Self-reflection is the ability of an AI system to evaluate its own reasoning, errors and decisions to improve the next execution. Instead of just generating an output, the system analyzes what worked, what did not work and how it could do differently on the next attempt.
How it works
The system generates a task, receives a response, and then critically evaluates that response. It asks: is this output plausible? Was there bias? Could it be better? Based on that evaluation, the system adjusts its next output or repeats the task with different guidance.
Practical example
A code agent writes a function, runs the tests and checks if the tests passed. If it fails, the agent analyzes the error, fixes the solution and re-runs it, until it reaches the expected result.
What it's used for
Reflection is essential for systems that need to operate autonomously, such as agents for complex tasks, programming assistants and reasoning models. It allows AI to learn from mistakes and improve over time without constant human supervision.
Related terms
- agente_de_ia: system that executes tasks with autonomy
- raciocinio: ability to make logical deductions
- verificacao: process of validating outputs
How to test today
Define a simple problem, ask an AI to solve it and then ask it: what did you do wrong and how would you improve? Compare the previous answer with the new one.
Frequently asked questions
- How does an agent reflect on its own reasoning? It generates the response, inspects it and evaluates the absence of contradictions, missing factors and possible improvements.
- Does reflection always improve the result? In general yes, but it can consume more tokens and time. In simple tasks, the overhead may not be worth it.
- This term is proposed by the site footer key: In agent systems, reflecting on results allows smooth evolution and reduces the need for manual correction.