claire Vo, founder of ChatPRD, compared the process to hiring a worker. "This is the time for the manager. You are designing a job. Just imagine you're onboarding an employee, that employee could be an assistant, a customer service agent, or a software engineer."Examples of these systems include one AI agent writing code while another reviews it, or autonomous coding tools repeatedly checking repositories and assigning new tasks without human intervention.However, the approach is not without challenges. Running multiple agents and sub-agents can dramatically increase AI costs. Experts warn that businesses may need to carefully balance automation with token usage and infrastructure expenses.Cherny also took aim at another popular AI buzzword: "vibe coding." While the term has become widely associated with AI-assisted programming, he argues it no longer reflects the complexity of modern AI development. For now, though, even Cherny admits he hasn't found a better replacement.