Microsoft Build 2026: Building agentic apps with Microsoft Fabric and Microsoft Databases

1 week ago 2

Microsoft Build 2026 highlights advancements successful app improvement with Microsoft Fabric and Microsoft Databases, emphasizing a unified information and AI level for scalable, agentic applications.

AI is driving a cardinal displacement successful however enactment gets done and however applications are built. As the 2026 Microsoft Work Trend Index Report highlights, a increasing stock of workers are moving beyond asking questions to handing disconnected full tasks and orchestrating multi-agent systems.

This displacement introduces a caller constraint. The situation is nary longer exemplary capability, but consistent, shared information discourse crossed the business.

Developers and concern users cognize what they privation to automate, and today’s models tin deliver. The bottleneck is context. Every caller cause starts from zero, relearning however the concern works, wherever information lives, and what rules to follow. Without a accordant foundation, agents can’t coordinate oregon scale.

That’s the situation we are solving with Microsoft Fabric. It provides a unified information and AI level that empowers you to bring unneurotic information and determination from isolated AI experiments to production-ready cause systems, successful which each caller cause builds connected shared organizational context. This imaginativeness is already driving beardown momentum among the millions of developers gathering connected Fabric and Microsoft Databases.

At Microsoft Build, we are extending this instauration with caller capabilities that assistance developers determination from prototype to accumulation faster. These include Rayfin, a caller bundle improvement kit (SDK) and command-line interface (CLI) designed to marque Fabric a production-ready exertion backend, and Azure HorizonDB, a caller PostgreSQL database designed for AI‑powered applications, present successful nationalist preview.

Introducing Rayfin: From punctual to accumulation backend

Coding agents are accelerating app development. Moving those applications from prototype to production, however, remains a challenge. Agent-created oregon not, each production-ready exertion inactive relies connected a backend to negociate data, enforce individuality and permissions, coordinate state, and run reliably implicit time. Existing software-service platforms were either not designed for agents oregon bash not afloat conscionable endeavor requirements for deployment, security, and governance.

Rayfin, a caller open-source SDK and CLI, is designed to adjacent that gap. It lets developers and coding agents picture what to physique and get an enterprise-grade exertion backend straight into the exertion code, including a database, authentication, and more. Rayfin past deploys straight to Microsoft Fabric, giving each exertion enterprise-grade information and standard from time one. Developers and AI agents tin present determination from punctual to accumulation without managing infrastructure.

With Rayfin, developers enactment done acquainted GitHub‑based workflows to specify information models, backend logic, and entree policies wholly successful code, giving teams and agents a consistent, programmable interface for gathering and managing applications. Watch Rayfin successful action:

Because Rayfin tin beryllium deployed straight connected Fabric, exertion information lands straight successful OneLake, wherever it is instantly disposable to the afloat Fabric information stack, unified with analytics, operational and real-time data, and AI engines by default. This enables developers to physique their endeavor apps connected trusted concern logic, integrate with semantic models, and embed affluent information visuals.

We’re excited to spouse with Replit, a starring AI coding platform, to assistance customers physique enterprise-grade apps successful the interface they cognize and emotion portion keeping app, data, and services managed successful their ain Fabric tenant.

Rayfin unlocks a caller improvement exemplary for our users. Agents constitute the code. Fabric ships it rapidly and safely. Together, we’re giving developers thing they’ve ne'er had before: a way from thought to enterprise-grade accumulation that’s measured successful hours, not months.

—Amjad Masad, Chief Executive Officer (CEO) of Replit

Learn much astir Rayfin by watching the Microsoft Build league “BRK225 – Data, apps, and agents: the aboriginal of app dev with Microsoft Fabric” connected Wednesday, June 3, 2026, astatine 1:30 PM PT.

Microsoft Databases, designed for AI applications

For decades, databases person been the backbone of endeavor applications. As applications go much intelligent and agent‑powered, we are evolving Microsoft Databases into the instauration optimized for real‑time, AI‑ready, and operationally affluent experiences.

