Boris Cherny, co-founder of Anthropic, predicts that manual prompt engineering is becoming obsolete, to be replaced by a methodology he describes as loop engineering. In this approach, users shift from crafting specific prompts to designing systemic workflows that enable AI agents to manage instructions independently. Cherny explains the process by noting that he no longer writes prompts directly; instead, he interacts with an AI agent that coordinates and generates the necessary instructions for other AI systems to execute tasks autonomously.

Loop engineering relies on creating recurring workflows or loops that allow AI to refine its own instructions toward a defined goal. Industry figures supporting this transition include developer Peter Steinberger, who suggests building loops to guide coding agents automatically, and Google's Addy Osmani, who defines the practice as an integration of automations, connectors, plugins, skills, and specialized sub-agents. Claire Vo, founder of ChatPRD, likens this management-focused workflow to onboarding a human employee, where the user acts as a manager defining roles for assistant, customer service, or software engineering agents.

Practical implementations of these systems involve multiple agents performing collaborative tasks, such as one agent writing code while another reviews it, or autonomous tools continuously monitoring repositories to assign tasks without human intervention. While proponents see these loops as a way for AI to operate more like independent staff, the model faces significant operational challenges. Experts warn that utilizing multiple agents and sub-agents can substantially increase AI costs, requiring businesses to manage token usage and infrastructure expenses carefully. Additionally, Cherny argues that the popular term "vibe coding" no longer adequately describes the technical complexity required for modern AI development, though a universally accepted replacement term remains elusive.

Boris Cherny, co-founder of Anthropic, predicts that manual prompt engineering is becoming obsolete, to be replaced by a methodology he describes as loop engineering. In this approach, users shift from crafting specific prompts to designing systemic workflows that enable AI agents to manage instructions independently. Cherny explains the process by noting that he no longer writes prompts directly; instead, he interacts with an AI agent that coordinates and generates the necessary instructions for other AI systems to execute tasks autonomously.

Loop engineering relies on creating recurring workflows or loops that allow AI to refine its own instructions toward a defined goal. Industry figures supporting this transition include developer Peter Steinberger, who suggests building loops to guide coding agents automatically, and Google's Addy Osmani, who defines the practice as an integration of automations, connectors, plugins, skills, and specialized sub-agents. Claire Vo, founder of ChatPRD, likens this management-focused workflow to onboarding a human employee, where the user acts as a manager defining roles for assistant, customer service, or software engineering agents.

Practical implementations of these systems involve multiple agents performing collaborative tasks, such as one agent writing code while another reviews it, or autonomous tools continuously monitoring repositories to assign tasks without human intervention. While proponents see these loops as a way for AI to operate more like independent staff, the model faces significant operational challenges. Experts warn that utilizing multiple agents and sub-agents can substantially increase AI costs, requiring businesses to manage token usage and infrastructure expenses carefully. Additionally, Cherny argues that the popular term "vibe coding" no longer adequately describes the technical complexity required for modern AI development, though a universally accepted replacement term remains elusive.