AWS Certified AI Practitioner (AIF-C01) in 2026
Part 1 of 6 in the AIF-C01 exam prep series for developers. Next: Domain 1: Fundamentals of AI and ML.
If you write software for a living, the AWS Certified AI Practitioner exam looks like an easy weekend. It's foundational, there's no code, and you probably already call an LLM API from somewhere in your stack.
Two things make it trickier than it looks.
First, it's a judgment exam. Almost every question gives you several answers that would technically work and asks for the best one. The best one is decided by AWS's view of the world: managed over self-built, cheapest option that meets the requirement, responsible by default.
Second, the exam changed in 2026. Exam guide v1.1 added seven objectives, threaded agentic AI through four of the five domains, and named services most prep courses never mention: Amazon Bedrock AgentCore, Strands Agents, Kiro, and Amazon Quick. Around the same time, AWS moved several services those courses do teach into maintenance mode.
What this certification is (and isn't)#
AIF-C01 validates foundational knowledge of AI, machine learning, and generative AI concepts, and of the AWS services built around them. AWS describes the target candidate as someone with up to six months of exposure to AI/ML on AWS who uses AI/ML solutions but doesn't necessarily build them.
For developers, that description is the most useful sentence in the guide. The exam never asks you to write or read code. It asks you to recognize a scenario and pick the approach, service, or metric that fits it.
The trap is engineering instinct. Your reflex might be to build the thing yourself, fine-tune a model, or reach for the most powerful option. The exam usually rewards the opposite.
The exam at a glance#
| Item | Detail |
|---|---|
| Exam code | AIF-C01 |
| Level | Foundational |
| Questions | 65 total: 50 scored, 15 unscored |
| Duration | 90 minutes |
| Cost | 100 USD |
| Delivery | Pearson VUE test center or online proctored |
| Result | Pass/fail, scaled score from 100 to 1,000 |
| Passing score | 700 |
| Scoring model | Compensatory: no per-domain minimum |
Question types#
| Type | Format | Credit |
|---|---|---|
| Multiple choice | One correct answer out of four | Right or wrong |
| Multiple response | Two or more correct out of five or more | You must select all correct options |
| Ordering | Put 3–5 steps in the correct sequence | Only the complete correct order counts |
| Matching | Match 3–7 prompts to responses | Only a fully correct set of pairs counts |
Ordering and matching are all-or-nothing, so partial knowledge scores zero on them.
How scoring works#
- Scaled score. Your result is a scaled score, which lets AWS compare exam forms of slightly different difficulty.
- Unscored questions. The 15 unscored questions are ones AWS is trialing for future exams. They aren't marked, so treat every question as if it counts.
- Compensatory model. You need 700 overall. A weak domain can be offset by strong ones.
- No penalty for guessing. Unanswered questions are scored as incorrect, and wrong answers cost nothing extra, so never leave a question blank.
The five domains#
Figure: Domains 2 and 3 together are more than half the exam. Generative AI and foundation models are where the points are.
| Domain | Weight | Approx. scored questions | Covered in |
|---|---|---|---|
| 1. Fundamentals of AI and ML | 20% | ~10 | Post 2 |
| 2. Fundamentals of GenAI | 24% | ~12 | Post 3 |
| 3. Applications of Foundation Models | 28% | ~14 | Post 4 |
| Agentic AI (spans D1, D2, D3, D5) | Post 5 | ||
| 4. Guidelines for Responsible AI | 14% | ~7 | Post 6 |
| 5. Security, Compliance, and Governance | 14% | ~7 | Post 6 |
What changed in the current guide#
The guide's revision history lists v1.0 (March 26, 2026) and v1.1 (April 30, 2026). AWS publishes guide updates about a month before they reach the live exam, so v1.1 content has been on the exam since around the end of May 2026.
Most courses were recorded against the original 2024 guide, which is why this section matters.
Seven brand-new objectives#
| Objective | What it asks | Covered in |
|---|---|---|
| 1.2.6 | When to use traditional ML vs. a foundation model | Post 2 |
| 2.1.4 | Token-based pricing and its effect on cost and performance | Post 3 |
| 2.1.5 | The role of context engineering | Post 5 |
| 2.1.6 | Agentic AI concepts: multi-agent patterns, MCP, memory, tools, orchestration | Post 5 |
| 3.2.5 | Prompt versioning with Amazon Bedrock Prompt Management | Post 4 |
| 3.4.5 | Business alignment metrics: task completion rate, user satisfaction, cost per interaction | Post 4 |
| 5.1.5 | Hallucination detection and grounding techniques | Post 6 |
Rewritten objectives worth knowing#
- 1.1.1 and 1.1.2 add agentic AI to the terms you must define and distinguish.
