Microsoft Construct 2026 highlights developments in app improvement with Microsoft Cloth and Microsoft Databases, emphasizing a unified information and AI platform for scalable, agentic functions.
On this article
AI is driving a basic shift in how work will get accomplished and the way functions are constructed. Because the 2026 Microsoft Work Pattern Index Report highlights, a rising share of employees are transferring past asking inquiries to handing off total duties and orchestrating multi-agent programs.
This shift introduces a brand new constraint. The problem is not mannequin functionality, however constant, shared information context throughout the enterprise.
Builders and enterprise customers know what they wish to automate, and at this time’s fashions can ship. The bottleneck is context. Each new agent begins from zero, relearning how the enterprise works, the place information lives, and what guidelines to comply with. And not using a constant basis, brokers can’t coordinate or scale.
That’s the problem we’re fixing with Microsoft Cloth. It supplies a unified information and AI platform that empowers you to convey collectively information and transfer from remoted AI experiments to production-ready agent programs, by which every new agent builds on shared organizational context. This imaginative and prescient is already driving robust momentum among the many thousands and thousands of builders constructing on Cloth and Microsoft Databases.
At Microsoft Construct, we’re extending this basis with new capabilities that assist builders transfer from prototype to manufacturing sooner. These embrace Rayfin, a brand new software program improvement package (SDK) and command-line interface (CLI) designed to make Cloth a production-ready utility backend, and Azure HorizonDB, a brand new PostgreSQL database designed for AI‑powered functions, now in public preview.
Introducing Rayfin: From immediate to manufacturing backend
Coding brokers are accelerating app improvement. Shifting these functions from prototype to manufacturing, nevertheless, stays a problem. Agent-created or not, each production-ready utility nonetheless depends on a backend to handle information, implement id and permissions, coordinate state, and function reliably over time. Present software-service platforms have been both not designed for brokers or don’t totally meet enterprise necessities for deployment, safety, and governance.
Rayfin, a brand new open-source SDK and CLI, is designed to shut that hole. It lets builders and coding brokers describe what to construct and get an enterprise-grade utility backend immediately into the applying code, together with a database, authentication, and extra. Rayfin then deploys on to Microsoft Cloth, giving each utility enterprise-grade safety and scale from day one. Builders and AI brokers can now transfer from immediate to manufacturing with out managing infrastructure.
With Rayfin, builders work by acquainted GitHub‑primarily based workflows to outline information fashions, backend logic, and entry insurance policies solely in code, giving groups and brokers a constant, programmable interface for constructing and managing functions. Watch Rayfin in motion:
As a result of Rayfin might be deployed immediately on Cloth, utility information lands immediately in OneLake, the place it’s instantly out there to the total Cloth information stack, unified with analytics, operational and real-time information, and AI engines by default. This allows builders to construct their enterprise apps on trusted enterprise logic, combine with semantic fashions, and embed wealthy information visuals.
We’re excited to accomplice with Replit, a number one AI coding platform, to assist prospects construct enterprise-grade apps within the interface they know and love whereas conserving app, information, and providers managed in their very own Cloth tenant.
Rayfin unlocks a brand new improvement mannequin for our customers. Brokers write the code. Cloth ships it rapidly and safely. Collectively, we’re giving builders one thing they’ve by no means had earlier than: a path from thought to enterprise-grade manufacturing that’s measured in hours, not months.
—Amjad Masad, Chief Govt Officer (CEO) of Replit
Be taught extra about Rayfin by watching the Microsoft Construct session “BRK225 – Knowledge, apps, and brokers: the way forward for app dev with Microsoft Cloth” on Wednesday, June 3, 2026, at 1:30 PM PT.
Microsoft Databases, designed for AI functions
For many years, databases have been the spine of enterprise functions. As functions grow to be extra clever and agent‑powered, we’re evolving Microsoft Databases into the muse optimized for actual‑time, AI‑prepared, and operationally wealthy experiences.