Azure HorizonDB: Enterprise-ready PostgreSQL built for the demands of AI applications

As a starring PostgreSQL committer, Microsoft has agelong invested successful the PostgreSQL community. But arsenic AI‑powered applications spot caller strains connected scale, latency, and resilience, the request for a caller people of PostgreSQL database is clear.

Azure HorizonDB is that adjacent step. Now disposable successful nationalist preview, HorizonDB is simply a afloat managed, PostgreSQL‑compatible database that combines PostgreSQL familiarity with cloud‑scale architecture. It’s portion resilient by default and delivers elastic retention that scales to 128 TB, monolithic scale‑out compute up to 3,072 vCores, and tin prolong sub‑millisecond, multi‑zone perpetrate latency for demanding transactional scenarios.

As our information demands person expanded exponentially due to the fact that of our usage of Azure AI to chat with our data, HorizonDB has travel astatine the cleanable clip to conscionable the performance, scale, and information we request to displacement into this caller satellite of AI-enabled data.

—Rand Morimoto, President of Convergent Computing (CCO)

Beyond scalability, we’ve besides infused Azure HorizonDB with experiences designed specifically for AI applications similar vector search, integrated AI exemplary management, and nonstop connectivity to Microsoft Foundry and Fabric. These features supply a modern instauration for building, modernizing, and scaling AI‑powered applications with confidence. Mohsin Shafqat, Director of Software Engineering astatine NASDAQ, mentioned, “What stood retired with HorizonDB is that it aligns intimately with however we already deliberation astir the problem. Instead of stitching unneurotic aggregate components, it brings transactional data, vector search, and AI capabilities into a azygous platform, which simplifies the architecture without forcing a implicit rethink.”

Learn much by watching the Microsoft Build league “BRK223 – From rows to reasoning: Designing databases for AI apps and agents” connected Tuesday, June 2, 2026, astatine 2:30 PM PT.

New information and migration tooling for Azure Database for PostgreSQL

As Azure HorizonDB powers a caller people of high-scale, mission‑critical applications built for AI, Azure Database for PostgreSQL remains a trusted, open‑source instauration for modernizing and operating existing PostgreSQL workloads.

Today’s updates present two meaningful enhancements. First is the Microsoft Defender for Cloud integration, present successful preview, which delivers continuous information and compliance assessments to assistance teams place misconfigurations and trim risk. Second, we’ve released caller find and appraisal tooling to assistance you much confidently program migrations to Azure Database for PostgreSQL. These tools measure Oracle and PostgreSQL environments and supply readiness insights, sizing guidance, and outgo estimates. Learn much astir this migration tooling successful the Microsoft Build on-demand league “OD822 – Smarter PostgreSQL migrations to powerfulness modern, intelligent apps.”

Powering intelligent, multi‑agent systems astatine planetary standard with Azure Cosmos DB

Azure Cosmos DB is Microsoft’s AI-ready NoSQL and vector database for gathering responsive applications and intelligent AI experiences astatine immoderate scale. OpenAI, for example, chose Azure Cosmos DB arsenic its superior operational database “because of its automatic scaling and schema-less flexibility, allowing america to iterate quickly,” said Nick Cooper, elder method unit subordinate astatine OpenAI.

At Microsoft Build, we are focused connected improving developer productivity and AI quality. The Azure Cosmos DB Linux Emulator is present mostly available, enabling developers to build, test, and validate applications locally crossed Linux, macOS, and Windows without a unreality dependency. New AI capabilities are besides present successful preview, including semantic reranking, which improves hunt relevance utilizing built‑in contextual understanding. In addition, a caller cause representation toolkit helps developers standardize persistent representation for AI agents utilizing Azure Cosmos DB, Azure Durable Functions, and Microsoft Foundry models. Learn much successful the Microsoft Build on‑demand league “OD820 – Designing reliable multi-agent apps with Azure Cosmos DB.”

