AWS now has AI credentials at very different depths. AIF-C01 validates foundational AI and generative-AI knowledge with an emphasis on practical business applications. MLA-C01 defined the associate machine-learning engineering scope, while AIP-C01 validates advanced design, implementation, and deployment of generative-AI solutions on AWS.
There is an important current-status wrinkle. AWS ended English-language MLA-C01 testing on September 28, 2026 and opened the MLA-C02 beta in English on September 29. MLA-C01 continues during the beta period in Korean, Japanese, and Simplified Chinese. That means the MLA-C01 scope remains useful for understanding the associate ML-engineering role, but English-language candidates should check the current MLA-C02 beta path rather than assume MLA-C01 is still the live English exam.
The three paths should not be treated as a mandatory ladder. They address different audiences. A business or technology professional may need AI fluency without becoming an ML engineer. An ML engineer can specialize in data preparation, model development, deployment, and monitoring without focusing primarily on generative-AI application patterns. A generative-AI developer needs deep application, retrieval, agentic, safety, evaluation, and operational knowledge even when traditional model training is not the center of the job.
The wider AWS certifications portfolio is most useful when the credential is matched to the work. Choose by the artifact you are responsible for: an AI-informed business decision, a production ML workflow, or a generative-AI application and its lifecycle.
AIF-C01 builds the language needed to discuss AI responsibly
AWS describes AIF-C01 as a foundational exam for people who want to demonstrate understanding of AI concepts and AWS AI tools, with practical business applications in view. Its domains cover AI and ML fundamentals, generative AI, foundation-model applications, responsible AI, and security, compliance, and governance.
That is useful for professionals who participate in AI projects without building the technical system. Product owners, managers, analysts, sales engineers, project leads, risk teams, and cloud practitioners need to distinguish training from inference, understand what foundation models can and cannot do, recognize common generative-AI patterns, and ask sensible questions about data, governance, cost, security, and responsible use.
MLA-C01 owns the machine-learning engineering lifecycle
MLA-C01 remains important for comparison because it defined the associate-level machine-learning engineering scope immediately before the English-language MLA-C02 beta transition. Its guide is organized around data preparation, ML model development, deployment and orchestration of ML workflows, and monitoring, maintenance, and security. For English-language candidates in October 2026, AWS has already moved new testing into the MLA-C02 beta, so preparation should be checked against that current beta outline before scheduling. That makes it a production lifecycle exam rather than a general AI-awareness credential.
The ML engineer has to make data usable, select and develop appropriate modeling approaches, automate training or deployment processes, and keep a model healthy after release. Drift, data quality, failed pipelines, scaling, observability, access control, and reproducibility become operational concerns. A model that performs well in a notebook is only the beginning of the engineering problem.
Services such as Amazon SageMaker are relevant because they connect experimentation to managed training, deployment, and lifecycle capabilities. The important skill is not memorizing a console; it is understanding which stages need control and how evidence from production feeds the next iteration.
AIP-C01 is centered on production generative-AI application engineering
AIP-C01 is a professional-level credential for advanced generative-AI work. AWS describes the target around designing, implementing, and deploying AI solutions, with exam content covering foundation-model integration, data and compliance, implementation and integration, safety and governance, optimization, and testing and troubleshooting. Retrieval-augmented generation, vector search, prompt engineering, agents, evaluation, infrastructure as code, and CI/CD all belong in the professional conversation.
That scope is different from simply knowing what generative AI is. A professional developer has to decide how context reaches the model, how tools are invoked, how sensitive information is protected, how output is evaluated, how latency and cost are controlled, and what happens when the system fails or produces an unsafe result. The AI behavior becomes a software-system responsibility.
Data depth separates foundational fluency from engineering work
AIF-C01 candidates should understand that AI systems depend on data and that data quality, privacy, bias, and governance affect outcomes. MLA-C01 goes much deeper because the engineer prepares datasets and has to reason about transformations, features, labels, training and validation behavior, reproducibility, and data changes over time.
AIP-C01 has another data problem: grounding generative systems with enterprise context. Retrieval pipelines, document preparation, embeddings, vector stores, permissions, freshness, and citation or traceability strategies can determine whether an application produces useful answers. The work is not identical to traditional ML feature engineering, but it still demands disciplined data architecture.
Model development is central to MLA-C01 but not the main story in AIF-C01
AIF-C01 is designed to recognize AI and ML concepts rather than to validate hands-on model engineering. MLA-C01 expects the candidate to work through model development decisions and the surrounding engineering lifecycle. That can include selecting approaches, preparing training data, evaluating performance, tuning, and deciding how the model should be deployed and monitored.
