# Milestone: Passing the AWS Certified AI Practitioner (AIF-C01) Exam As a staff engineer working primarily on AWS, I've spent the last few years going deeper on the parts of the cloud I already knew: architecture, infrastructure as code, networking, security. This certification was the first in a while that asked me to start from close to zero. By early 2025 I was sitting in too many design discussions where someone would say "we could use an LLM for this," and I could follow the infrastructure conversation (where it runs, what it costs, how it's secured), but not the question that actually mattered: whether the approach was right at all. Passing the [AWS Certified AI Practitioner (AIF-C01)](https://www.credly.com/badges/0d13c714-b052-4343-9919-ae6fb36b1b88/) exam last week was how I closed that gap. ## Why AI Practitioner? * **The conversations had already started.** AI was arriving in architecture diagrams whether or not I felt ready to evaluate it, and "I'll pick it up as it comes" is not a strategy I've had much luck with. * **The foundational tier was the right fit.** The exam targets people who *use* AI/ML on AWS rather than build it. Coming from infrastructure, that framing was exactly right. I didn't need to train a model; I needed to know which service fits which problem. * **It's mostly generative AI, not classic ML.** Generative AI and foundation models make up over half the scored content. That surprised me, and it's the reason the certification was relevant to my actual work rather than an academic detour. * **Responsible AI is treated as a first-class skill.** Governance, bias, and transparency carry real weight here, not a token mention, which matched how these questions were already coming up in review. ## My Preparation Journey 1. **Started with** [**Stephane Maarek's AIF-C01 course**](https://www.udemy.com/course/aws-ai-practitioner-certified/) **on Udemy.** The standard recommendation, and it earns it: efficient coverage that doesn't pretend to more depth than a foundational exam needs. Watched at 1.5x, stopping wherever a concept was new rather than merely unfamiliar. 2. **Worked through** [**Master AWS Certified AI Practitioner AIF-C01**](https://www.educative.io/courses/aws-certified-ai-practitioner-aif-c01) **on Educative.** This is the one that made it stick. Maarek gave me the map; Educative made me walk it. 3. **Used the cloud labs to get hands on foundation models.** Actually prompting a model and moving inference parameters is a different kind of knowing from watching someone else do it. Reading about temperature and top-p taught me far less than watching output shift as I changed them. 4. **Treated the built-in quizzes and assessments as my practice tests.** They were enough to expose what I'd nodded along to without following, which is why I never went looking for a separate question bank. 5. **Kept it to two resources and a daily hour.** About three weeks, roughly an hour a day before work. Past exams have taught me that stacking five resources mostly buys the feeling of preparing. The deliberate overlap between two courses on the same syllabus was the real technique. The parts I had to slow down for on the second pass were exactly the parts I'd have got wrong. ## Key Takeaways * **The vocabulary was the point, not the certificate.** Being able to treat fine-tuning, RAG, and prompt engineering as cost and maintenance decisions rather than buzzwords has transferred into real work more directly than I expected from a foundational exam. * **RAG before fine-tuning, most of the time.** The right answer is usually the cheaper, more maintainable one: retrieval over retraining, prompt engineering over both when it will do. That maps onto an instinct I already trusted from infrastructure work: don't take on operational burden you haven't justified. * **Security and governance don't get a new rulebook.** Coming through the Security Specialty eight months earlier, the AI-specific material sat on foundations I already had: least privilege for model and data access, encryption in transit and at rest, data residency, audit trails. What's genuinely new is narrower than the discourse suggests: prompt injection as an input-validation problem, training-data provenance, model output as untrusted content, and whether your data trains someone else's model. The rest is the shared responsibility model wearing a new hat. ## Certification Impact * **Better questions in design reviews.** The practical change is being able to challenge an AI proposal on cost, maintenance, and data exposure rather than nodding through it. * **A map instead of fragments.** The real value of a foundational certification is structure, a coherent view of a domain I'd otherwise have assembled from conference talks and vendor blog posts. * **A bridge from my existing work.** Security, networking, and cost discipline all carry over. This didn't replace what I know; it gave me somewhere new to apply it. ## What's Next * **Build something real with Bedrock.** A certification proves recognition, not capability. The gap closes by shipping. * **Go deeper on the ML side.** The Machine Learning Engineer – Associate is the natural next step if this direction keeps proving useful. * **Bring responsible AI into how we review designs.** The governance material is the part most worth turning into team practice rather than personal knowledge. ## Advice for Aspiring Candidates Be honest with yourself about why you're taking it. After SAP-C02 and two Specialty exams, this was comfortably the least difficult exam in this series. The questions test recognition and judgement rather than deep architecture, and three weeks at an hour a day was enough. The cost of finding out is low. The exam lists at $100, and the 50% discount voucher AWS issues after passing any of their certifications brought mine down to $59 including tax. If you have passed an AWS exam before, check your certification account for that benefit before paying full price. So the value here isn't the challenge. It's that you come out with a structured view of a domain that is otherwise mostly noise. That only works if you resist the urge to finish fast. The concepts stayed abstract for me until I'd touched the services, and the labs were where preparation turned into understanding. And if you already hold a security or infrastructure certification, you are further along than you think. Much of this builds on ground you have already covered. Every certification in this series so far has deepened knowledge I already had. This one was about becoming a beginner again, which was less comfortable and probably more useful. AI is going to keep showing up in the systems we design, and I would rather meet it with a map than with opinions.