NEW DELHI: The next big breakthrough in coding may not come from writing better prompts, it could come from teaching AI how to improve itself. As artificial intelligence takes on a larger share of software development, Google Brain co-founder Andrew Ng believes the industry’s focus is shifting from prompt engineering to what he calls “loop engineering,” a system of feedback loops that enables AI agents to repeatedly test, refine and improve their own work.
In an open letter shared on X, Ng described loop engineering as one of the most important emerging ideas in AI-powered software development, arguing that the future of coding lies in designing iterative learning systems rather than relying on one-off interactions with AI.
The concept has gained momentum in recent weeks following discussions by Boris Cherny, creator of Claude Code, and Peter Steinberger, creator of OpenClaw, both of whom have argued that autonomous AI agents are fundamentally changing how software is built.
According to Ng, modern AI development now revolves around three interconnected feedback loops, allowing software to evolve through continuous refinement rather than a single coding session.
The first, the Agentic Coding Loop, enables developers to provide AI agents with product specifications and evaluation datasets. The AI then writes code, tests it, identifies bugs and repeatedly improves the software until it satisfies the desired requirements.
Ng said this approach has dramatically expanded the amount of development work AI systems can perform independently. He cited a recent example in which an AI coding assistant spent almost an hour autonomously building and refining a typing application for his daughter, repeatedly testing the software through a web browser before human intervention became necessary.
The second stage, the Developer Feedback Loop, shifts engineers away from routine debugging towards higher-level product decisions. Instead of fixing code, developers increasingly focus on refining product specifications, prioritising features, designing user interfaces and steering the overall direction of applications.
Ng argued that while AI can generate code and analyse large volumes of customer feedback, humans continue to possess what he described as a “context advantage,” an understanding of users, business priorities and operating environments that current AI systems cannot fully replicate.
The final stage, the External Feedback Loop, extends development beyond engineering teams by incorporating insights from early users, alpha testing, A/B experiments and live production environments. Although this process unfolds more slowly than AI-generated coding, Ng said it provides the real-world feedback needed to shape future product iterations.
The emergence of loop engineering also reflects a broader evolution in the software industry, where developers are increasingly expected to bridge product strategy, customer understanding and technical execution rather than simply write code.
As AI agents become more capable of testing, debugging and refining their own software, Ng suggests the competitive edge will belong less to those who write the best prompts and more to those who design the smartest systems for continuous improvement.
“Loop engineering” is a hot buzzphrase after mentions of it by Boris Cherny (Claude Code’s creator) and Peter Steinberger (OpenClaw's creator) went viral on social media. Loops are now a key part of how we get AI agents to iterate at length to build software. In this letter, I’d… pic.twitter.com/bhuRw8lrFC
— Andrew Ng (@AndrewYNg) June 30, 2026