DP-800 Study Guide: How to Pass the SQL AI Developer Associate Exam
Domain weights straight from Microsoft's official DP-800 study guide, a 4-week plan, and the overlapping features (Always Encrypted vs. masking, ANN vs. ENN, CDC vs. CES) that actually decide whether you pass.
TL;DR: DP-800 (Developing AI-Enabled Database Solutions) is Microsoft's Associate-level exam for the SQL AI Developer Associatecredential, requiring a score of 700+. Three domains: design and develop database solutions (35-40%), secure/optimize/deploy database solutions (35-40%), and implement AI capabilities in database solutions (25-30%) — the smallest domain by weight but the one with the least prior-knowledge overlap for most SQL developers, since it covers vector search, embeddings, and RAG built directly into T-SQL. Skills measured as of March 12, 2026, per Microsoft's official study guide (checked September 7, 2026).
Who DP-800 Is Actually For
Microsoft's official audience profile describes someone with subject-matter expertise designing and developing AI-enabled database solutions across Microsoft SQL platforms — Microsoft SQL Server, Azure SQL, and SQL databases in Microsoft Fabric. You're expected to already write T-SQL comfortably and have built databases on these platforms before you start studying for the AI-specific content.
Beyond T-SQL, the profile calls out three things a lot of experienced database developers haven't touched yet: CI/CD practices in GitHub applied to database schemas (not just application code), AI-assisted development tools inside the database tooling itself, and core AI concepts — embeddings, vectors, and models — as first-class T-SQL objects rather than external services you call from an app.
You'll work across four responsibilities on the real job and the real exam: designing solutions for structured and semi-structured data, integrating AI features into scalable applications, securing/optimizing/deploying what you build, and implementing AI capabilities natively in the database — often while coordinating with DBAs, architects, AI engineers, and DevSecOps engineers who own adjacent pieces of the same system.
Positioned against neighboring Microsoft certifications, DP-800 sits closer to the database-developer end of the data and AI path than DP-300 (Administering Relational Databases on Microsoft Azure), which is squarely a DBA-operations exam, or AI-102 (Azure AI Engineer), which builds AI applications from outside the database rather than embedding AI capabilities inside T-SQL itself. If your background is application-facing AI engineering rather than database development, expect the domain 1 and domain 2 content — specialized tables, advanced T-SQL, CI/CD for schema changes — to take noticeably more study time than the AI domain.
| Detail | Information |
|---|---|
| Exam code | DP-800 |
| Exam title | Developing AI-Enabled Database Solutions |
| Credential | Microsoft Certified: SQL AI Developer Associate |
| Level | Associate (Intermediate) |
| Product / Role | Azure / Developer |
| Passing score | 700 (Microsoft's scaled scoring system) |
| Prerequisites | None formally required; T-SQL and SQL platform experience assumed |
| Renewal | Annual, free online assessment on Microsoft Learn |
| Skills measured as of | March 12, 2026 |
The Three DP-800 Domains and How Much Each Is Worth
| Domain | Weight | What it tests |
|---|---|---|
| Design and develop database solutions | 35-40% | Tables and specialized table types, programmability objects, advanced T-SQL (JSON, regex, fuzzy matching, graph queries), AI-assisted development tooling |
| Secure, optimize, and deploy database solutions | 35-40% | Data security and compliance, performance tuning, CI/CD via SQL Database Projects, integrating SQL solutions with Azure services (Data API builder, Azure Monitor, change handling) |
| Implement AI capabilities in database solutions | 25-30% | Models and embeddings, intelligent search (full-text, vector, hybrid), retrieval-augmented generation (RAG) |
Design and develop database solutionsis the domain most experienced SQL developers will recognize fastest, but it goes further than routine schema design. Beyond ordinary tables, indexes, and constraints, it expects you to know when a specialized table type earns its complexity — a ledger table for tamper-evident audit trails, a temporal table for point-in-time history, an in-memory table for extreme throughput, or a graph table for relationship-heavy data queried with the MATCH operator. The advanced T-SQL portion adds function families most developers haven't needed before: fuzzy string matching (JARO_WINKLER_DISTANCE, EDIT_DISTANCE) for deduplication-style problems, and regular-expression functions for pattern extraction that used to require CLR code or an application-side library.
Secure, optimize, and deploy database solutions is really three sub-skills bundled into one domain: locking down data (encryption, masking, row-level security, auditing), keeping it fast (execution plans, DMVs, resolving blocking and deadlocks), and shipping schema changes safely (SQL Database Projects with source control, schema drift detection, and deployment pipeline controls). The Azure-integration slice — Data API builder, Azure Monitor, and change-handling mechanisms like CDC and Change Tracking — is where this domain overlaps most with the AI domain, since embedding maintenance depends on the same change-detection tooling.
Implement AI capabilities in database solutions carries the lowest weight of the three but the highest concentration of genuinely new material for most candidates. This is where T-SQL stops being just a data-manipulation language and starts acting as an AI orchestration layer: generating and maintaining embeddings, running vector and hybrid search with reciprocal rank fusion, and calling an external language model directly from a stored procedure via sp_invoke_external_rest_endpoint to build a retrieval-augmented generation flow without ever leaving the database.
