The way developers interact with AI coding assistants is fundamentally changing. Instead of writing detailed prompts to guide AI agents through each step, engineers are now designing automated loops that allow AI systems to generate and refine their own instructions. This shift represents a major departure from how people have worked with artificial intelligence over the past few years.

What Are Loops and Why Do They Matter?

Loops are recurring systems that guide AI agents without requiring constant human input. Rather than asking an AI model to complete one task and then waiting for the next instruction, loops enable agents to work more autonomously. For example, a simple command like /goal can instruct an AI model to keep working until a task is complete, rather than requiring step-by-step prompts from a human operator.

Peter Steinberger at OpenAI explained the practical shift in how developers should approach AI agents. "You shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents," Steinberger stated. This represents a fundamental change in how engineers think about their relationship with AI tools.

"You shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents," said Peter Steinberger.

Peter Steinberger, OpenAI

The concept resonates with how companies actually manage human employees. Claire Vo, founder of ChatPRD, drew this parallel: "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." In other words, instead of micromanaging an AI agent with constant instructions, developers now design the system and let it operate more independently.

How to Build Effective AI Loops: Key Components

Real-world implementations of loops are already emerging across the AI industry. Addy Osmani, director at Google Cloud, described the five key components that make loops function effectively:

• Automations: Processes that run without manual intervention to handle routine tasks and decisions.
• Worktrees: Organizational structures that help agents manage multiple branches of work simultaneously and track progress.
• Skills: Specific capabilities that agents can develop and apply to different problems across various domains.
• Plugins and connectors: Tools that allow agents to integrate with external systems, APIs, and services seamlessly.
• Sub-agents: Smaller AI agents that can be deployed to handle specialized tasks within the larger system architecture.

Practical examples of loops in action include scenarios where one agent writes code while another agent checks it for quality and correctness. Another example involves systems that maintain code repositories by periodically directing work into organized threads, ensuring continuous progress without human oversight.

What Are the Cost Implications of Loop Engineering?

While loops reduce the amount of human effort required to manage AI agents, they introduce a new concern: token budgets. Tokens are units of text that AI models process, and running multiple agents and sub-agents simultaneously can quickly become expensive. Each interaction between agents consumes tokens, which translates directly to operational costs.

To address this challenge, industry leaders have offered practical guidance. Steinberger advised using longer intervals, such as hourly or daily checks, rather than constant monitoring to reduce the number of tokens consumed. Addy Osmani from Google Cloud added an important caveat: sub-agents should only be deployed when a second opinion is genuinely worth the additional expense. This means companies need to carefully evaluate whether adding another agent to a loop will provide enough value to justify the cost.

Why Is This Shift Happening Now?

Boris Cherny, co-founder of Anthropic and creator of Claude Code, has been at the forefront of declaring that traditional software engineering practices are evolving. After previously stating that "software engineering is dead," Cherny now emphasizes that the era of manually writing AI prompts is ending. His perspective reflects a broader industry recognition that AI agents have become sophisticated enough to manage their own workflow generation.

"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," explained Boris Cherny.

Boris Cherny, Co-founder of Anthropic

At Anthropic, the shift is already underway. According to Cherny, no one at the company has touched a line of code manually since at least November, demonstrating that loop-based workflows are not theoretical but already operational in production environments.

The transition from prompt engineering to loop engineering represents a maturation of AI development practices. As AI agents become more capable, the bottleneck shifts from the agent's ability to execute tasks to the human's ability to design effective systems. Loop engineering addresses this by allowing developers to focus on architecture and system design rather than constant instruction writing. This shift has significant implications for how companies will hire, train, and deploy AI engineers in the coming years. The skills required to work effectively with AI agents are changing from prompt crafting to system design and loop architecture.