AIF-C01 study plan: AWS AI Practitioner without the maths

What the exam is really testing

AIF-C01 is a literacy exam. Questions describe a business situation — a company wants to summarise support tickets, ground a chatbot in its own documentation, or reduce the cost of an inference workload — and ask which approach or which AWS service fits. You are never asked to train a model, write code, or reason about an equation.

What it does demand is precision with terms that get used loosely everywhere else. Fine-tuning and retrieval-augmented generation are not interchangeable, a token is not a word, temperature and top-p do different things, and an embedding is not a vector database. The distractors are built from exactly these confusions.

Phase 1 — Build the vocabulary (about a week)

Start with the AI and machine-learning fundamentals domain even though it is not the heaviest, because every later domain assumes it. Supervised versus unsupervised learning, training versus inference, what a model actually is, where AI sits relative to machine learning and deep learning: these are the terms the rest of the exam builds sentences out of.

Answer a short practice session cold before you begin, purely for calibration. If you already work near AI you will clear this phase quickly; if the field is new, do not rush it, because a shaky foundation here shows up as guesswork in every later domain.

Phase 2 — Foundation models and their applications (the bulk of the plan)

The generative-AI fundamentals and foundation-model applications domains carry the most weight between them, and this is where the majority of your time belongs. Work through what a foundation model is, how prompting shapes its output, what fine-tuning changes and what it costs, and how retrieval-augmented generation grounds a model in your own data without retraining anything.

Then map the concepts onto the AWS services that implement them, because that mapping is what the questions are made of. The current exam guide also covers the agentic material — agents, tool use, the emerging interoperability standards and the managed services around them — along with token-based pricing and context engineering. This is a fast-moving area, so study against the current guide version rather than a study set written when the exam launched.

• Generative-AI fundamentals: foundation models, tokens, embeddings, prompting, context windows, and the cost model.
• Applications of foundation models: prompt engineering, retrieval-augmented generation, fine-tuning and when each is the right answer.
• Agents and tool use: what an agent adds over a bare model, and the managed services for building one.
• Evaluation: how you tell whether a generative system is working, and what the metrics do and do not capture.

Phase 3 — Responsible AI, security and governance (about a week)

The two remaining domains are the smallest but they are cheap marks, and candidates from a technical background routinely neglect them. Responsible AI covers bias, fairness, transparency, explainability and the guardrail mechanisms that constrain model output. Security and governance covers data protection for AI workloads, the compliance framing, and who is accountable for what.

Treat these as vocabulary domains too. The questions ask you to identify which principle a scenario violates or which control addresses a stated risk, so being able to name and distinguish the concepts is the whole task. Rebuild practice sessions from your incorrect and flagged questions as you go.

Phase 4 — Mock week: rehearse, then book

Sit full-length timed mocks under real conditions. One honest caveat about format: the real exam includes ordering and matching question types alongside standard multiple choice and multiple response, and multiple-choice practice approximates those rather than reproducing them. Expect them, and read the instruction for each item rather than assuming its shape.

Between mocks, spend a session on the incorrect answers. Book once you are clearing the pass mark comfortably on consecutive mocks with different draws.

Signals you are ready

Book the exam when all of these are true:

• Consecutive full-length mocks above the pass mark, not one result.
• You can explain the difference between prompting, retrieval-augmented generation and fine-tuning — and say when each is the cheaper answer.
• The responsible-AI and governance domains are not your weakest; they are small and easy to close.
• You can name the AWS service behind each capability in the guide without reaching for a search.

Questions, answered