Artificial intelligence tools are everywhere today—from chatbots and research assistants to coding copilots.
But anyone who uses them regularly knows they sometimes repeat the same mistakes, hallucinate information, or follow inefficient workflows.
A prompt circulating online claims to tackle that problem head-on.
Shared by tech enthusiast @PerSolana, the prompt proposes a structured “self-improvement system” that allows an AI agent to track errors, learn from them, and gradually refine its reasoning process.
The idea is simple but powerful: instead of ignoring mistakes, the AI records them, analyzes them, and converts them into permanent lessons.
Turning Mistakes Into Data
The first step in the system is error logging. Whenever the AI recognizes that it made a mistake—whether it’s faulty reasoning, a hallucinated fact, or a flawed strategy—it records the issue in a file called mistakes.md.
Each entry includes three elements:
-
What went wrong
-
Why it happened
-
The lesson that should prevent it from happening again
In software engineering and machine learning, logging errors is a well-known technique used to improve systems over time.
Major tech companies like OpenAI, Google, and Microsoft rely heavily on monitoring and error tracking to refine AI models and digital services.
Extracting Lessons From Every Failure
After logging a mistake, the system instructs the AI to transform the insight into a reusable rule stored in lessons.md.
Instead of remembering just one error, the AI creates a general principle that can apply to future situations. For example:
-
A mistake about citing unreliable sources could become a rule about verifying sources before answering.
-
A reasoning failure might become a rule about double-checking assumptions before making conclusions.
This approach mirrors how humans learn.
Cognitive scientists describe learning as the process of converting specific experiences into general patterns that guide future decisions.
Building a Core Set of Principles
Not every lesson deserves to become a permanent rule.
That’s why the system introduces a third file: soul.md.
This file stores only foundational behavioral principles—the guiding rules that shape how the AI thinks and operates.
Examples might include:
-
Prioritize accuracy over speed
-
Always verify uncertain claims
-
Break complex problems into smaller steps
By keeping this file focused on core principles rather than minor tweaks, the AI maintains a clear reasoning framework rather than drowning in thousands of tiny rules.
Spotting Patterns in Repeated Errors
Another key feature of the system is pattern detection.
The AI periodically reviews the mistakes.md log to identify recurring problems.
If several errors share the same root cause—such as misinterpreting ambiguous questions—the AI can create a higher-level rule that prevents that entire category of mistakes.
In many ways, this resembles how machine learning engineers analyze model failures to improve performance.
Instead of fixing individual errors, they redesign systems to eliminate the root cause.
Optimizing Workflows for Better Results
The prompt also introduces playbook.md, a file dedicated to documenting improved workflows and strategies.
For instance, if the AI discovers that a structured research process produces better answers than quick guesses, that process becomes part of its playbook.
In business and technology, this approach is often called knowledge management—turning successful methods into repeatable systems.
Companies from startups to global firms rely on similar playbooks to maintain consistent performance across teams.
Impact and Consequences
If implemented effectively, a system like this could reshape how AI assistants operate:
-
AI agents could improve their reasoning quality over time rather than repeating mistakes.
-
Developers might create more transparent systems where errors and lessons are documented.
-
AI workflows could become more efficient as successful strategies accumulate.
-
The concept may inspire new approaches to AI training and autonomous agents.
-
However, poorly designed self-improvement systems could introduce biased or incorrect rules if mistakes are misinterpreted.
What’s Next?
The idea of self-improving AI is already gaining attention in research labs and tech communities. Autonomous AI agents capable of analyzing their own behavior are becoming a major focus in artificial intelligence development.
Future systems could combine structured prompts like this with advanced reinforcement learning, enabling AI models to refine themselves continuously.
However, experts emphasize that human oversight will remain essential.
Without careful supervision, automated self-learning systems might reinforce flawed assumptions or drift away from accurate reasoning.
Summary
A viral prompt shared online proposes a structured system for making AI agents smarter over time by turning mistakes into lessons.
The process involves logging errors, extracting principles, detecting patterns, and refining workflows through dedicated documentation files.
While the concept reflects practices already used in software engineering and machine learning, it highlights a growing interest in AI systems that can continuously improve their reasoning and performance.
Key Takeaways
-
A viral prompt suggests a framework for continuous AI self-improvement.
-
The system records errors in mistakes.md and converts them into lessons in lessons.md.
-
Foundational principles are stored in soul.md to guide core reasoning.
-
Pattern detection helps prevent entire categories of recurring mistakes.
-
Workflow improvements are documented in playbook.md to boost efficiency.
-
The concept reflects broader trends toward autonomous, self-learning AI systems.
