---
title: 'Learning path: language models and agents'
url: https://doc.liz6.com/en/ai/00-learning-path
locale: en
area: ai
tags:
- LLM
- Agent
- Models & agents
date: 2026-09-09
modified: 2026-09-09
description: Learn generation and resource use, establish input/output and evidence contracts for one request, build a recoverable agent, then validate the complete system. Basic programming experience is assumed; the articles explain their matrix and probability examples. Deployment notes remain in Homelab.
---

# Learning path: language models and agents

Learn generation and resource use, establish input/output and evidence contracts for one request, build a recoverable agent, then validate the complete system. Basic programming experience is assumed; the articles explain their matrix and probability examples. Deployment notes remain in Homelab.

## Model foundations

1. [Tokens, probabilities and sampling](/ai/01-model-foundations/01-tokens-and-sampling)
2. [Attention, feed-forward networks and MoE](/ai/01-model-foundations/02-attention-and-moe)
3. [Inference, KV cache and memory](/ai/01-model-foundations/03-inference-and-kv-cache)
4. [Inference-time compute, candidates and verification](/ai/01-model-foundations/04-reasoning-and-verification)

## Requests and evidence

5. [Prompts and output contracts](/ai/02-context-and-interfaces/01-prompt-and-output-contracts)
6. [Context engineering](/ai/02-context-and-interfaces/02-context-engineering)
7. [Retrieval, RAG and evidence](/ai/02-context-and-interfaces/03-retrieval-and-evidence)

## Agent execution systems

8. [Agent loops and executors](/ai/03-agent-systems/01-agent-loop)
9. [Tool interfaces and MCP](/ai/03-agent-systems/02-tools-and-mcp)
10. [Skills and reusable task methods](/ai/03-agent-systems/03-skills-and-task-context)
11. [Memory, state and recovery](/ai/03-agent-systems/04-memory-and-recovery)
12. [Authority, trust and tool boundaries](/ai/03-agent-systems/05-authority-and-tool-boundaries)
13. [Multi-agent orchestration and integration](/ai/03-agent-systems/06-multi-agent-orchestration)

## Evaluation and production

14. [Evaluation, trials and observability](/ai/04-evaluation-and-production/01-evaluation-and-observability)
15. [Cost, capacity and reliability](/ai/04-evaluation-and-production/02-cost-performance-and-reliability)
16. [Release validation, canaries and recovery](/ai/04-evaluation-and-production/03-release-and-recovery)

## How to use this path

For an application-first route, start with output contracts and return to model foundations when sampling or memory questions arise. Validate single requests before adding tools; establish a single-agent baseline, recovery and authorization before orchestration. Read evaluation early when defining acceptance criteria.

Use interactions for computable mechanisms and failure boundaries, and static diagrams for overviews. Predict an outcome, change a parameter or action order, then explain the result using the formula or code. The inline models require no model download and do not benchmark a real service.