AI Agents and Axiomate
This chapter is intended for readers new to AI agents. It introduces three fundamental concepts: what AI is, what an agent adds on top of it, and what Axiomate is. Specific features are covered in later chapters.
What is AI?
The AI referred to here is a system powered by a large language model. Given text, code, or an error log, it interprets the intent and produces an answer, a plan, or a follow-up suggestion.
Its strengths lie in processing language and knowledge: understanding, summarizing, explaining, and planning. It has two inherent limitations, however—it cannot perceive the current state of a project, and it cannot act on its own, neither reading files nor executing commands by itself. To make it genuinely useful for software engineering, it must be connected to the repository, the terminal, and real tools. This is precisely what an agent addresses.
What is an AI agent?
A standard AI conversation follows a question-and-answer format; an AI agent, by contrast, works toward a goal autonomously once it is given, until that goal is complete.
The core difference is that an agent operates through an execution loop: understand the goal, call tools to gather information or take action, evaluate the result, decide the next step, and continue iterating until the goal is met. Rather than merely advising, it carries out the work.
For example, given the goal "fix the test that keeps failing," an agent reads the relevant code, identifies the cause of failure, applies the change, re-runs the test to confirm it passes, and presents the diff for review—all without step-by-step direction. The user's role is to set the goal and evaluate the result.
The ability to act also entails responsibility: an agent modifies files and runs commands, and therefore requires a clear goal, appropriate permissions, and a means to review its actions and undo them when necessary.
What is Axiomate?
Axiomate is an AI agent CLI that runs in the terminal and is purpose-built for software engineering. Given a goal, it reads context, calls tools, edits code, and runs verification within the repository—applying the execution loop above to a real project.
Axiomate is designed around two essential principles:
- Capable: it supports connecting any model and provider (with no lock-in to a single service) and provides full coding tools, desktop control, skills, plugins, and MCP extensions to handle real tasks.
- Controllable: permissions, checkpoints, and explicit tool calls keep every step visible, reviewable, and reversible—a prerequisite for bringing an agent into a real engineering workflow.
In summary, Axiomate serves as an engineering collaborator in the terminal: the user sets the goal, exercises judgment, and makes the final decision, while Axiomate handles context, execution, and moving the work forward.
Later chapters cover installation, model configuration, permissions, tool usage, and more specific use cases.