- 1.1.3 expands inference types to include asynchronous and serverless.
- 1.1.5 is reworded to cover types of AI/ML learning generally, rather than naming only the three classic ones.
- 1.3.1 now asks you to differentiate pipeline components, not just describe them.
- 1.3.4 now names Amazon Bedrock, Amazon Q, Amazon Quick, Kiro, and SageMaker AI as pipeline services.
- 1.3.6 swaps AUC for precision and recall.
- 2.2.1 drops simplicity from the GenAI advantages and adds conversational capabilities and content generation.
- 2.2.3 adds cost, latency, and model complexity as model selection factors.
- 2.2.4 adds ROI to the business value metrics.
- 2.3.1 now lists Bedrock, SageMaker AI, SageMaker JumpStart, Amazon Quick, Kiro, Strands Agents, and Bedrock AgentCore.
- 3.1.5 adds model distillation to the customization cost trade-offs.
- 3.1.6 no longer names Amazon Bedrock Agents. It now asks you to define the role of AI agents and their business applications.
- 3.4.1 and 3.4.2 add human-in-the-loop evaluation and LLM-as-a-judge.
- 4.2.2 and 4.2.4 add explainability tools and name user-feedback mechanisms and AI decision transparency.
- 5.1.1 adds AgentCore Identity, Policy in AgentCore, and Bedrock Guardrails to the security services.
- 5.1.4 adds data leakage prevention, output filtering and validation, audit trails for AI interactions, and toxicity.
In-scope services: what's new and what's gone#
Added in v1.1: Amazon Aurora, Amazon Bedrock AgentCore, Kiro, Strands Agents, Amazon SageMaker JumpStart, AWS Transform, and Amazon Quick.
Removed in v1.1: Amazon MemoryDB.
Taught by many courses but not on the current in-scope list: Amazon Kendra, Amazon Augmented AI (A2I), Amazon Fraud Detector, Amazon Mechanical Turk, Amazon Comprehend Medical, Amazon Transcribe Medical, AWS HealthScribe, PartyRock, AWS Audit Manager, and Amazon Q.
The list is officially non-exhaustive, so they can still appear as distractors. Know each in one line and move on.
Services on the list that most courses skip#
You won't get deep questions on these, but one line each is enough to eliminate a distractor.
| Service | One line |
|---|---|
| AWS KMS | Creates and manages encryption keys |
| AWS Secrets Manager | Stores and rotates credentials such as API keys |
| AWS Glue | Serverless ETL plus a Data Catalog of your datasets |
| AWS Glue DataBrew | Visual, no-code data cleaning and preparation |
| AWS Lake Formation | Builds data lakes with fine-grained access control |
| Amazon EMR | Managed big-data frameworks such as Spark |
| Amazon Redshift | Data warehouse for analytics queries |
| AWS Data Exchange | Marketplace for third-party datasets |
| Amazon Aurora | Managed relational database; the PostgreSQL edition can store vectors |
| Amazon RDS | Managed relational databases; RDS for PostgreSQL can store vectors |
| Amazon DynamoDB | Serverless NoSQL key-value database |
| Amazon DocumentDB | MongoDB-compatible document database |
| Amazon ElastiCache | Managed in-memory cache |
| Amazon Neptune | Graph database |
| Amazon ECS / Amazon EKS | Container orchestration (AWS-native / Kubernetes) |
| Amazon CloudFront | Content delivery network |
| AWS Budgets / AWS Cost Explorer | Spend alerts / spend analysis and forecasting |
| AWS Well-Architected Tool | Reviews workloads against AWS best-practice pillars |
The service maintenance shake-up#
On June 30, 2026, AWS moved a batch of AI services and features into maintenance mode:
- Existing customers keep using them and still get bug fixes and security updates.
- No new features are built.
- New customers are blocked, starting July 30, 2026.
Several of these are central to older prep material.
| Service or feature | What happened | What to answer now |
|---|---|---|
| Amazon Bedrock Agents | Renamed Bedrock Agents Classic; closed to new customers from July 30, 2026 | Bedrock AgentCore (see Post 5) |
| Amazon Kendra | Maintenance mode; closed to new customers from July 30, 2026 | Amazon Bedrock Knowledge Bases, including the new Managed Knowledge Base |
| Amazon Q Business | Closed to new customers from July 31, 2026 | Amazon Quick |
| Amazon Q Developer IDE plugins | New signups blocked from May 15, 2026; end of support April 30, 2027 | Kiro |
| SageMaker AI features (Clarify, Model Monitor, Ground Truth, A2I, Debugger, and others) | Maintenance mode | The concepts stay testable; see below |
Bedrock itself is unaffected: models, Knowledge Bases, and Guardrails continue as normal.
