If you are preparing for the AWS Certified Developer – Associate right now, you are preparing for an exam that is about to change under your feet. AWS has published the DVA-C03 exam guide, registration for the new version opened on 27 October 2026, and the current DVA-C02 can be taken for the last time on 30 November 2026. From 1 December 2026 every Developer – Associate exam delivered is DVA-C03.
Most write-ups of the change stop at “AI was added”. That is true but not useful. If you are mid-preparation, the questions you actually have are: which domains gained and lost questions, which of the things I already studied moved or disappeared, what exactly is the new AI content and how deep does it go, which services were added to and removed from scope, and should I rush to sit DVA-C02 before the deadline or re-plan for DVA-C03.
This post answers those questions by putting the two official exam guides side by side, task statement by task statement and service list by service list. Everything below comes from the published DVA-C02 and DVA-C03 guides, so you can verify any line against the source. Our existing DVA-C02 exam guide breakdown and exam topics walkthrough remain accurate for the outgoing version. This post is the delta.
The transition at a glance
| Item | DVA-C02 | DVA-C03 |
|---|---|---|
| Last day to take | 30 November 2026 | Live from 1 December 2026 |
| Registration for new version | n/a | Opened 27 October 2026 |
| Scored questions | 50 | 50 |
| Unscored questions | 15 | 15 |
| Total questions | 65 | 65 |
| Duration | 130 minutes | 130 minutes |
| Passing score | 720 of 100–1,000 | 720 of 100–1,000 |
| Question types | Multiple choice, multiple response | Multiple choice, multiple response |
| Scoring model | Compensatory (overall only) | Compensatory (overall only) |
| Domains | 4 | 4 |
| Skills listed in guide | 101 | 72 |
| Credential name | AWS Certified Developer – Associate | AWS Certified Developer – Associate |
The mechanics did not change. Same number of questions, same time, same passing score, same two question formats, same compensatory scoring (you do not need to pass each domain, only the whole exam). What changed is the content, and the way the content is described.
One structural point before the detail: the DVA-C03 guide lists 72 skills across its four domains against 101 in DVA-C02 version 2.1. AWS is explicit that this is a consolidation, not a reduction of emphasis. The comparison page states that the consolidations “do not result in a decreased emphasis on essential skills” and that the exam was “reorganized to reduce repetition and ensure consistency in level of detail across all domains”. Treat a DVA-C03 skill as a heading that still covers everything the two or three DVA-C02 skills beneath it covered.
Domain weights: Troubleshooting gains, Development and Deployment give
| Domain | DVA-C02 | DVA-C03 | Change |
|---|---|---|---|
| Domain 1: Development with AWS Services | 32% | 30% | −2 |
| Domain 2: Security | 26% | 26% | 0 |
| Domain 3: Deployment → Testing and Deployment | 24% | 22% | −2 |
| Domain 4: Troubleshooting and Optimization | 18% | 22% | +4 |
On a 50-scored-question exam, two percentage points is one question. Domain 4 grows from roughly 9 to roughly 11 scored questions; Domains 1 and 3 each lose about one. That is not dramatic, but it reverses the DVA-C02 pattern where Troubleshooting and Optimization was the smallest domain and could be treated as the one to skim. On DVA-C03 it is tied with Deployment and is where several of the new AI-tool skills live.
Domain 3 is also renamed from “Deployment” to “Testing and Deployment”, which reflects a real move: the unit-testing skill that DVA-C02 kept in Domain 1 (Task 1.1) now lives in Domain 3, and the two DVA-C02 testing tasks were merged into one.
The headline addition: Task 2.3, securing AI service usage
The single biggest change is a brand-new task in the Security domain. DVA-C02 had three security tasks: authentication and authorization, encryption, and sensitive data. DVA-C03 merges encryption and sensitive data into one task and uses the freed slot for this:
Task 2.3: Identify and mitigate security risks associated with using and integrating AI services into application development.
Its five skills, verbatim from the guide:
- Skill 2.3.1: Request and manage access to AI services.
- Skill 2.3.2: Use data privacy controls to transmit data to AI services (for example, by using VPC endpoints to establish private connectivity, ensuring that inputs and outputs are not used to train AI models).
