AWS folds generative AI and agent skills into SAP-C03 and DVA-C03 certifications

The changes at a glance

AWS is revising two of its most widely taken certifications, Solutions Architect Professional and Developer Associate, and both new exams will test generative AI and agent skills. According to a dev.to write-up by AWS instructor Maruchin Tech, which compiles details from AWS's official blog and certification pages, SAP-C02 will be retired in favour of SAP-C03 on November 17, 2026, and DVA-C02 will give way to DVA-C03 on December 1. Registration opens October 27, 2026 in all languages, and the full exam guides with detailed task statements arrive the same day, so the current picture rests only on announcement-level detail.

The exam formats are unchanged. SAP-C03 runs 180 minutes with 75 questions at $300. DVA-C03 runs 130 minutes with 65 questions, of which 50 are scored and 15 unscored, at $150 with a passing score of 720 out of 1000.

SAP-C03 gains a large AI section and post-quantum crypto

The professional-level exam is reorganized from four domains into five: design and implementation of cloud-native architectures; security, compliance and governance; cost-optimized design; resilience, migration and business continuity; and operational excellence and automation. Domain weightings have not been published yet.

The single largest addition is generative AI and agentic AI, with 11 new skills covering integration design with Amazon Bedrock, agent architectures built on Bedrock AgentCore, retrieval-augmented generation, content filtering via Bedrock Guardrails, and human-oversight workflows for AI operations. The AWS blog, as cited in the article, justifies the overhaul by saying today's architects integrate generative AI, agentic architectures and post-quantum cryptography rather than focusing only on highly available, cost-effective systems.

Other new areas include resilience engineering (Fault Injection Service, Resilience Hub, Application Recovery Controller and automated Systems Manager runbooks), seven cloud-native skills such as durable Step Functions workflows and service mesh with VPC Lattice and ECS Service Connect, six DevSecOps and observability skills including pipeline vulnerability scanning, and three data skills such as lakehouses built on Lake Formation and Apache Iceberg. Post-quantum cryptography also enters the exam through ML-DSA and ML-KEM support in AWS KMS. The target candidate needs two or more years designing and implementing AWS solutions.

DVA-C03 makes AI security a first-class topic

For the developer exam, AWS frames AI-assisted development as a core competency rather than an optional extra. A brand-new task called AI Security brings five skills: managing access to AI services, data privacy controls using VPC endpoints and guarantees that model inputs and outputs are not used for training, content filtering and prompt-injection defences, securing agent interactions through tool-use authorization, session isolation and human-in-the-loop approvals, and keeping sensitive information out of monitoring logs.

AI-assisted development also runs through every domain, from code generation and automated review in domain one, to automated testing and CI/CD deployment approvals in domain three, to AI-driven error analysis and performance suggestions in domain four. Container coverage grows, with ECR, ECS, EKS and Fargate now explicit, and Amazon Bedrock, Bedrock AgentCore, Amazon Q, Kiro, Amazon Data Firehose and AWS PrivateLink enter scope. To make room, older content is trimmed hard: DynamoDB drops from four skills to one, and the security, testing, observability and optimization areas are all consolidated. Domain weights stay at 30, 26, 22 and 22 percent.

Notably, prompt engineering, RAG design, model selection, SageMaker AI and AI governance frameworks are explicitly out of scope. That is a deliberate boundary with the AI-focused MLA-C02 exam: DVA-C03 tests using, integrating and securing AI, not building AI systems.

Watch the retirement dates

The dev.to article flags a discrepancy worth checking: the AWS blog lists the last day to sit SAP-C02 as November 16 while the certification page says November 17, and the same one-day gap exists for DVA-C02 (November 30 versus December 1). Candidates close to finishing a C02 exam should confirm the real cutoff on the registration screen.

Why it matters

The significance is that AI content has reached certifications that were never AI-specific. AWS had already announced an AI Business Strategist certification and an update to MLA-C02 as of September 1, 2026, but SAP and DVA are the mainstream credentials, so this revision touches far more candidates.

For anyone mid-way through preparing for a C02 exam, there is now a hard deadline: finish by mid-November for SAP-C02 or the end of November for DVA-C02, or restart against a different blueprint. For those targeting the C03 versions, the sensible move is to wait for the October 27 exam guides before booking, since domain weightings and task statements are still unpublished.

It also signals what AWS expects from these roles: architects should be able to design AI-integrated and agentic systems, and developers should work with AI tooling while handling prompt injection, data privacy and agent authorization, concerns that mirror what teams already face in production. For training providers, a significant share of existing C02 course material becomes outdated the moment the new exams go live.