Azure HorizonDB: Enterprise-ready PostgreSQL constructed for the calls for of AI functions
As a number one PostgreSQL committer, Microsoft has lengthy invested within the PostgreSQL group. However as AI‑powered functions place new strains on scale, latency, and resilience, the demand for a brand new class of PostgreSQL database is obvious.
Azure HorizonDB is that subsequent step. Now out there in public preview, HorizonDB is a totally managed, PostgreSQL‑appropriate database that mixes PostgreSQL familiarity with cloud‑scale structure. It’s zone resilient by default and delivers elastic storage that scales to 128 TB, large scale‑out compute as much as 3,072 vCores, and might maintain sub‑millisecond, multi‑zone commit latency for demanding transactional situations.
As our information calls for have expanded exponentially due to our use of Azure AI to speak with our information, HorizonDB has come on the good time to satisfy the efficiency, scale, and safety we have to shift into this new world of AI-enabled information.
—Rand Morimoto, President of Convergent Computing (CCO)
Past scalability, we’ve additionally infused Azure HorizonDB with experiences designed particularly for AI functions like vector search, built-in AI mannequin administration, and direct connectivity to Microsoft Foundry and Cloth. These options present a contemporary basis for constructing, modernizing, and scaling AI‑powered functions with confidence. Mohsin Shafqat, Director of Software program Engineering at NASDAQ, talked about, “What stood out with HorizonDB is that it aligns carefully with how we already take into consideration the issue. As a substitute of sewing collectively a number of elements, it brings transactional information, vector search, and AI capabilities right into a single platform, which simplifies the structure with out forcing an entire rethink.”
Be taught extra by watching the Microsoft Construct session “BRK223 – From rows to reasoning: Designing databases for AI apps and brokers” on Tuesday, June 2, 2026, at 2:30 PM PT.
New safety and migration tooling for Azure Database for PostgreSQL
As Azure HorizonDB powers a brand new class of high-scale, mission‑important functions constructed for AI, Azure Database for PostgreSQL stays a trusted, open‑supply basis for modernizing and working current PostgreSQL workloads.
At this time’s updates introduce two significant enhancements. First is the Microsoft Defender for Cloud integration, now in preview, which delivers steady safety and compliance assessments to assist groups determine misconfigurations and scale back threat. Second, we’ve launched new discovery and evaluation tooling that will help you extra confidently plan migrations to Azure Database for PostgreSQL. These instruments consider Oracle and PostgreSQL environments and supply readiness insights, sizing steerage, and price estimates. Be taught extra about this migration tooling within the Microsoft Construct on-demand session “OD822 – Smarter PostgreSQL migrations to energy trendy, clever apps.”
Powering clever, multi‑agent programs at international scale with Azure Cosmos DB
Azure Cosmos DB is Microsoft’s AI-ready NoSQL and vector database for constructing responsive functions and clever AI experiences at any scale. OpenAI, for instance, selected Azure Cosmos DB as its major operational database “due to its computerized scaling and schema-less flexibility, permitting us to iterate rapidly,” mentioned Nick Cooper, senior technical employees member at OpenAI.
At Microsoft Construct, we’re centered on enhancing developer productiveness and AI high quality. The Azure Cosmos DB Linux Emulator is now usually out there, enabling builders to construct, take a look at, and validate functions domestically throughout Linux, macOS, and Home windows with out a cloud dependency. New AI capabilities are additionally now in preview, together with semantic reranking, which improves search relevance utilizing constructed‑in contextual understanding. As well as, a brand new agent reminiscence toolkit helps builders standardize persistent reminiscence for AI brokers utilizing Azure Cosmos DB, Azure Sturdy Features, and Microsoft Foundry fashions. Be taught extra within the Microsoft Construct on‑demand session “OD820 – Designing dependable multi-agent apps with Azure Cosmos DB.”
Unifying databases and Cloth on a single platform
Microsoft Databases might be centrally managed by the brand new Database Hub in Cloth, at present in personal preview, and mirrored into OneLake, bringing operational and analytical information onto a single basis. From there, you should use Cloth to make it trusted, contextual, and prepared for AI.
