What the DP-800 Exam Covers

DP-800, Developing AI-Enabled Database Solutions, sits directly between Microsoft SQL development and applied AI. The role is not an AI engineer who only happens to query a database, nor a database developer who can ignore embeddings and models. Microsoft expects candidates to design SQL solutions, secure and deploy them, and implement AI capabilities where they create value.

The DP-800 exam currently has three weighted domains. Microsoft has also published a minor skills refresh effective October 19, 2026. The exam’s overall shape remains stable, but candidates should verify the live guide close to exam day because detailed wording around database objects, AI-assisted SQL, models, embeddings, and intelligent search can change.

The candidate profile is still grounded in Microsoft SQL

Candidates should be comfortable with SQL Server, Azure SQL, and SQL databases in Microsoft Fabric. T-SQL development is central. The blueprint reaches database objects, programmability, advanced querying, security, performance, deployment, integration, models, embeddings, search, and RAG.

This matters because the AI features sit on top of the database foundation. A vector representation cannot repair a bad key design, and a language model cannot make an incorrect query trustworthy.

Design and develop database solutions is 35–40 percent

The first domain includes database object design, programmability, advanced T-SQL, and AI-assisted development. Tables, data types, indexes, specialized table types, JSON, constraints, sequences, and partitioning all belong here.

Views, functions, stored procedures, and triggers remain important. Advanced query work includes CTEs, window functions, JSON operations, pattern functions, fuzzy matching, graph queries, correlated queries, and error handling.

AI-assisted SQL development is part of the core developer role

DP-800 explicitly includes GitHub Copilot and Copilot in Fabric, model and MCP options, project instruction files, and connections to MCP server endpoints such as SQL Server or Fabric lakehouse services.

The general tool-use principles are useful because once an assistant can reach a database through a tool, authorization and validation matter as much as prompt quality.

Secure, optimize, and deploy is another 35–40 percent

Security topics include encryption, masking, row-level security, object permissions, passwordless access, auditing, model endpoint protection, and secure REST, GraphQL, and MCP endpoints.

Performance topics include database configuration, isolation and concurrency, execution plans, dynamic management information, Query Store, performance analysis, blocking, and deadlocks. You should be able to diagnose a condition from evidence, not merely recognize a product name.

Database CI/CD is a first-class exam skill

SQL Database Projects, tests, reference data, source control, branching, pull requests, conflict resolution, secret management, schema drift, deployment, and pipeline controls all appear in the guide.

The broader CI/CD fundamentals provide the delivery pattern. DP-800 adds stateful database concerns such as schema compatibility, controlled rollout, and recovery.

Integration with Azure services expands the SQL boundary

Data API builder configuration, REST and GraphQL exposure, caching, pagination, search, filtering, Azure Monitor, Application Insights, and change-handling patterns all appear in this domain.

The database therefore participates in a wider application architecture. Candidates need to recognize where SQL responsibility ends and an API, event path, function, or monitoring service begins.

AI capabilities account for 25–30 percent

The final domain covers external models, embeddings, intelligent search, and retrieval-augmented generation. It is smaller than either database domain, but it joins several disciplines: AI concepts, SQL implementation, data freshness, search quality, and endpoint security.

Know how models are evaluated and managed, how embedding inputs are chosen and maintained, and how search behavior is measured rather than assumed.

Intelligent search includes lexical, vector, and hybrid choices

The guide includes full-text search, vector data and indexing, vector-related SQL functions, nearest-neighbor approaches, hybrid search, ranking fusion, and performance evaluation. Exact terminology can evolve, so confirm the live function list before exam day.

The vendor-neutral architecture in embeddings, vector databases, and RAG is useful background, but DP-800 expects implementation judgment in Microsoft SQL.

RAG is built from SQL results outward

DP-800 includes identifying RAG use cases, preparing prompt content from SQL, converting structured data for language-model processing, sending results to a model, and extracting the response.

That makes data shaping, security, and result handling part of the RAG design. RAG is not a separate chatbot layer floating above the database.