Unifying databases and Fabric connected a azygous platform

Microsoft Databases tin beryllium centrally managed done the caller Database Hub successful Fabric, presently successful backstage preview, and mirrored into OneLake, bringing operational and analytical information onto a azygous foundation. From there, you tin usage Fabric to marque it trusted, contextual, and acceptable for AI.

Building an AI‑ready information instauration with Microsoft Fabric

In the epoch of AI, information is the fuel, but information unsocial is not enough. Equally important is however that information is understood: the definitions of customers, orders, products, revenue, and the relationships betwixt them. Today, that knowing is fragmented crossed lawsuit narration absorption (CRM) and endeavor assets readying (ERP) systems, productivity tools, and spreadsheets, and excessively often it does not question with the data. Organizations person agelong relied connected radical to recreate this context. But arsenic agents instrumentality connected much responsibility, this spread becomes critical. Without a shared knowing of the business, agents cannot reliably reason, coordinate, oregon act.

Microsoft IQ addresses this missing furniture by unifying endeavor quality into a shared instauration built to activate AI agents. It enables accordant reasoning and enterprise-scale impact, alternatively than isolated interactions and and brings unneurotic 4 interconnected capabilities: Work IQ captures however enactment happens, Fabric IQ models however the concern operates, Foundry IQ enables agents to observe and reuse knowledge, and the caller Web IQ, announced contiguous astatine Microsoft Build, adds real-time planetary discourse from the web.

Fabric IQ is cardinal to this system. It powers a continuous operational loop wherever radical and agents observe unrecorded signals, crushed implicit shared context, and instrumentality governed enactment successful the infinitesimal crossed analytics, operations, and the productivity tools wherever enactment happens. It provides 3 integrated layers of concern context:

  1. Unified data: OneLake unifies the organization’s information estate, spanning analytical and operational information into a single, accessible layer.
  2. Business intelligence: Semantic models supply structured, governed representations of that information which organizations already trust connected for trusted concern metrics and analyse their business.
  3. Operational intelligence: Ontologies seizure operational discourse by defining concern entities and their relationships truthful agents tin crushed successful the connection of the business. This discourse tin see unrecorded signals from Fabric Real-Time Intelligence, enabling organizations and their agents to recognize what is happening close present and enactment successful clip to alteration outcomes.
Fabric IQ layered architecture showing Unified Data arsenic the foundation, Business Intelligence (semantic models) supra it, and Operational Intelligence (ontologies) connected top.

This three-tiered instauration helps guarantee that each cause starts with the aforesaid knowing of the concern and tin use it correctly crossed workflows. But Frontier organizations cannot commencement astatine the IQ layer. Building this capableness requires a unified information foundation. Microsoft Fabric delivers this done 4 halfway capabilities:

  1. Unifying your information estate
  2. Processing and harmonizing data
  3. Curating semantic meaning  
  4. Empowering AI agents to act

Learn much astir however Fabric tin assistance you make an AI-ready information instauration successful the Microsoft Build on-demand league “OD811 – Powering the adjacent AI frontier with a unified information platform.”

1. Unifying your information property with Microsoft OneLake

Most organizations conflict to spot their full information estate. It is dispersed crossed systems, duplicated successful aggregate places, and owned by antithetic teams, making it hard to cognize what exists, wherever it lives, and however it connects. Microsoft OneLake brings it unneurotic into a single, AI‑ready information water that unifies your multi‑cloud property and enables organization‑wide entree for analytics and AI.

We are making it easier to link existing information to OneLake without moving oregon duplicating it with the merchandise of shortcuts to SharePoint and OneDrive into wide availability and the quality to make shortcuts straight from Fabric Data Warehouses, present successful preview. We are besides adding the preview of workspace-level Azure Private Link enactment for mirrored information sources.

The wide availability of the OneLake catalog successful Microsoft Foundry

At the aforesaid time, we are making it easier to link that information to AI. With the recent wide availability of the OneLake catalog successful Microsoft Foundry, you tin observe trusted data, research affluent metadata, and link it straight to AI solutions. The catalog is embedded wrong Foundry’s Knowledge experience, making it elemental to determination from information find to AI improvement successful a azygous workflow.

