---
title: "AI Agents: From models to systems"
description: "Understand the mechanisms, compare real implementations, and follow the frontier of agent systems."
---

> Documentation Index
> Fetch the complete documentation index at: https://latentlit.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Agents: From models to systems

import AgentCards from '../../../components/agents/AgentCards.astro';
import AgentLandscape from '../../../components/agents/AgentLandscape.astro';
import FocusCanvas from '../../../components/FocusCanvas.astro';

An AI model can describe how to investigate a problem. An agent system can
work through it: choose an action, inspect the result and decide what to do next.
The useful unit is the **whole system: model, harness, tools, environment and
feedback**.[^react]

This collection connects the mechanisms with the systems people actually use.
Choose a route first; the chapters supply the depth.

**[Core Concepts](/agents/foundations/)** — Understand

Follow the agent loop, then study tools, context and coordination. Finish with a complete coding workflow.

**[Reliability & Control](/agents/evaluation-reliability/)** — Design for trust

Learn how to evaluate outcomes, recover interrupted work and bound what a system can change.

**[Agent Deep Dives](/agents/deep-dives/)** — Study implementations

Explore Claude Code, Codex, Pi, OpenCode and DeepSeek Harness, with an architectural comparison and frontier research.

## A map of the system

The model proposes decisions. The **harness** is the surrounding software that
assembles context, manages the loop and coordinates execution. Persistent
session records and execution environments can have different lifecycles;
they need not live inside one process.[^managed][^api]
[Figure 1](#figure-1) connects that work loop to the wider landscape: packaged
products, extensible harnesses, frameworks and managed runtimes. Select a label
to explore the corresponding chapter or study.

<a id="figure-1"></a>

**The agent landscape, at a glance**

**Human direction:** Goal · authority · acceptance criteria

**How agent work unfolds:** Manages context, actions and stopping.

1. **[Context](/agents/context-memory-state/)**: Goal, instructions and relevant evidence.
2. **[Model](/agents/foundations/)**: Choose an action or response.
3. **[Action](/agents/tools-skills-protocols/)**: Tools act in files, apps or services.
4. **[Evidence](/agents/evaluation-reliability/)**: Observe results; check the outcome.

**Feedback:** Results update context: continue, revise or stop.

**Inside the harness**

- **[State & memory](/agents/context-memory-state/)**: Context, history and recovery.
- **[Coordination](/agents/orchestration/)**: Workflows and scoped workers.
- **[Skills & integrations](/agents/tools-skills-protocols/)**: Procedures, protocols and plugins.

**Across every step**

- **[Control throughout](/agents/security-control/)**: Bound access and actions.
- **[Evaluation throughout](/agents/evaluation-reliability/)**: Check outcomes, cost and reliability.

**Where systems sit**

- **[Coding products](/agents/deep-dives/)**: Package an end-user workflow. Examples: [Claude Code](/agents/deep-dives/claude-code/), [Codex](/agents/deep-dives/codex/), [OpenCode](/agents/deep-dives/opencode/).
- **[Extensible harnesses](/agents/ecosystem/#reusable-harnesses-and-frameworks)**: Provide reusable agent machinery. Examples: [Pi](/agents/deep-dives/pi/), [DeepSeek Harness](/agents/deep-dives/deepseek-harness/), [Deep Agents](/agents/ecosystem/#reusable-harnesses-and-frameworks).
- **[Frameworks & SDKs](/agents/ecosystem/#reusable-harnesses-and-frameworks)**: Build an application-owned system. Examples: [LangGraph](/agents/ecosystem/#reusable-harnesses-and-frameworks), [Google ADK](/agents/ecosystem/#reusable-harnesses-and-frameworks), [Agent SDKs](/agents/ecosystem/#reusable-harnesses-and-frameworks).
- **[Managed runtimes](/agents/ecosystem/#managed-runtime-ownership)**: Operate harness and session infrastructure. Examples: [OpenAI Agents API](/agents/ecosystem/#managed-runtime-ownership), [Anthropic Managed Agents](/agents/deep-dives/#durable-sessions-beyond-one-running-process).

