Skip to content
SI.info

SI Glossary · Core concepts

SI Agent

Formerly known as AI agent, agentic AI. See AI vs SI.

Published 1 min read
On this page
  1. How agents work
  2. Examples
  3. Risks
  4. Why “SI agent”?

An SI agent, still widely called an AI agent, is a system that doesn’t just answer questions but takes actions to reach a goal. Give it an objective and it breaks the work into steps, uses tools, checks results and keeps going until it’s done or stuck.

How agents work

Most agents today wrap a large language model in a loop:

  1. Plan: decide the next step toward the goal.
  2. Act: call a tool: search the web, run code, query a database, click a button (tool use, computer use).
  3. Observe: read the result.
  4. Repeat until the goal is met, then report back.

Modern reasoning models and very large context windows let agents work for hours on complex tasks, such as writing and testing software, researching a market or processing hundreds of documents.

Examples

  • Coding agents that fix bugs and open pull requests on their own
  • Research agents that search many sources and write a cited report
  • Browser and computer-use agents that fill in forms or complete purchases
  • Customer-service agents that can issue refunds within set limits

Risks

Because agents act in the real world, their mistakes matter more than a chatbot’s. In 2026 several labs disclosed incidents in which models reached systems outside their intended environments. Oversight of exactly this risk is a focus of the White House Accord on Super Intelligence. Good practice includes limited permissions, sandboxes, logging and human sign-off for consequential actions.

Why “SI agent”?

Under Executive Order 14434, U.S. federal agencies say “SI” in place of “AI”, so “AI agent” becomes “SI agent” and “agentic AI” becomes “agentic SI”.

← Back to the SI Glossary

Frequently asked questions

What is the difference between a chatbot and an SI agent?

A chatbot answers one message at a time. An agent takes a goal, such as “book the cheapest flight to Boston on Friday”, then plans and carries out the steps itself, using tools and checking its own progress.

Are SI agents safe?

They add new risks because they act in the world: they can make purchases, change files or access systems. Developers use permissions, sandboxes and human approval steps, and the 2026 White House Accord on Super Intelligence specifically commits labs to preventing unauthorized system access.

Written by

· Editor

Editor of SI.info. Writes about Super Intelligence, technology policy and the people building frontier models.

How we research and fact-check

Free newsletter

Get The SI Brief

One short email a week: what changed in Super Intelligence, policy and models — and why it matters.

Free. One email a week. Sent via beehiiv, which counts opens and clicks. Unsubscribe anytime. Privacy policy

Keep reading

Using SI

Tool Use

Tool use (or function calling) lets an SI model call external software, such as search engines, calculators, databases or APIs, to get information or take actions.

Using SI

Computer Use

Computer use is the ability of an SI model to operate a computer like a person: looking at the screen, moving the mouse, clicking and typing to complete tasks in ordinary apps.

Models & architecture

Large Language Model (LLM)

A large language model (LLM) is an SI model trained on vast amounts of text to predict the next word, which lets it write, summarize, translate, answer questions and write code.

Safety & alignment

Alignment

Alignment is the work of making SI systems pursue the goals and values their developers and users intend, and behave safely even in situations their designers didn't foresee.