- Skill 2.3.3: Filter and control AI model inputs and outputs (for example, by applying content filtering, sensitive information detection, PII/PHI redaction, denied topic policies, prompt injection and manipulation protection).
- Skill 2.3.4: Secure AI agent interactions (for example, by using tool-use authorization, session isolation, human-in-the-loop approval flows for sensitive actions).
- Skill 2.3.5: Protect sensitive content in monitoring logs during interactions with AI services.
Read these as a developer, not as an ML engineer, because the guide fences off the ML-engineering side explicitly (see the out-of-scope job tasks below). What each skill is actually asking you to know, mapped to the in-scope services:
| Skill | What it means in practice | Services that answer it |
|---|---|---|
| 2.3.1 Request and manage access | Foundation model access must be enabled per account and Region before invocation; invocation is authorised by IAM actions on model and inference-profile resources; least-privilege policies for the application role | Amazon Bedrock, IAM |
| 2.3.2 Data privacy controls | Keep traffic to the model API off the public internet with interface VPC endpoints; know the vendor statement that customer prompts and completions are not used to train the base models and are not shared with model providers | AWS PrivateLink, Amazon Bedrock, Amazon VPC |
| 2.3.3 Filter inputs and outputs | Guardrail-style policies applied around a model call: content filters by category and strength, denied topics, sensitive-information filters that block or mask PII, word filters, and prompt-attack detection for injection and jailbreak attempts | Amazon Bedrock |
| 2.3.4 Secure agent interactions | An agent that can call tools is an identity that can take actions; scope which tools it may invoke, isolate one user’s session from another’s, and require a human approval step before irreversible or sensitive actions | Amazon Bedrock AgentCore, IAM |
| 2.3.5 Protect sensitive content in logs | Model invocation logging captures prompts and completions; decide whether to log them at all, where (CloudWatch Logs or S3), and apply log data-protection policies that detect and mask sensitive data | Amazon Bedrock, Amazon CloudWatch Logs |
The exam-style shape of these questions is familiar even though the subject is new. A scenario describes a chatbot that occasionally leaks another customer’s data, and the answer is session isolation plus a sensitive-information filter, not “fine-tune the model”. A scenario describes compliance requiring that prompts never traverse the public internet, and the answer is an interface endpoint, not an S3 bucket policy. The distractors will be the ML-engineering options the guide has placed out of scope.
The other nine AI skills, task by task
Task 2.3 is the only new task, but AI appears in five other tasks as added skills. DVA-C02 version 2.1 already had two Amazon Q Developer skills (1.1.11 “Use Amazon Q Developer to assist with development” and 3.3.6 “Use Amazon Q Developer to generate automated tests”) and, since late 2025, listed six “emerging topics” that could appear only as unscored pretest questions. DVA-C03 promotes those emerging topics into scored skills and generalises the wording from one product to “AI-assisted development tools” and “AWS AI tools”.
| Location | New DVA-C03 skill (verbatim) |
|---|---|
| Task 1.1 | Skill 1.1.9: Use AI-assisted development tools to generate, review, and optimize code (for example, spec-driven code generation, automated code reviews, intelligent code completion, refactoring suggestions, security scanning). |
| Task 1.1 | Skill 1.1.10: Integrate managed AWS AI services into applications. |
| Task 2.3 | Skills 2.3.1 to 2.3.5 (above). |
| Task 3.2 | Skill 3.2.5: Use AWS AI tools to generate tests and automate repetitive testing workflows (for example, automated test execution, test result analysis, regression testing automation, test coverage). |
| Task 3.3 | Skill 3.3.9: Use AWS AI tools to support CI/CD workflows (for example, automated deployment approvals, environment provisioning, post-deployment validation). |
| Task 4.1 | Skill 4.1.5: Use AWS AI tools to analyze errors and to generate troubleshooting suggestions. |
| Task 4.3 | Skill 4.3.6: Use AWS AI tools to identify optimization opportunities (for example, performance bottleneck detection, resource usage optimization, code efficiency improvement). |
That is ten AI-flavoured skills out of 72, spread across all four domains. Two observations help you calibrate the depth.
