FACT WIKI · Concept
AI agents
AI agents
AI agents are software systems that interpret context and a user's request, plan actions, carry out tasks autonomously and adapt when necessary. Their role extends beyond answering prompts to performing work.
How it works
Coding agents can invoke external development tools, including compilers, debuggers, performance profilers and version control systems. These connections let them interact with a development environment throughout a software task.
An agent's task loop involves evaluating a goal, breaking it into steps, using tools or connected systems to act, observing the result and adjusting the plan when necessary.
Scope and limitations
LLM-based coding agents can generate programs or modules from natural-language specifications, diagnose and fix bugs using compiler or test feedback, write and execute tests, and refactor code for readability or performance.
Background and related concepts
The survey AI Agentic Programming identifies long-context limitations, a lack of persistent memory across tasks, safety, alignment with user intent and collaboration with human developers as challenges for coding agents. These are broader research concerns rather than an explanation of any particular agent's actions.
The survey AI Agentic Programming identifies a difficulty with development tools designed primarily for human users: abstractions can hide the internal states and transformation steps an agent needs to understand a failure. An error message alone may not explain which code change caused a build to fail.
References
Related stories
FactSnap stories are secondary summaries.