Analyst Miles Deutscher called the recently published internal guide from Anthropic on so-called loop engineering—building AI agent work cycles—the most valuable artificial intelligence guide he has read in the past year. The document, although its leak has not yet been officially confirmed, has already sparked heated discussion in the professional community.

Deutscher highlighted five key principles on which each such cycle is built. The first is detection: the agent independently finds tasks for itself through test failures, open tasks, and commits. The second is isolation: each task receives a separate working branch in the version control system.

The third principle is verification: the code generator should never evaluate its own work. The fourth is resilience: saving state to disk so that it persists between runs. The fifth is scheduling: the system operates on a timer, even while the user is asleep.

Deutscher called the separation of generator and evaluator the most important rule. It is necessary to use two agents: one writes code, while the second acts as a skeptical judge. The evaluator must take action—run tests, press buttons, take screenshots. According to him, this is what truly cuts off poor results.

Deutscher also listed the risks of this approach: "verification debt," loss of understanding of one's own codebase, explosive growth in token costs, and "cognitive capitulation"—when a person stops thinking, relying on the idea that "the cycle will handle it on its own."

To control costs, the analyst proposed a "barbell" approach based on the 80/20 principle. For the most complex tasks requiring maximum intelligence, expensive models like Opus should be used, while the remaining 80% of routine work should be performed by cheap open-source models within the Claude Code toolkit. As an example for code execution, he cited the GLM-5.2 model.

Karpathy's Personal Settings File

Engineer Khairallah al-Awadi raised a similar topic. According to him, AI developer Andrej Karpathy joined Anthropic five weeks ago, and an acquaintance from his team showed him the Claude.md file that he supposedly uses in his work. Al-Awadi integrated this file into his configuration, and the model's very first response was completely different. According to him, Claude stopped giving generic answers and began working in the logic that the user themselves thinks in.

The discussions reflect a real trend: developers are increasingly building automated work cycles for AI agents and seeking ways to improve their efficiency. Key topics remain the division of roles between models, quality control, and managing token costs.

Cryptalist Expert Opinion: Loop engineering is a natural stage in the evolution of AI agents. Separating the generator and evaluator not only improves code quality but also creates an architectural foundation for scalable autonomous systems. However, "cognitive capitulation" is a real threat: the smarter agents become, the higher the risk that developers will lose critical thinking. Cost control and conscious task distribution between models will become key skills in 2024-2025.