First, the phrase “spec-driven code generation” in Skill 1.1.9 is the vocabulary of Kiro, the AWS agentic IDE, which appears in the new in-scope list. Expect questions that describe a developer workflow (write a specification, let the tool generate code and tests, review suggested refactors, run a security scan on the result) and ask which capability or tool fits. You will not be asked to debug a generated function.
Second, Skill 1.1.10, “Integrate managed AWS AI services into applications”, is the developer-side counterpart of Task 2.3. In practice it means calling a model through the SDK: the runtime API that sends a prompt and receives a completion, the unified conversation API that works across models, streaming responses, and the fact that an inference call is just another signed AWS API request your application role must be allowed to make. Our SDK retries, pagination and presigned URLs guide covers the SDK mechanics that carry over unchanged to this new call.
Non-AI additions you might miss
Three additions have nothing to do with AI and are easy to overlook because they are listed under existing tasks.
Containers become explicit. DVA-C02 mentioned container images only as a packaging example. DVA-C03 adds two dedicated skills:
- Skill 3.1.2: Build and manage container images for deployment (for example, by using Amazon ECR, Docker files, container image tagging, lifecycle policies).
- Skill 3.3.2: Deploy containerized applications by using AWS container orchestration services (for example, Amazon ECS task definitions, Amazon EKS manifests, AWS Fargate launch types, rolling updates).
ECR, ECS and EKS were already in scope for DVA-C02, so this is a promotion of implicit content to named skills rather than new services. Expect at least one question that turns on ECR tag immutability or a lifecycle rule expiring untagged images, and one on an ECS task definition or rolling update. Our ECS, ECR and Fargate task definitions guide covers exactly this material.
Presigned URLs get their own skill. Skill 2.1.4, “Use presigned URLs to provide secure, time-limited access (for example, by using Amazon S3 and Amazon CloudFront)”, is new as a named skill. Presigned S3 URLs were always a favourite DVA-C02 question; the addition of CloudFront signed URLs and signed cookies next to them is the new angle. Know that an S3 presigned URL inherits the permissions and the session lifetime of the credentials that signed it, and that CloudFront signed URLs and cookies are the equivalent for content served through a distribution with a trusted key group.
Certificate services are named. Skill 2.2.2, “Manage certificates by using AWS services (for example, AWS Certificate Manager, AWS Private CA)”, upgrades DVA-C02’s “Describe certificate management” and is matched by both ACM and Private CA appearing in the in-scope list for the first time.
What moved: the recategorizations
AWS lists five relocations. The content is the same; it is tested under a different task. This matters if you organise your notes by domain.
| What moved | From (DVA-C02) | To (DVA-C03) |
|---|---|---|
| Unit testing by using AWS SAM | Domain 1, Task 1.1 | Domain 3, Task 3.2 |
| AWS Lambda function performance tuning | Domain 1, Task 1.2 | Domain 4, Task 4.3 |
| Serialize and deserialize data | Domain 1, Task 1.3 | Domain 1, Task 1.1 |
| Amazon API Gateway stages and custom domains | Domain 3, Task 3.4 | Domain 3, Task 3.1 |
| IaC templates (CDK, AWS SAM, AWS CloudFormation) | Domain 3, Task 3.3 | Domain 3, Task 3.3 |
The second row is the one with a scoring consequence. Lambda tuning (memory versus CPU allocation, cold starts, provisioned concurrency, SnapStart) moves from the domain that shrank to the domain that grew. If you studied Lambda performance as a Domain 1 topic, it is now contributing to the 22% Domain 4 instead, which is part of why Domain 4 grew.
The last row looks like a no-op because the task number is the same, but the DVA-C02 Task 3.3 was “Automate deployment testing” and the DVA-C03 Task 3.3 is “Deploy code by using AWS CI/CD services”. IaC templates moved from a testing context to a deployment context.
What was consolidated: 101 skills become 72
The comparison page names five consolidation areas. Here is each one with the DVA-C02 skills that were folded together, so you can confirm nothing you studied actually disappeared.