AIP-C01 often assumes the developer is integrating foundation models rather than training a new large model from scratch. The difficult engineering moves toward model selection, prompting, grounding, tool use, agent orchestration, evaluation, safety, and application integration. Understanding that distinction prevents candidates from studying deep traditional ML theory for the wrong role or, in the opposite direction, assuming generative-AI integration removes the need for engineering rigor.
RAG and agents are a professional application concern in AIP-C01
Retrieval-augmented generation and agentic workflows are powerful because they let a generative system use current enterprise information and perform controlled actions. They also create new failure modes. Retrieval can return irrelevant or unauthorized context. An agent can call the wrong tool, repeat an action, or follow an ambiguous instruction. Evaluation has to test behavior across realistic cases rather than one impressive demo.
Hands-on AWS generative AI labs can help candidates see those problems directly. The valuable practice is not just getting an answer from a model; it is measuring whether the answer is grounded, whether access is correct, whether the workflow fails safely, and whether the system can be observed and improved after deployment.
Operations matter in both MLA-C01 and AIP-C01, but the signals differ
ML systems need monitoring for data quality, model performance, drift, pipeline failures, resource use, and security. Generative-AI applications add behavioral measures such as groundedness, answer quality, tool success, safety outcomes, token usage, latency, and cost. Both require controlled deployment and rollback, but the evidence used to judge a release can be different.
This is why professional AI work converges with software and cloud operations. Automated testing, infrastructure as code, access controls, observability, and release discipline keep experimentation from turning into unpredictable production behavior. The AI-specific metrics add to engineering practice rather than replace it.
Evaluation becomes a different discipline in each path
AIF-C01 candidates should understand evaluation conceptually: AI output can be wrong, biased, unsafe, or inappropriate for a business use case, so important decisions need suitable oversight. MLA-C01 engineers need quantitative model evaluation tied to data, task metrics, validation strategy, and production monitoring. AIP-C01 developers need behavioral evaluation for generative systems, including groundedness, relevance, safety, tool use, and consistency across representative prompts.
Those differences are useful when choosing a path. If your work stops at asking whether an AI capability is appropriate, foundational knowledge may be enough. If you decide whether a model is ready for deployment, you need ML engineering depth. If you decide whether an agent or RAG application behaves reliably under complex user interactions, professional generative-AI evaluation becomes central.
Security moves from awareness to architecture and enforcement
All three exams include security or governance, but the expected action changes. AIF-C01 candidates should understand why data privacy, responsible AI, compliance, and access control matter. MLA-C01 engineers must protect training and inference workflows, data stores, endpoints, credentials, pipelines, and monitoring. AIP-C01 developers additionally have to secure retrieval context, model access, tool permissions, agent actions, and sensitive prompts or outputs.
The most important professional principle is least privilege for the complete AI workflow. An agent should not receive broad access merely because it needs one business action, and a retrieval system should not bypass the permissions that apply to source information. AI can amplify the impact of weak authorization, so security architecture has to be designed before the system reaches production.
Cost is another dimension that changes by role. AIF-C01 candidates should understand that AI services have usage and business-value implications. MLA-C01 engineers need to consider training resources, inference patterns, endpoint utilization, storage, and the cost of repeated experimentation. AIP-C01 developers have to watch token use, retrieval infrastructure, agent tool calls, latency, model choice, and the possibility that a seemingly small interaction becomes expensive at production volume.
Cost optimization should not be separated from quality. A cheaper model that produces unreliable results may create more human review and lower business value. A higher-quality model can still be wasteful if prompts carry unnecessary context or an agent loops through tools. Mature AI engineering measures quality, latency, and cost together so that optimization does not silently degrade the user outcome.
Hands-on practice should mirror those role boundaries. For AIF-C01, analyze use cases and risks. For MLA-C01, build a reproducible training-to-deployment workflow and monitor it. For AIP-C01, build a grounded generative application, test difficult inputs, constrain tool access, and observe quality, latency, and cost. The artifact you practice on should look like the artifact your target role owns.
That makes practice role-specific rather than generic.
Choose the path by what you are accountable for shipping
Choose AIF-C01 when you need credible AI and generative-AI fluency for business, product, risk, sales, cloud, or project decisions without owning the implementation lifecycle. Choose MLA-C01 when you prepare data, develop models, deploy ML workflows, and monitor production machine-learning systems. Choose AIP-C01 when you design and ship sophisticated generative-AI applications, retrieval systems, agents, integrations, and the controls needed to operate them at professional scale.
There is useful overlap, but no requirement to earn all three. An ML engineer may add AIP-C01 when generative-AI application engineering becomes a major responsibility. A generative-AI developer may study MLA concepts to strengthen evaluation or ML foundations. A business professional may stop at AIF-C01 because deeper engineering does not match the role. The best path is the one that increases competence around the system you actually own.