A useful guide gets you to a working result, names the failure modes, and gives you a test for completion. These guides are written for people who want to use AI deliberately rather than collect vocabulary.

Start with interactive use if you are new. Move to API work only when software needs repeatable access. Add retrieval, tools, or agents one capability at a time, after the simpler version has a measurable baseline.

Explore this section

Beginner guideHow to start using AI and get a result you can trust

A complete beginner's guide to using ChatGPT, Claude, Gemini, or Grok for a real task, including prompt structure, source checking, privacy, and a practical first exercise.

Open page →Developer access guideHow to get AI API access without leaking a key or losing cost control

Create OpenAI, Anthropic, Google Gemini, or xAI API access safely, with server-side secrets, project ownership, budgets, usage alerts, and a first-call checklist.

Open page →Model evaluation guideHow to choose an AI model using your own work instead of benchmarks

Build a practical AI model evaluation, compare efficient, balanced, and flagship models, score quality and cost, and create a routing policy for production.

Open page →Retrieval guideHow to build document Q&A that cites sources and admits uncertainty

A practical retrieval-augmented generation guide covering document preparation, chunking, search, reranking, citations, permission boundaries, and evaluation.

Open page →Agent reliability guideHow to evaluate an AI agent before giving it real authority

Evaluate AI agents across task completion, tool use, permissions, evidence, retries, cost, latency, and safe stopping before production deployment.

Open page →Local inference guideHow to run a useful language model locally with Ollama or LM Studio

Install a local LLM with Ollama or LM Studio, choose a model that fits memory, test quality and speed, protect the local endpoint, and build a first private workflow.

Open page →

A sensible learning order

First, learn to describe an outcome, supply evidence, set constraints, and review a result. Second, compare models on the same task. Third, use an API with budgets and logs. Fourth, connect approved knowledge through retrieval. Only then give a model tools or authority to take actions.

This order is not less ambitious. It separates failure modes. When a tool-using agent fails, you need to know whether the problem came from retrieval, instructions, permissions, tool behavior, or the model's decision.

  • Interactive use: learn instruction and review habits.
  • API use: learn secrets, schemas, budgets, and observability.
  • Retrieval: learn evidence quality and permission boundaries.
  • Agents: learn state, approvals, retries, and safe stopping.

What every guide should leave you with

Keep a small test set, the exact prompt or policy, the chosen model ID, the expected output, and a record of important failures. That package is more valuable than a screenshot of one impressive answer because it can be repeated after a model or workflow changes.

Do not automate a result you have not learned to review.

Primary sources

Verify before you commit money or architecture