DynamoDB: four skills become one
DVA-C02 Task 1.3 listed four separate DynamoDB skills (1.3.1 high-cardinality partition keys, 1.3.2 consistency models, 1.3.3 query versus scan, 1.3.4 keys and indexing). DVA-C03 has one:
Skill 1.3.4: Configure DynamoDB to support applications (for example, by selecting high-cardinality partition keys, applying read consistency models, performing query and scan operations, by defining keys, attributes, and indexes).
Every DVA-C02 bullet survives inside the parenthesis. Our DynamoDB expressions deep dive and DynamoDB guide for the Developer Associate apply unchanged.
Security: 13 skills become 7
DVA-C02 Tasks 2.2 (encryption, 7 skills in version 2.1) and 2.3 (sensitive data, 6 skills) are merged into DVA-C03 Task 2.2 with 7 skills. The merged list drops the standalone “Use encryption across account boundaries”, “Enable and disable key rotation”, “Encrypt environment variables that contain sensitive data” and “Implement data access patterns for multi-tenant applications” as separate lines, folding them into “Apply encryption at rest and in transit”, “Use encryption keys to encrypt and decrypt data” and “Identify, mask, and limit access to sensitive data”. KMS key rotation and cross-account key grants are still fair game; they are now examples under a broader skill rather than headlines. Our Developer Associate security guide still covers them.
Testing: 11 skills become 5
DVA-C02 Task 3.2 “Test applications in development environments” (5 skills) and Task 3.3 “Automate deployment testing” (6 skills) become DVA-C03 Task 3.2 “Implement and automate application testing in AWS environments” with 5 skills: develop and validate tests with SAM/CDK/CloudFormation, mock APIs with API Gateway, create JSON test events, create application environments (Lambda aliases, image tags, Amplify branches, Copilot environments, API Gateway APIs), and the new AI testing skill.
Observability: 8 skills become 5
DVA-C02 Task 4.2 “Instrument code for observability” had 8 skills including the definitional “Describe differences between logging, monitoring, and observability” and the separate “Add annotations for tracing services”. DVA-C03 Task 4.2 “Apply logging, monitoring, and observability best practices” has 5: logging strategy with structured logging as the example, custom metrics, notification alerts, tracing, and workload health with liveness and readiness probes. Annotations did not vanish; they are how you “implement tracing”. Our X-Ray internals post covers annotations versus metadata in the depth the questions require.
Optimization: 9 skills become 6
DVA-C02 Task 4.3 had nine skills, several of them near-duplicates (“Profile application performance”, “Analyze application performance issues”, “Use application logs to identify performance bottlenecks”). DVA-C03 Task 4.3 has six, and the wording is notably more concrete: concurrency now names “AWS Lambda provisioned concurrency, Amazon EC2 Auto Scaling, thread pool management, connection pooling”; caching names “Amazon ElastiCache, Amazon DynamoDB Accelerator [DAX], Amazon CloudFront”; messaging names “Amazon SQS queues, Amazon SNS fan-out patterns”. DVA-C02’s “Use subscription filter policies to optimize messaging” and “Cache content based on request headers” are absorbed into those broader skills.
What quietly disappeared from the skill list
Consolidation means some DVA-C02 skills have no exact DVA-C03 counterpart. None of these is a topic you should stop studying, because the underlying services remain in scope, but the guide no longer calls them out:
- 1.1.1 “Describe architectural patterns (event-driven, microservices, monolithic, choreography, orchestration, fanout)” and 1.1.2 “stateful and stateless” are gone as separate skills. Event-driven patterns survive as 1.1.8; coupling survives as 1.1.1.
- 1.1.13 “Implement resilient application code for third-party service integrations (retry logic, circuit breakers)” is folded into 1.1.3 “Create and maintain fault-tolerant and resilient applications (retry with backoff, graceful degradation)”.
- 1.2.7 “Use Lambda functions to process and transform data in near real time” is gone; streaming survives as 1.1.6 and event source mappings as 1.2.4.
- 1.3.9 “Use specialized data stores based on access patterns (for example, Amazon OpenSearch Service)” becomes the broader 1.3.2 “Choose AWS data stores based on access patterns”. OpenSearch Service is still in scope.