How to Split Your DP-800 Study Time by Domain
The two largest domains are near-identical in weight, but they aren't equally unfamiliar. Most candidates already write T-SQL daily, so weight extra hours toward the AI domain and the newer parts of domain 2 rather than splitting evenly:
| Domain | Suggested hours (of ~28 total) |
|---|---|
| Design and develop database solutions (35-40%) | ~9 hours |
| Secure, optimize, and deploy (35-40%) | ~9 hours |
| Implement AI capabilities (25-30%) | ~10 hours |
The AI domain gets slightly more hours than its raw weight implies because vector search, embedding maintenance, and RAG in T-SQL are the parts least likely to overlap with what an experienced SQL developer already knows.
A Four-Week Plan to Prepare for DP-800
Tables including in-memory, temporal, external, ledger, and graph types; JSON columns/indexes; constraints; SEQUENCES; partitioning; views, functions, stored procedures, triggers.
CTEs, window functions, JSON functions (JSON_OBJECT, OPENJSON, JSON_VALUE), regex functions, fuzzy matching (JARO_WINKLER_DISTANCE), graph MATCH queries, error handling; GitHub Copilot and Copilot in Fabric with MCP tool configuration.
Always Encrypted, Dynamic Data Masking, Row-Level Security, object-level permissions, auditing; execution plans, DMVs, Query Store; SQL Database Projects with schema drift detection; Data API builder and Azure Monitor integration.
Embedding maintenance methods, chunking, full-text/vector/hybrid search, ANN vs. ENN, reciprocal rank fusion, RAG with sp_invoke_external_rest_endpoint. Finish with full-length practice questions and revisit any domain scoring under 80%.
For hands-on practice, MSCertQuiz maintains a full DP-800 practice test (40 questions free, 500 total) weighted against the domains above — see the practice questions and cheat sheet that go with this guide. Microsoft's own Microsoft SQL documentation and Microsoft Fabric documentation cover the product surface directly.
Where DP-800 Candidates Actually Lose Points
Always Encrypted vs. Dynamic Data Masking vs. Row-Level Security
Three security features that get grouped together and tested against each other. Always Encrypted hides data from anyone reading the column, including DBAs. Masking hides a value in query results without touching storage. RLS restricts which rows a query can return at all. Confusing "hide a column" with "hide a row" is the single most common miss.
ANN vs. ENN vector search
Approximate Nearest Neighbor trades a small amount of precision for speed at scale; Exact Nearest Neighbor guarantees the true closest matches but doesn't scale the same way. Scenario questions describe the data volume and accuracy requirement, not the acronym — you have to infer which one fits.
CDC vs. Change Tracking vs. CES vs. Azure Functions SQL trigger binding
All four detect that a row changed, but they answer different questions. Change Tracking says what changed. CDC captures before/after values for auditing. CES streams changes continuously to a target. A SQL trigger binding fires a function the moment a specific change happens. The exam tests picking the cheapest mechanism that satisfies the stated latency and detail requirement.
REST vs. GraphQL in Data API builder
DAB can expose the same database object as both a REST and a GraphQL endpoint, including GraphQL relationships across tables. Questions about "expose related data efficiently in one request" usually point to GraphQL, not REST.
SQL Database Projects vocabulary
If your CI/CD experience predates SDK-style database projects, terms like schema drift detection and reference/static data in source control won't be familiar even though the underlying idea (keep the database in sync with source control) is not new.
MSCertQuiz sells practice-exam access for DP-800 and other Microsoft certifications; this guide is written by the same team that builds those questions. The official Microsoft Learn resources linked above are what you need at minimum, free.
DP-800 Study Guide FAQ
Does DP-800 require a prerequisite certification?
No prerequisite exam is listed. Real T-SQL and SQL platform experience, plus basic AI concepts, are assumed background rather than a gate.
How is DP-800 different from DP-300?
DP-300 is DBA operations (backup, HA/DR, monitoring). DP-800 is developer-focused: schema design, advanced T-SQL, and building AI features into the database itself.
Does DP-800 cover training machine learning models?
No — it covers calling and evaluating external models you didn't train: embeddings, vector/hybrid search, and RAG. Model training is out of scope.
Which SQL platforms does DP-800 cover?
Microsoft SQL Server, Azure SQL, and SQL databases in Microsoft Fabric, treated as one portable T-SQL surface.
What are the DP-800 exam price, duration, and question count?
Not published on the DP-800-specific pages as of this check. Microsoft's general Associate-exam policy (roughly 100 minutes, 40-60 questions, ~$165 USD) is the closest public baseline, but it isn't confirmed for this exam specifically.
Does the DP-800 certification expire?
Yes, annually, renewed free through a Microsoft Learn assessment — standard policy for Microsoft associate certifications.
Is there an official DP-800 practice assessment?
Yes, through AI Skills Navigator, linked directly from Microsoft's DP-800 exam page (sign-in required).
More DP-800 Resources
Full exam details and the complete 500-question practice bank.
Scenario questions with full rationale, grouped by domain.
Every domain as a quick-reference table, plus decision tables for the exam's most-confused feature pairs.
A data-fundamentals starting point if DP-800 feels like a big jump.
See How Your DP-800 Knowledge Holds Up
Start with 40 free DP-800 questions calibrated to the domains above — no card required.
Start Free DP-800 Practice