As AI-powered software development rapidly evolves, the term "loop engineering" has emerged as one of the technology industry's latest buzzwords. In an open letter, Andrew Ng, co-founder of Google Brain, explained how AI agents are transforming software development through continuous feedback loops—while emphasizing that human developers remain indispensable.

The discussion follows viral comments from Boris Cherny, creator of Claude Code, and Peter Steinberger, who argued that traditional coding is giving way to AI-driven software creation.

What Is Loop Engineering?

According to Ng, loop engineering is a development approach in which AI systems continuously generate, test, refine, and improve software through multiple feedback cycles instead of relying solely on manual programming.

He outlines three interconnected loops that power modern AI-assisted software development.

1. Agentic Coding Loop

The Agentic Coding Loop forms the foundation of AI-driven programming.

In this workflow:

• Developers provide a product specification and, optionally, evaluation benchmarks.
• An AI coding agent writes the software.
• The agent automatically tests its own code.
• It detects bugs and repeatedly revises the program until it satisfies the required specifications.

Ng describes this as a major breakthrough because AI agents can now work autonomously for extended periods without constant human supervision.

As an example, he shared that while building a typing application for his daughter, his AI coding agent independently wrote code, tested it through a web browser, and refined the application for nearly an hour before requiring further input.

2. Developer Feedback Loop

The second stage is the Developer Feedback Loop, where humans shift from writing every line of code to acting more like product managers or creative directors.

Rather than manually identifying coding errors, developers now focus on:

• Refining product specifications.
• Improving user interface design.
• Prioritizing new features.
• Making architectural decisions.
• Guiding the AI toward the desired product vision.

Ng noted that as coding agents become better at self-testing, developers spend significantly less time performing traditional quality assurance and more time making higher-level design decisions.

He argues that humans possess a "context advantage"—a deeper understanding of users, business objectives, and real-world requirements—that AI currently lacks.

3. External Feedback Loop

The third and slowest stage is the External Feedback Loop, where software is evaluated in real-world environments.

This includes:

• Sharing prototypes with friends.
• Conducting alpha testing.
• Running A/B experiments.
• Deploying products to production environments.
• Collecting user behavior and customer feedback.

Insights gathered from real users help refine the overall product vision, which then informs updated specifications for the AI coding agent, creating a continuous cycle of improvement.

Humans Still Play a Critical Role

Despite rapid advances in autonomous coding, Ng rejects the notion that software engineers are becoming obsolete.

Instead, he believes developers are evolving into professionals who:

• Define product strategy.
• Translate ideas into AI-readable specifications.
• Interpret customer needs.
• Balance technical implementation with user experience.
• Incorporate knowledge unavailable to AI systems.

While AI increasingly assists with analyzing customer feedback, usage data, and competitor products, Ng argues that human judgment remains essential because people possess richer contextual understanding than current AI models.

Why It Matters

The emergence of loop engineering reflects a broader shift in software development:

• AI agents increasingly automate coding, debugging, and testing.
• Human developers spend more time on product vision and strategic decisions.
• Continuous feedback—from both developers and users—becomes central to building better software.
• Engineering and product management roles are becoming more closely integrated.

Rather than signaling the end of programming, Ng's framework suggests that the future of software development lies in human-AI collaboration, where AI accelerates implementation while humans provide direction, creativity, and contextual decision-making.