- 2.1.2 “Secure applications by using bearer tokens” and 2.1.8 “Handle cross-service authentication in microservice architectures” are absorbed into 2.1.2 “Implement access control (fine-grained permissions, security tokens, MFA)”.
- 3.1.2 “Organize files and a directory structure” and 3.1.4 “Apply application requirements for resources (memory, cores)” have no direct heir; resource sizing lives in 4.3.2.
- 3.4.10 “Use existing runtime configurations to create dynamic deployments (staging variables from API Gateway in Lambda)” is absorbed into 3.1.4 “Manage environment-specific application configuration” and 3.1.5 “Use API Gateway features and resources (stage-specific URLs, endpoints)”.
- 4.1.4 “Implement custom metrics (CloudWatch embedded metric format [EMF])” loses the EMF example but survives as 4.2.2 “Implement code that emits custom metrics”.
In-scope services: what was added and removed
This is the list that candidates skim and should not. The DVA-C03 in-scope list has a new category and several new entries.
Added to the in-scope list
| Category | New in DVA-C03 | Note |
|---|---|---|
| Analytics | Amazon Data Firehose | Previously implied under “Amazon Kinesis”; now listed separately |
| Machine Learning and AI (new category) | Amazon Bedrock | Managed foundation-model access; the service behind Skill 1.1.10 and Task 2.3 |
| Machine Learning and AI | Amazon Bedrock AgentCore | The agent runtime and services behind Skill 2.3.4 (tool-use authorization, session isolation) |
| Machine Learning and AI | Amazon Quick | Listed in scope while Amazon Quick Sight remains out of scope; treat as awareness-level |
| Machine Learning and AI | Kiro | The AWS agentic IDE; “spec-driven code generation” in Skill 1.1.9 is its vocabulary |
| Management and Governance | Amazon CloudWatch Logs | Listed separately from CloudWatch; log queries and data protection are the hooks |
| Management and Governance | AWS SAM | Now listed explicitly (it was named in DVA-C02 skills but not in the service list) |
| Networking and Content Delivery | AWS PrivateLink | The mechanism behind Skill 2.3.2’s “VPC endpoints to establish private connectivity” |
| Security, Identity, and Compliance | AWS Certificate Manager (ACM) | Matches the new wording of Skill 2.2.2 |
| Security, Identity, and Compliance | AWS Private CA | Was named in a DVA-C02 skill but absent from the DVA-C02 service list |
| Storage | Amazon S3 Glacier | Matches Skill 1.3.1 “Manage data lifecycles (S3 Lifecycle policies, TTL, retention periods)“ |
Removed from the in-scope list
| Removed | Category in DVA-C02 | What to make of it |
|---|---|---|
| AWS AppSync | Application Integration | GraphQL APIs are no longer a named service. API Gateway remains the API service in scope. |
| Amazon Q Developer | Developer Tools | Replaced by the generic phrasing “AI-assisted development tools” and by Kiro in the service list. Q Developer skills you learned (code generation, test generation) map one-to-one onto Skills 1.1.9 and 3.2.5. |
The Amazon Q Developer removal is the most interesting line in the whole comparison. AWS added Q Developer to the DVA-C02 in-scope list in the December 2024 revision and wrote two skills around it; eleven months later the DVA-C03 guide names Kiro instead and describes the capability generically. The lesson for exam preparation is to learn the capabilities (completion, generation from a spec, automated review, test generation, security scanning, troubleshooting suggestions) rather than a product name, which is exactly how the skills are now worded.
Out-of-scope list changes
| Change | Service |
|---|---|
| Added to out of scope | Amazon SageMaker AI |
| Added to out of scope | Strands Agents |
| Added to out of scope | AWS Transfer Family |
| Removed from out of scope | Amazon Chime |
The SageMaker AI and Strands Agents lines draw the boundary of the AI content precisely. You consume models and build around managed agent services; you do not train models or build agents from an open-source framework. Combined with the out-of-scope job tasks below, this tells you how deep to go.
