AI Agents Enter The Loop: Autonomous Systems Take Control

The AI world is getting 'loopy' as autonomous agents begin to work continuously in the background, improving code architecture, fixing bugs, and optimizing processes. This development marks a significant shift in the way AI agents are being utilized, from simple code generation to complex chains of tasks where agents assign tasks to one another and monitor the results. According to Cherny, a figure behind this technology, 'Two years ago, we wrote source code by hand. We started to transition so agents write the code. And now we’re transitioning to the point where agents are prompting agents that then write the code.'

The Rise of Agentic Loops

Agentic loops are a new concept in AI, where multiple agents work together in a continuous loop to improve processes. Cherny emphasizes that this technology is a significant step forward, allowing agents to work autonomously without human intervention. In traditional programming, recursive loops repeat until a specific condition is met. However, loops in the AI world operate differently, with the decision-making process not being deterministic; the sub-agent itself decides when to stop the work. For example, in a method called 'Ralph Loop,' the model accounts for all the work it has performed and independently verifies whether the set goal has been achieved.

Applications of Agentic Loops

Agentic loops have various applications, including improving code architecture, fixing bugs, and optimizing processes. According to Cherny, one agent is continually looking for ways to improve the code architecture, while another looks for duplicated abstractions that can be unified. They submit pull requests like any other coder, and since the code is constantly changing, they never stop running. This technology is expected to become a primary tool for major IT companies in the coming years, taking efficiency to a new level not only in programming but also in data analysis and other digital fields.

Enterprise AI and Governance

While public attention focuses on 'AI theater' like social media bots, the real impact of AI agents lies in quiet, enterprise applications. Companies like Pactum, co-founded by Kaspar Korjus, deploy agents to autonomously negotiate supplier contracts for major corporations such as Walmart and Henkel. These agents operate within strict guardrails, demonstrating undeniable value by closing significant deals and optimizing processes at scale. Effective enterprise AI requires layered governance, AI-driven supervision, and clear, non-conflicting objectives to prevent chaos and mitigate new cybersecurity risks.

The Future of AI Agents and Responsibility

As AI agents become more autonomous, the question of responsibility becomes increasingly important. When an AI agent sends an email to a regulator, who is responsible? The answer is not straightforward, and it requires a rethinking of control and governance in agentic systems. The provider typically controls the model architecture, training data, guardrails, and update cycles, while the customer controls the deployment context. These are not interchangeable, and a clear understanding of control is essential for managing risk and ensuring accountability.

Recursive Self-Improvement and the Future of AI Research

The development of agentic loops is part of a broader trend in AI research, including recursive self-improvement. Sakana AI has launched a Recursive Self-Improvement Lab, which aims to create AI systems that can improve themselves with less compute. This includes the development of systems like the AI Scientist, which can perform fully automated, open-ended scientific discovery, and LLM-Squared, which uses language models to design better training methods for other language models.

What this means

The emergence of agentic loops marks a significant shift in the way AI agents are being utilized, from simple code generation to complex chains of tasks where agents assign tasks to one another and monitor the results. This development has important implications for the future of AI research, enterprise applications, and governance. As AI agents become more autonomous, it is essential to ensure that they are designed and deployed in a way that is transparent, accountable, and secure. The future of AI depends on getting this right.