alderson Sign up Contact Download
Menu

Articles

01 Core concepts

What is an AI agent?

An agent lets a model choose tools and direct a multi-step process using feedback from its environment.

8 min read
02 Architecture

What is an agent harness?

A harness supplies the instructions, guardrails, context, and runtime controls for model-directed work.

8 min read
03 Interfaces

What are agent tools?

Tools turn structured model requests into validated, authorized operations in deterministic software.

8 min read
04 Security

What is agent sandboxing?

How process isolation, containers, userspace kernels, and microVMs restrict untrusted agent-controlled code.

9 min read
05 State

What is agent memory?

Context, run state, and long-term memory store different information and require different controls.

8 min read
06 Intelligence

What is a world model?

A world model predicts planning-relevant changes in an environment from a state or history and a possible action.

8 min read
07 Quality

What are agent evaluations?

Agent evaluation tests task outcomes, required constraints, and consistency across controlled trials.

8 min read
08 Operations

What is agent observability?

Traces record operations and confirmed effects while limiting collection of prompts and customer content.

8 min read