llearn

Independent applied practice

Practical AI Guide

Nine topics from an applied learning effort: learn frontier AI in depth, then turn an LLM into a reliable working partner in any domain.

Start at the stage that matches your work, or read the guide in order. Add only the smallest change that earns its keep.

Level 1 · Conversation

Clarify what a model can do for a specific task, then communicate and evaluate the work well.

  1. Model Analysis & Selection

    Choose by goal, privacy, capabilities, reliability, and cost.

  2. Prompt Engineering

    Use communication, domain knowledge, and anti-sycophancy guardrails to steer work well.

Level 2 · Agent-assisted work

Move from a conversation to files, tools, APIs, and appropriate safety boundaries.

  1. Harness Engineering

    Choose the tool layer around a model according to workflow, cost, and control.

  2. LLM Sandboxing

    Choose a safety boundary appropriate to the model’s access to files and data.

  3. MCP

    Understand where the Model Context Protocol sits relative to direct APIs.

Level 3 · Connected knowledge

Keep trusted sources and decisions available without flooding a model’s active context.

  1. Context Engineering

    Keep active model context small and retrieve only the information the task needs.

  2. RAG

    Use retrieval-augmented generation when context engineering must operate at larger scale.

Level 4 · Durable systems

Make automated work repeatable, inspectable, and economical across sessions.

  1. Loop Engineering

    Use measurable completion gates and durable state to make automation reliable.

  2. Cost Optimisation

    Reduce spend through model choice, context, isolation, and structured tooling.