Services that did not change
Everything else carries over: Athena, Kinesis, OpenSearch Service; EventBridge, SNS, SQS, Step Functions; EC2, Elastic Beanstalk, Lambda; ECR, ECS, EKS; Aurora, DynamoDB, ElastiCache, RDS; Amplify, CloudShell, CodeArtifact, CodeBuild, CodeDeploy, CodePipeline, X-Ray; AppConfig, CDK, CloudFormation, CloudTrail, CloudWatch, AWS CLI, Systems Manager; API Gateway, CloudFront, ELB, Route 53, VPC; Cognito, IAM, KMS, Secrets Manager, STS, WAF; EBS, EFS, S3. Note that the DVA-C03 list writes well-known services by their short names (Amazon SNS, not Amazon Simple Notification Service), which the guide says is also how they appear on the exam to reduce reading load.
The out-of-scope job tasks: where the AI content stops
The DVA-C03 target-candidate section adds three job tasks to the “not expected to perform” list. These are your depth limiter for the AI material:
- Select foundation models (FMs), engineer AI prompts, and design Retrieval Augmented Generation (RAG) architectures
- Design AI safety frameworks, implement guardrails, and implement AI governance
- Design and create extract, transform, and load (ETL) pipelines
Read the second one carefully against Skill 2.3.3. The exam expects you to apply input and output controls that exist (content filters, denied topics, PII redaction, prompt-attack protection) and to recognise which control solves which scenario. It does not expect you to design a safety framework or a governance programme. Likewise, you need to know that a model must be invoked through a signed API call with the right permissions and that it can be reached privately; you do not need to compare model families or build a retrieval pipeline. If you also hold or are studying the AI Practitioner, our Bedrock Knowledge Bases internals post is deeper than DVA-C03 needs; the parts on access and private connectivity are the overlap.
The recommended-knowledge section also changed in tone. DVA-C02 asked for “Experience in using development tools”; DVA-C03 asks for “Experience in using application development tools, including AI tools” and “Familiarity with AI-assisted development workflows”, and adds “Integrate managed AI services into applications” to the recommended AWS knowledge. The target candidate is still someone with one or more years of hands-on AWS development experience.
What did not change
It is worth being explicit, because panic about an exam update usually comes from overestimating the churn.
- The Lambda material is intact: VPC access (1.2.1), configuration including memory, concurrency, timeout, runtime, handler, layers and extensions (1.2.2), destinations and dead-letter queues (1.2.3), event source mappings (1.2.4), packaging as .zip, container image or layers (3.3.1), provisioned concurrency (4.3.1). Our Lambda versions, aliases and concurrency guide and the SAM serverless deployments guide remain current.
- The CI/CD material is intact and slightly more explicit: CodeDeploy revisions (3.3.4), CodePipeline orchestration from linked repositories through multiple environments (3.3.5), blue/green, canary and rolling strategies (3.3.6), automatic or manual rollback with CloudFormation and CodeDeploy (3.3.7), labels and branches (3.3.8). See the CI/CD deployment guide.
- API Gateway is intact: request/response transformations and validation (1.1.4), stage-specific URLs and endpoints (3.1.5), mock integrations for testing (3.2.2). See the API Gateway REST vs HTTP integrations and authorizers guide.
- Security fundamentals are intact: identity federation (2.1.1), programmatic access (2.1.3), authenticated calls (2.1.5), IAM permissions (2.1.6), assuming a role (2.1.7), encryption at rest and in transit, client-side versus server-side, KMS keys, secrets management.
- Troubleshooting fundamentals are intact: debugging code and integration issues (4.1.1), interpreting metrics, logs and traces (4.1.2), querying logs (4.1.3), deployment failure logs (4.1.4). See the monitoring, troubleshooting and optimization guide.
- Elastic Beanstalk, Kinesis, Step Functions, Cognito, ElastiCache, DAX and CloudFront caching all remain in scope with unchanged depth.
If you have been studying for DVA-C02 for a while, roughly 85% of your preparation transfers directly. The remaining 15% is Task 2.3, the AI-tool skills, and the explicit container skills.
DVA-C02 or DVA-C03: a decision framework
You have until 30 November 2026 to sit DVA-C02. Here is how to decide.
Sit DVA-C02 before the deadline if:
- You can realistically be ready in the time left. Our Developer Associate study plan assumes six to eight weeks; if you are past the midpoint and scoring consistently in the high 70s on full-length mock exams, finishing on the version you prepared for is the lower-risk path.