The October 2026 update is a minor refresh

Microsoft’s change log shows no change to the main functional groups. Minor updates affect detailed objectives within database objects, AI-assisted SQL, models and embeddings, and intelligent search.

If your exam is after the effective date, compare the updated guide with the earlier version and focus review on the changed bullets instead of rebuilding your study plan from zero.

Readiness requires both coding and architecture judgment

You should be able to write and review T-SQL, but also choose between security controls, index strategies, deployment patterns, integration mechanisms, retrieval methods, and model-enabled features.

A candidate who studies only syntax will struggle with architecture questions. A candidate who studies only AI concepts will struggle with database mechanics and operational evidence.

Know the technologies Microsoft expects you to recognize

The role spans T-SQL, Microsoft SQL platforms, Git and source control, CI/CD, Azure integration services, AI-assisted development, external models, embeddings, vector search, and RAG. You do not need to become a specialist in every surrounding Azure product, but you should know why the database solution integrates with them.

This breadth explains why isolated memorization performs poorly. The exam repeatedly connects database mechanics to application delivery and AI behavior.

Expect scenario questions that cross domain boundaries

A single scenario can involve database design, endpoint security, performance, and AI retrieval at once. For example, a search feature may require a vector representation, a filter on customer scope, a secure model endpoint, and a deployment strategy that keeps embeddings current.

Practice identifying the primary requirement first, then eliminate answers that violate security, correctness, or maintainability even if they technically satisfy one part of the scenario.

Do not over-focus on preview terminology

AI database features are evolving quickly. Microsoft may rename functions, refine supported index types, or adjust objective wording as features mature. Learn the architecture and purpose behind the feature so a naming change does not invalidate your understanding.

Before exam day, review the current study guide and current documentation for the specific functions and configuration names that Microsoft expects candidates to recognize.

Use the certification role as the organizing principle

The SQL AI Developer role is responsible for building a database solution that other applications can rely on. That means data integrity, secure access, performance, deployability, observability, and AI features all matter together.

If a study topic cannot be connected to a developer responsibility—designing, coding, securing, tuning, exposing, deploying, or adding AI capability—recheck whether you are spending time on material outside the exam’s practical scope.

Security and AI objectives often appear in the same scenario

A language-model feature may depend on an external endpoint, database permission, API layer, or retrieval path. That means an AI question can also test secure authentication, endpoint protection, or least privilege. Do not isolate the final domain from the security domain while studying.

Practice asking which identity performs each operation and which control protects the data before it reaches the model. That habit eliminates many attractive but unsafe answer choices.

Performance still matters after AI is added

Vector search, embedding refresh, API exposure, and model calls can add new latency, but conventional database performance remains part of the user experience. Slow joins, blocking, and poor indexing can make an AI-enabled application appear to have a model problem when the bottleneck is SQL.

Use ordinary database evidence first. DP-800 expects you to reason about the whole solution, not to blame every new workload issue on the AI layer.

Expect deployment questions to include secrets and environment differences

Database projects and pipelines often move between development, test, and production environments with different endpoints, identities, connection settings, and model configurations. Keep those values out of source code and use controlled environment-specific configuration.

When reviewing a scenario, distinguish the versioned database definition from the deployment secrets and runtime configuration that should be injected separately.

Use current documentation for fast-moving AI features

The database fundamentals in DP-800 are relatively stable, while model, embedding, vector, and AI-assisted development features can move faster. Treat the Microsoft Learn study guide as the final authority for exam scope and use current feature documentation to confirm syntax or support status close to exam day.

This keeps preparation current without letting preview-era details dominate the durable database concepts.

Use the blueprint as a dependency map

Database design supports integrity and performance. Security protects data and model endpoints. CI/CD governs change. Integration exposes data safely. Embeddings depend on source quality. Search feeds RAG. The domains reinforce one another.

Studying those relationships is more effective than treating each bullet as a separate fact. DP-800 measures the ability to build AI-enabled database solutions that remain secure, maintainable, and operable.