How to Pass AIF-C01: The AI Practitioner Guide

AIF-C01 (AI Practitioner) is AWS's newest foundational certification — the AI-era equivalent of Cloud Practitioner. 65 questions, 90 minutes, $100. It tests AI/ML vocabulary and Bedrock service knowledge, not math or model training. No hands-on ML experience needed — this is accessible to managers, developers, and anyone who needs to talk credibly about AI on AWS. Two weeks at 1–1.5 hours/day is realistic for most candidates.

What the exam tests

• AI/ML fundamentals (20%): AI vs ML vs deep learning vs GenAI, learning types, common terminology
• GenAI fundamentals (24%): foundation models, tokens/context windows, prompt engineering, RAG vs fine-tuning
• FM applications (28% — biggest): Amazon Bedrock end-to-end, agents, guardrails, inference modes, evaluation
• Responsible AI (14%): bias, transparency, explainability, human oversight, AI Service Cards
• Security & governance (14%): data privacy in Bedrock, IAM for AI services, compliance

Week 1 — AI/ML and GenAI concepts (days 1–7)

• Days 1–2: Core vocabulary — AI/ML/deep learning/GenAI relationships, supervised/unsupervised/reinforcement, training vs inference
• Days 3–4: GenAI mechanics — foundation models, tokens, context windows, embeddings, temperature/top-p
• Days 5–6: Adaptation ladder — prompt engineering → RAG → fine-tuning, and when each is right
• Day 7: first practice set — vocabulary calibration

Week 2 — Bedrock depth + governance (days 8–14)

• Days 8–9: Bedrock end-to-end — model access, invocation modes (on-demand vs provisioned vs batch), playgrounds
• Days 10–11: Bedrock capabilities — Knowledge Bases (managed RAG), Agents (multi-step actions), Guardrails (content filtering)
• Day 12: Responsible AI — bias sources, transparency (AI Service Cards), human oversight patterns
• Day 13: security & governance — the Bedrock data-privacy facts (prompts don't train base models), IAM controls
• Day 14: timed mock + mistake review — the exam is vocabulary-heavy, so wrong answers are usually memorization gaps

The concepts that trip people up

• Knowledge Bases = managed RAG (retrieval); Agents = orchestrated actions — different capabilities
• Guardrails filter BOTH inputs and outputs, across models — not a model feature
• Temperature controls randomness; top-p controls vocabulary pool; max tokens controls length — know which knob does what
• On-demand = pay per token; Provisioned Throughput = fixed cost for steady load; Batch = offline async jobs
• Bedrock does NOT use your prompts to train foundation models — the most-tested data-privacy fact
• Few-shot prompting = examples in the prompt; zero-shot = instructions only

What 'ready' looks like

You're ready when you can answer 'which Bedrock feature' and 'which prompt lever' questions at 75%+ on timed sets. AIF-C01 is genuinely attainable in two weeks — its value is giving you structured GenAI vocabulary plus a certification that barely existed a year ago, making it a differentiation play while it's still rare.