- Your preparation material is DVA-C02-specific and you do not want to add the AI material under time pressure.
- Test-centre or online-proctoring availability in your region still exists for the dates you need. Late November demand will be high.
Re-plan for DVA-C03 if:
- You are early in preparation. The new content is not large, and preparing for the live version avoids the scenario where a failed November attempt forces you to re-prepare for a different exam in December anyway (there is a mandatory 14-day wait between attempts, which lands you on DVA-C03 regardless).
- You already work with Bedrock or an AI coding assistant day to day. Task 2.3 and the AI skills will be the easiest part of the exam for you.
- You want the credential to reflect the current guide. Both versions award the identical certification, valid for three years, so this is a preference rather than a requirement.
Either way, the credential is the same. Passing DVA-C02 on 30 November and passing DVA-C03 on 1 December produce the same AWS Certified Developer – Associate, with the same three-year validity and the same recertification path described in our renewal guide. Nobody will ask which version you sat.
A DVA-C03 study delta for DVA-C02 candidates
If you decide to switch, here is the incremental list. Everything else in your existing plan stands.
- Read Task 2.3 and build a one-page map of each of the five skills to a concrete control: model access enablement and IAM; interface VPC endpoints; guardrail policies (content filters, denied topics, sensitive-information filters, prompt-attack filters); agent tool authorisation, session isolation and human approval; invocation logging decisions and log data protection.
- Make one Bedrock call from code. Enable a model in a sandbox account, grant your role the invoke permission, call the runtime API from the SDK you use, and then make the same call through an interface endpoint with a policy that denies other principals. Thirty minutes of hands-on work covers Skills 1.1.10, 2.3.1 and 2.3.2 and makes the question wording recognisable.
- Attach a guardrail to that call and send it a prompt containing a fake phone number and a denied topic. Observe the blocked and masked outputs. That is Skill 2.3.3 in practice.
- Review the AI-assisted development vocabulary: spec-driven generation, automated code review, intelligent completion, refactoring suggestions, security scanning, generated tests, deployment approvals, error analysis, optimization suggestions. Expect scenario questions where the answer is “use the AI tool’s X capability” and the distractors are manual steps.
- Refresh ECR and ECS. Tag immutability, lifecycle policies that expire untagged or old images, task definitions, Fargate launch type and rolling updates are now named skills (3.1.2 and 3.3.2).
- Add CloudFront signed URLs and cookies next to S3 presigned URLs in your notes (2.1.4), and ACM versus Private CA for public versus private certificates (2.2.2).
- Re-weight your mock-exam review. Spend proportionally more time on Domain 4 misses than you did for DVA-C02.
For the mock-exam step, the AWS Developer Associate mock exam bundle gives you full-length timed exams whose questions follow the exam-guide skills with explanations for every option. Use the per-domain breakdown after each attempt to see whether your Domain 4 and security misses are shrinking, and use the explanations to turn each miss into a one-line note in the map from step 1. If you are sitting DVA-C02 before the deadline, the same exams cover the shared 85%; if you are switching, treat Task 2.3 as the area to supplement with the hands-on steps above.
Common misconceptions about the update
- “DVA-C03 is an AI exam now.” Ten of 72 skills mention AI tools or services, and the explicitly out-of-scope tasks exclude model selection, prompt engineering, RAG design, safety frameworks and governance. It is a developer exam that acknowledges developers use AI tools.
- “The exam got easier because there are fewer skills.” The guide says the opposite: fewer, broader skills with the same emphasis. A single DVA-C03 skill can generate the same questions as the three DVA-C02 skills it replaced.
- “My DVA-C02 certificate will be worth less.” Both versions award the same credential with the same validity. Exam versions are not printed on the certificate.
- “AppSync questions are gone, so GraphQL is gone.” AppSync left the in-scope list. API design in general (1.1.4) and API Gateway remain, and a question could still describe a GraphQL requirement as context. Do not spend study time on AppSync specifics.