**Frontier questions**

- [Durable sessions](/agents/deep-dives/#durable-sessions-beyond-one-running-process)
- [Evolving context](/agents/deep-dives/#evolving-context-with-accountable-updates)
- [Useful coordination](/agents/deep-dives/#coordination-that-fits-the-task)
- [Outcome feedback](/agents/deep-dives/#feedback-that-can-actually-distinguish-improvement)

Roles overlap; examples are not a quality ranking. DeepSeek Harness is a developer preview.

**Figure 1.** Latent Light synthesis: the work loop, overlapping ecosystem roles and frontier questions. Implementation and research sources are collected in the [deep-dive comparison](/agents/deep-dives/).[^react][^managed][^api][^landscape]

A fixed workflow chooses its path in advance. An agent delegates more path
selection to the model. Many useful systems combine both: deterministic steps
handle known work, and a model handles uncertainty.[^patterns] More freedom is
useful only when the system can observe and verify what it does.

## Learn the mechanisms

| Question | Chapter |
| --- | --- |
| How does a response become a sequence of actions? | [How Agents Work](/agents/foundations/) |
| What can the system do, and how does it connect to other components? | [Tools & Skills](/agents/tools-skills-protocols/) |
| What is visible now, what persists, and how does work resume? | [Context & Memory](/agents/context-memory-state/) |
| When should control live in code, one agent or several workers? | [Workflows & Coordination](/agents/orchestration/) |
| What does the whole process look like on a repository change? | [A Coding Workflow](/agents/deep-dives/coding-agents/) |
| How do we know the outcome is correct and repeatable? | [Evaluation](/agents/evaluation-reliability/) |
| How do we bound access, actions and external effects? | [Security & Control](/agents/security-control/) |

For a first reading, follow the first three chapters. Before designing a
system that changes external state, read evaluation and control as well.
The coding workflow connects those ideas in one worked example.

## Study the systems

The [deep-dive overview](/agents/deep-dives/) compares the five initial studies
by execution ownership, persistence, extension mechanisms and control. It also
connects them to the [broader ecosystem](/agents/ecosystem/), including frameworks,
SDKs and managed runtimes.

The studies ask different questions. What does Claude Code package into its
workflow? How does Codex expose a controllable session to clients? What does Pi
leave to its host and extensions? What does OpenCode's client/server boundary
make possible? What changes when DeepSeek makes the loop itself a plugin?
Follow a system from its design to its consequences, rather than counting features.

## Follow the frontier

The [frontier discussion](/agents/deep-dives/#frontier) examines durable sessions,
evolving context, coordinated workers and feedback that can genuinely grade
an outcome. Research findings and vendor engineering accounts have different
limits; a documented feature does not establish reliable task completion.

The individual studies identify their inspected documentation or repository
snapshot. Source checks for the five studies are **October 11, 2026**. The
broader directory retains its separately dated checks. These pages explain
implementations and research; they do not report a new head-to-head benchmark.

[^react]: Shunyu Yao et al. [ReAct: Synergizing Reasoning and Acting in Language Models](https://arxiv.org/abs/2210.03629v3), ICLR 2023.
[^patterns]: Anthropic. [Building effective agents](https://www.anthropic.com/engineering/building-effective-agents), 2024; workflows and agents.
[^managed]: Anthropic. [Scaling Managed Agents: Decoupling the brain from the hands](https://www.anthropic.com/engineering/managed-agents), April 8, 2026. Session, harness and sandbox responsibilities; checked October 11, 2026.
[^api]: OpenAI. [Agents API overview](https://developers.openai.com/api/docs/guides/agents-api/overview), managed harness and application integration; checked October 11, 2026.

[^landscape]: Latent Light. [Agent Deep Dives](/agents/deep-dives/), synthesis of the five inspected implementation studies, ecosystem roles and bounded frontier evidence; source checks October 11, 2026. Categories overlap and do not indicate measured quality.

Source: https://latentlit.com/agents/index.mdx