How to use this series#
| Post | Focus |
|---|---|
| 1. This post | The exam, v1.1 changes, study plan |
| 2. Domain 1 | AI/ML terms, learning types, inference types, managed AI services, ML lifecycle, metrics |
| 3. Domain 2 | Tokens and embeddings, foundation model lifecycle, token pricing, GenAI trade-offs, AWS GenAI services |
| 4. Domain 3 | Model selection, RAG, customization, prompt engineering, fine-tuning, evaluation |
| 5. Agentic AI | Agents, MCP, memory, context engineering, multi-agent patterns, Strands, AgentCore, Kiro, Quick |
| 6. Domains 4 + 5 | Responsible AI, Guardrails, security services, hallucination grounding, governance, final cram sheet |
Each domain post follows the same layout:
- Domain at a glance: weight, task statements, and what's new in v1.1.
- Objective checklist: every objective in the domain and where the post covers it. Agentic objectives get a one-line summary and a link to Post 5, so each domain post is still a complete checklist.
- Core concepts: organized by task statement.
- Service cheat sheet: one line per service.
- Commonly confused: the answer pairs that decide points.
- Practice questions: including ordering and matching, with answers hidden until you click.
- Key takeaways to revise from.
Four callout types recur:
- New in v1.1: content added or changed in the current guide.
- Exam tip: how a concept tends to show up in questions.
- Old name → new name: where older material uses a name the exam no longer does.
- Covered in depth in another post: a pointer to where a topic lives.
A study plan#
The domains aren't equally hard for developers. Domain 1 is mostly vocabulary, and Domain 4 is the smallest change since the original guide. Most of the effort belongs in Domains 2 and 3 and the agentic material.
| Week | Work |
|---|---|
| 1 | Posts 1–3. Drill the service-to-use-case table in Post 2 until it's reflexive. |
| 2 | Posts 4 and 5. These cover the most points and most of the new material. |
| 3 | Post 6, then timed practice exams. For every wrong answer, write down why the wrong option looked right. |
For practice questions, start with AWS's free official materials on AWS Skill Builder: the exam prep plan and the official practice question set. If you use third-party practice exams, check that they say they're aligned to guide v1.1.
Pitfalls specific to developers#
- Building instead of buying. If a managed service does the job, the managed service is the answer.
- Treating a bigger model or fine-tuning as the fix. Stale facts need RAG. Hallucinations need grounding. Cost problems need caching, batching, or distillation.
- Ignoring who's asking. When the question's stakeholder is a CFO, the answer is a business metric, even if a model metric is technically correct.
- Reading the latest AWS news into questions. Know both the old and new names, and use the one the guide uses.
Exam-day logistics#
- Extra time for non-native English speakers. In your AWS Certification account, request the "ESL +30" exam accommodation before scheduling. It adds 30 minutes and doesn't expire once approved. If you've already scheduled, reschedule after approval so the extra time applies.
- 50% off your next exam. After you pass any AWS certification, a 50% discount voucher appears under Benefits in your certification account.
- Pace yourself. 90 minutes for 65 questions is about 80 seconds each. Answer everything, flag anything uncertain, and come back to it at the end.
- Choose your format. Test center or online proctoring. Online requires a clean desk, a webcam, and a stable connection.
Where this certification leads#
AIF-C01 is the entry point to AWS's AI certifications. Developers usually go next to the AWS Certified Machine Learning Engineer – Associate (MLA-C01) or the AWS Certified Generative AI Developer – Professional (AIP-C01), both of which are hands-on and technical. Much of this series, particularly Posts 4 and 5, carries straight over.
Next: Domain 1: Fundamentals of AI and ML.
Sources#
- AWS Certified AI Practitioner (AIF-C01) exam guide and PDF version
- Exam guide revisions (v1.0 → v1.1)
- Out-of-scope AWS services
- AWS Certified AI Practitioner certification page
- AWS service availability updates, June 30, 2026
- Amazon Bedrock Agents Classic maintenance mode
- Amazon Bedrock Managed Knowledge Base general availability
- AWS DevOps Blog: Amazon Q Developer end-of-support announcement (April 30, 2026)
Originally published at https://iuriio.com/blog/posts/2026/09/aif-c01-part-1-overview