Learn much astir each of these OneLake announcements by watching the Microsoft Build on-demand league “OD815 – Unify your full information property connected a single, AI-ready information lake.”

2. Process information faster with a caller people of GPU-accelerated analytics

Once information is unified, the adjacent situation is turning it into insights quickly, reliably, and astatine scale. We are introducing GPU-acceleration built straight into Fabric Data Warehouse to unlock a caller level of show without adding complexity.

The probe down this innovation was precocious recognized by ACM SIGMOD arsenic the “Best Industry Paper of 2026.” This breakthrough establishes Fabric Data Warehouse arsenic the archetypal afloat managed information warehouse to connection GPU acceleration.

By integrating NVIDIA accelerated computing, query acceleration successful Fabric Data Warehouse fundamentally changes however accelerated queries tin run. In interior benchmarking conducted successful May 2026, the GPU-accelerated Fabric Data Warehouse delivered up to 7x faster show comparative to 3 comparable outer vendors for reporting and exertion workloads astatine 64-user concurrency.

This displacement reflects a broader alteration successful however modern information systems request to operate. As Ian Buck, Vice President of Hyperscale and HPC astatine NVIDIA, explains, “AI applications are redefining however a information warehouse needs to perform. As AI agents crushed implicit endeavor data, analytics systems request low-latency show for galore simultaneous users. With NVIDIA accelerated computing and customized CUDA kernels built straight into Microsoft Fabric Data Warehouse, Microsoft is bringing the SQL workflows customers already usage into the accumulation AI era.”

Customers are already seeing that impact. At UNC Health, “We’re seeing up to 5x betterment successful our query speeds, which allows our teams to walk little clip managing show and much clip delivering meaningful insights,” said Shaun McDonald, IT Manager.

Query acceleration works automatically wrong Fabric, speeding up queries without requiring immoderate query rewrites. The effect is consistently low‑latency analytics that powerfulness responsive applications, interactive reporting, and agent‑driven analysis, delivering fresher insights and greater assurance successful data.

Query acceleration volition beryllium disposable for an aboriginal entree preview successful the adjacent fewer weeks. You tin larn much by watching the Microsoft Build on-demand league “OD813 – Powering modern information analytics successful Fabric Data Warehouse.”

3. Curating semantic meaning with Fabric IQ, present mostly available

Once information is prepared, the adjacent situation is not entree but understanding. Most organizations deficiency a shared furniture of concern context, forcing each cause to relearn however the concern works from fragmented data. Fabric IQ, present mostly available, addresses this gap.

With information unified successful OneLake, Power BI’s industry-leading semantic models past supply structured representations of your information for trusted concern intelligence, serving arsenic an perfect instauration for grooming agents.

Ontologies successful Fabric IQ, expected to beryllium mostly disposable successful the coming months, widen semantic models by adding operational context. They specify concern entities, relationships, properties, rules, and actions, and link to unrecorded signals from Fabric Real-Time Intelligence. Operations agents, present mostly available, past crushed implicit shared unrecorded context, marque decisions based connected policy, and instrumentality enactment successful the moment. Running connected the governed instauration of Fabric and integrated with Microsoft Foundry, operations agents determination beyond answering questions to driving enactment and outcomes.

GIF of entities, relationships, properties, actions and concern objectives each being created based connected products, inventory, stores, shipments, buyers, and suppliers

Announcing the wide availability of graph and readying successful Fabric

We’re announcing the wide availability of graph successful Fabric, with wide availability of the readying successful Fabric coming aboriginal this month. Available today, graph introduces a highly scalable, relationship‑first exemplary that connects concern entities, systems, and signals truthful teams and agents tin recognize however changes propagate crossed the endeavor and enactment with afloat context. Planning extends this instauration by enabling organizations to make plans, budgets, forecasts, and script models connected apical of Fabric’s semantic models. Notably, readying successful Fabric is not conscionable static outputs. They tin beryllium written backmost into Fabric to thrust execution, enabling closed-loop alignment with the aforesaid strategy of information and context.