Constructing an AI‑prepared information basis with Microsoft Cloth
Within the period of AI, information is the gasoline, however information alone shouldn’t be sufficient. Equally vital is how that information is known: the definitions of shoppers, orders, merchandise, income, and the relationships between them. At this time, that understanding is fragmented throughout buyer relationship administration (CRM) and enterprise useful resource planning (ERP) programs, productiveness instruments, and spreadsheets, and too typically it doesn’t journey with the information. Organizations have lengthy relied on individuals to recreate this context. However as brokers tackle extra duty, this hole turns into important. And not using a shared understanding of the enterprise, brokers can’t reliably cause, coordinate, or act.
Microsoft IQ addresses this lacking layer by unifying enterprise intelligence right into a shared basis constructed to activate AI brokers. It allows constant reasoning and enterprise-scale affect, reasonably than remoted interactions and and brings collectively 4 interconnected capabilities: Work IQ captures how work occurs, Cloth IQ fashions how the enterprise operates, Foundry IQ allows brokers to find and reuse data, and the brand new Internet IQ, introduced at this time at Microsoft Construct, provides real-time international context from the online.
Cloth IQ is central to this technique. It powers a steady operational loop the place individuals and brokers observe dwell alerts, cause over shared context, and take ruled motion within the second throughout analytics, operations, and the productiveness instruments the place work occurs. It supplies three built-in layers of enterprise context:
Unified information: OneLake unifies the group’s information property, spanning analytical and operational information right into a single, accessible layer.
Enterprise intelligence: Semantic fashions present structured, ruled representations of that information which organizations already depend on for trusted enterprise metrics and analyze their enterprise.
Operational intelligence: Ontologies seize operational context by defining enterprise entities and their relationships so brokers can cause within the language of the enterprise. This context can embrace dwell alerts from Cloth Actual-Time Intelligence, enabling organizations and their brokers to grasp what is going on proper now and act in time to vary outcomes.
This three-tiered basis helps be sure that each agent begins with the identical understanding of the enterprise and might apply it accurately throughout workflows. However Frontier organizations can’t begin on the IQ layer. Constructing this functionality requires a unified information basis. Microsoft Cloth delivers this by 4 core capabilities:
Unifying your information property
Processing and harmonizing information
Curating semantic that means
Empowering AI brokers to behave
Be taught extra about how Cloth may also help you create an AI-ready information basis within the Microsoft Construct on-demand session “OD811 – Powering the subsequent AI frontier with a unified information platform.”
1. Unifying your information property with Microsoft OneLake
Most organizations battle to see their total information property. It’s unfold throughout programs, duplicated in a number of locations, and owned by totally different groups, making it troublesome to know what exists, the place it lives, and the way it connects. Microsoft OneLake brings it collectively right into a single, AI‑prepared information lake that unifies your multi‑cloud property and allows group‑large entry for analytics and AI.
We’re making it simpler to attach current information to OneLake with out transferring or duplicating it with the discharge of shortcuts to SharePoint and OneDrive into normal availability and the flexibility to create shortcuts immediately from Cloth Knowledge Warehouses, now in preview. We’re additionally including the preview of workspace-level Azure Personal Hyperlink help for mirrored information sources.
The final availability of the OneLake catalog in Microsoft Foundry
On the identical time, we’re making it simpler to attach that information to AI. With the latest normal availability of the OneLake catalog in Microsoft Foundry, you possibly can uncover trusted information, discover wealthy metadata, and join it on to AI options. The catalog is embedded inside Foundry’s Data expertise, making it easy to maneuver from information discovery to AI improvement in a single workflow.
Be taught extra about all of those OneLake bulletins by watching the Microsoft Construct on-demand session “OD815 – Unify your total information property on a single, AI-ready information lake.”