- “I should learn Kiro, Amazon Quick and AgentCore in depth.” The skills that reference them are capability-level (“spec-driven code generation”, “tool-use authorization”, “session isolation”). Know what each service is for and which skill it answers.
- “Amazon Q Developer is out of scope, so AI coding assistants are out.” Only the product name left the list. The capabilities it provided are now Skills 1.1.9 and 3.2.5, worded generically.
Frequently Asked Questions
When is the last day to take DVA-C02 and when does DVA-C03 start?
The DVA-C03 exam guide states that the last day to take DVA-C02 is 30 November 2026 and that DVA-C03 is in use beginning 1 December 2026. Registration for DVA-C03 opened on 27 October 2026.
Did the number of questions, time limit or passing score change?
No. Both versions have 65 questions (50 scored, 15 unscored), 130 minutes, a scaled score of 100 to 1,000 with 720 to pass, and multiple-choice and multiple-response question types only.
What are the DVA-C03 domain weights?
Development with AWS Services 30% (down from 32%), Security 26% (unchanged), Testing and Deployment 22% (down from 24%, and renamed from “Deployment”), Troubleshooting and Optimization 22% (up from 18%).
What is new in DVA-C03?
A new Task 2.3 on identifying and mitigating security risks of using and integrating AI services (five skills), nine further AI-tool skills across Tasks 1.1, 3.2, 3.3, 4.1 and 4.3, explicit skills on building container images with ECR and deploying with ECS, EKS and Fargate, a named presigned-URL skill covering S3 and CloudFront, and a named certificate-management skill covering ACM and Private CA.
Which services were added to and removed from scope?
Added: Amazon Bedrock, Amazon Bedrock AgentCore, Amazon Quick, Kiro, Amazon Data Firehose, Amazon CloudWatch Logs, AWS SAM, AWS PrivateLink, AWS Certificate Manager, AWS Private CA and Amazon S3 Glacier. Removed: AWS AppSync and Amazon Q Developer. Newly out of scope: Amazon SageMaker AI, Strands Agents and AWS Transfer Family.
Is DVA-C03 harder than DVA-C02?
The guide describes a reorganisation with equal emphasis rather than a change in difficulty. The practical difference is new subject matter (AI security and AI-assisted tooling) that candidates with no exposure to Bedrock or AI coding tools will need to learn, and a slightly heavier Troubleshooting and Optimization domain.
If I pass DVA-C02 in November, is my certification different from a DVA-C03 pass?
No. Both award the AWS Certified Developer – Associate credential, valid for three years, with the same recertification options.
Do I need to know prompt engineering or RAG for DVA-C03?
No. The guide lists selecting foundation models, engineering prompts, designing RAG architectures, designing AI safety frameworks, implementing guardrails as a design activity, and AI governance as job tasks out of scope for the target candidate. You need to apply existing controls and integrate managed services, not design them.
Should I switch from DVA-C02 to DVA-C03 preparation?
If you can be ready before 30 November 2026 on your current plan, finishing on DVA-C02 is lower risk. If you are early in preparation, prepare for DVA-C03, because a failed attempt in late November would require the 14-day wait and put you on DVA-C03 anyway. Roughly 85% of the content is shared.
Conclusion
The DVA-C03 update is a targeted revision, not a rewrite. The exam format is identical, the four domains keep their names with one honest rename, and most of the service list and skills carry over. The real changes are a four-point shift of weight toward Troubleshooting and Optimization, a genuinely new security task about using AI services responsibly, nine AI-tool skills that promote last year’s unscored “emerging topics” into scored content, and explicit container and presigned-URL skills that were previously implicit. On the service side, Bedrock, Bedrock AgentCore, Kiro, Amazon Quick, Data Firehose, PrivateLink, ACM, Private CA and S3 Glacier arrive; AppSync and Amazon Q Developer leave; SageMaker AI and Strands Agents are fenced off as out of scope.
If you are mid-preparation, decide by the calendar: finish on DVA-C02 if you can sit before 30 November 2026 with confidence, otherwise add the small AI and container delta to your plan and sit DVA-C03 from 1 December. Either way, the credential you earn is the same, and the developer fundamentals that make up the bulk of both exams are the same fundamentals you have been practising all along.