Artificial intelligence is entering a new phase. For the past few years, prompt engineering has been seen as one of the most valuable AI skills. Writing clear instructions helped large language models produce better results. Now, many AI experts believe the next evolution has arrived through AI Loops.
The idea gained attention after Boris Cherny, creator of Claude Code, explained that he no longer spends his time writing prompts. Instead, he designs automated loops that guide AI agents until a task is complete. Around the same time, OpenAI engineer Peter Steinberger shared a similar view, suggesting developers should design loops that prompt AI agents instead of prompting them directly.
What Is Prompt Engineering?
Prompt engineering is the process of writing instructions for an AI model. Every request contains the information the model needs, including its role, the task, examples and the desired output.
Once the AI responds, the process ends. If something is wrong, the user creates another prompt and tries again.
This approach works well for individual tasks such as writing content, answering questions or generating code. It gives people direct control over every interaction and is simple to understand.
However, prompt engineering relies heavily on human involvement. Users must constantly review responses, provide corrections and decide what happens next.
What Are AI Loops?
AI Loops move that responsibility from the user to an automated system.
Rather than waiting for a person to submit another prompt, the loop decides what should happen next. It sends prompts to the AI, checks the results, runs tools where needed and repeats the process until a predefined goal has been achieved.
Importantly, the loop does not stop simply because the AI says it has finished. Instead, it continues until objective checks confirm success, such as software tests passing or another verification process approving the result.
This changes the role of the human from prompt writer to system designer.
Pros and Cons
Prompt engineering remains valuable because it is straightforward, flexible and inexpensive. It is ideal for one off tasks where human judgement is important.
Its biggest weakness is that it does not scale. Every new task usually requires another prompt and another round of human review.
AI Loops are designed for repetitive, multi step workflows. They can automate coding, testing, customer support, quality assurance and many other business processes. Once configured, they require far less human intervention.
The trade off is complexity. Loops need triggers, completion checks, memory, safety controls and carefully selected tools. Without these safeguards they can waste computing resources, consume large numbers of AI tokens and even continue working long after they should have stopped.
Helping AI Run on Commodity Hardware
One important advantage of AI Loops is that they can help organisations make better use of commodity hardware.
Instead of relying on increasingly larger AI models running continuously on expensive infrastructure, loops allow smaller models to work iteratively. A model can complete one step, verify the outcome, use external tools where appropriate and continue only when necessary.
This means businesses can often achieve complex workflows using more affordable hardware rather than relying entirely on the largest frontier models. Processing can also be distributed across multiple machines, making AI systems more accessible to organisations without specialist computing resources.
While larger models still offer advantages for many tasks, intelligent orchestration through AI Loops can reduce unnecessary processing and improve efficiency.
A New AI Skill
Prompt engineering is unlikely to disappear. It remains the foundation of every interaction with an AI model.
However, as AI agents become more capable, experts increasingly believe the competitive advantage will come from designing the systems that manage those prompts automatically.
For developers and businesses alike, AI Loops represent a shift from manually directing every conversation to building intelligent systems that can plan, verify and complete work with minimal supervision. That makes loop engineering one of the most important AI skills to watch as autonomous AI continues to evolve.