Extending Fabric IQ to Microsoft Foundry and Agent 365

Today, we are extending Fabric IQ crossed the cause ecosystem truthful this shared knowing tin beryllium utilized consistently crossed each cause and application.

Microsoft Foundry and Agent 365 

Now successful preview, Ontologies are accessible straight from Microsoft Foundry arsenic cognition sources, bringing trusted concern discourse into some customized and built-in cause experiences. Also successful preview, Fabric IQ is present integrated with Microsoft Agent 365 arsenic a first-party exemplary discourse protocol (MCP) tool, enabling organizations to crushed agents successful shared meaning and guarantee accordant behaviour crossed their cause estate.

Microsoft 365 Copilot: Cowork and Copilot Chat 

Fabric IQ is besides extending into Microsoft 365 Copilot, including Cowork and Copilot Chat. This enables agents to entree governed Fabric data, starting with Power BI reports and semantic models, to crook insights into action. Instead of static dashboards, agents tin observe changes successful cardinal metrics, make updates, trigger follow-ups, and docket adjacent steps, each grounded successful trusted, governed data. The effect is faster, much accordant execution crossed the organization. These experiences are presently disposable customers successful Frontier with a Microsoft 365 Copilot license. Join the programme today.

GitHub Copilot CLI 

Using Agent Skills for Fabric, Fabric IQ tools and skills for information insights are accessible done GitHub Copilot CLI, bringing governed semantic discourse straight into the terminal. Now you tin besides query Power BI reports and semantic models from the bid line, grounded successful governed semantic discourse successful Fabric IQ. Teams tin inquire earthy connection questions astir usage, metrics, oregon lawsuit behaviour and get answers grounded successful Fabric information straight wrong their workflow, reducing back-and-forth and accelerating data-driven decisions. 

Learn much astir these enhancements successful the Build on‑demand league “OD812 – Bringing Enterprise Ontology Directly into the Developer Workflow.”

4. Empowering agents to enactment with operations agents and Copilot successful Fabric

Enterprises are progressively moving to multi-agent systems made up of specialized agents grounded successful circumstantial information oregon domain expertise. These agents tin beryllium reused crossed aggregate systems, making it easier to standard and present much accordant outcomes.

Announcing the wide availability of operations agents

Microsoft Fabric supports this displacement with autochthonal cause capabilities, including Fabric information agents and operations agents. I’m excited to stock operations agents are present mostly available. These agents are designed to continuously show real-time data, observe patterns oregon anomalies, and enactment based connected predefined concern logic.

Apply agentic analytics successful Power BI

In addition, we precocious released open-source Agent Skills for Fabric, designed to marque it easier for AI tools and agents to interact straight with Fabric. These capabilities present widen to Power BI, enabling developers and analysts to make reports utilizing earthy connection oregon adjacent a screenshot of the desired outcome, importantly accelerating however insights are built and shared.

Copilot successful Power BI tin present besides modify semantic models, with built-in recommendations that amended show and marque models much AI-ready. This helps teams iterate faster and present insights much rapidly without leaving the tools they already use.

Learn much by speechmaking the Power BI blog or by watching the Microsoft Build on-demand league “OD817 – Agentic analytics with Power BI and Microsoft Fabric.”

Watch these announcements from Microsoft Build

If you’re funny successful learning much and seeing unrecorded demos connected these announcements, I promote you to ticker the Microsoft Database and Microsoft Fabric sessions astatine Microsoft Build, disposable successful the Microsoft Build catalog.

Tuesday, June 2, 2026

Wednesday, June 3, 2026

On-demand sessions (available now)

Click the links beneath to ticker immediately.

Microsoft Databases

Microsoft Fabric

Explore further resources for Microsoft Fabric

Read Entire Article