2. Course of information sooner with a brand new class of GPU-accelerated analytics
As soon as information is unified, the subsequent problem is popping it into insights rapidly, reliably, and at scale. We’re introducing GPU-acceleration constructed immediately into Cloth Knowledge Warehouse to unlock a brand new stage of efficiency with out including complexity.
The analysis behind this innovation was lately acknowledged by ACM SIGMOD because the “Finest Trade Paper of 2026.” This breakthrough establishes Cloth Knowledge Warehouse as the primary totally managed information warehouse to supply GPU acceleration.
By integrating NVIDIA accelerated computing, question acceleration in Cloth Knowledge Warehouse basically modifications how briskly queries can run. In inside benchmarking carried out in Might 2026, the GPU-accelerated Cloth Knowledge Warehouse delivered as much as 7x sooner efficiency relative to a few comparable exterior distributors for reporting and utility workloads at 64-user concurrency.
This shift displays a broader change in how trendy information programs have to function. As Ian Buck, Vice President of Hyperscale and HPC at NVIDIA, explains, “AI functions are redefining how an information warehouse must carry out. As AI brokers cause over enterprise information, analytics programs want low-latency efficiency for a lot of simultaneous customers. With NVIDIA accelerated computing and customized CUDA kernels constructed immediately into Microsoft Cloth Knowledge Warehouse, Microsoft is bringing the SQL workflows prospects already use into the manufacturing AI period.”
Clients are already seeing that affect. At UNC Well being, “We’re seeing as much as 5x enchancment in our question speeds, which permits our groups to spend much less time managing efficiency and extra time delivering significant insights,” mentioned Shaun McDonald, IT Supervisor.
Question acceleration works mechanically inside Cloth, rushing up queries with out requiring any question rewrites. The result’s constantly low‑latency analytics that energy responsive functions, interactive reporting, and agent‑pushed evaluation, delivering brisker insights and larger confidence in information.
Question acceleration might be out there for an early entry preview within the subsequent few weeks. You possibly can be taught extra by watching the Microsoft Construct on-demand session “OD813 – Powering trendy information analytics in Cloth Knowledge Warehouse.”
3. Curating semantic that means with Cloth IQ, now usually out there
As soon as information is ready, the subsequent problem shouldn’t be entry however understanding. Most organizations lack a shared layer of enterprise context, forcing each agent to relearn how the enterprise works from fragmented information. Cloth IQ, now usually out there, addresses this hole.
With information unified in OneLake, Energy BI’s industry-leading semantic fashions then present structured representations of your information for trusted enterprise intelligence, serving as a great basis for coaching brokers.
Ontologies in Cloth IQ, anticipated to be usually out there within the coming months, lengthen semantic fashions by including operational context. They outline enterprise entities, relationships, properties, guidelines, and actions, and connect with dwell alerts from Cloth Actual-Time Intelligence. Operations brokers, now usually out there, then cause over shared dwell context, make selections primarily based on coverage, and take motion within the second. Working on the ruled basis of Cloth and built-in with Microsoft Foundry, operations brokers transfer past answering inquiries to driving motion and outcomes.
Asserting the overall availability of graph and planning in Cloth
We’re saying the overall availability of graph in Cloth, with normal availability of the planning in Cloth coming later this month. Out there at this time, graph introduces a extremely scalable, relationship‑first mannequin that connects enterprise entities, programs, and alerts so groups and brokers can perceive how modifications propagate throughout the enterprise and act with full context. Planning extends this basis by enabling organizations to create plans, budgets, forecasts, and state of affairs fashions on prime of Cloth’s semantic fashions. Notably, planning in Cloth is not only static outputs. They are often written again into Cloth to drive execution, enabling closed-loop alignment with the identical system of knowledge and context.
Extending Cloth IQ to Microsoft Foundry and Agent 365
At this time, we’re extending Cloth IQ throughout the agent ecosystem so this shared understanding can be utilized constantly throughout each agent and utility.
Microsoft Foundry and Agent 365
Now in preview, Ontologies are accessible immediately from Microsoft Foundry as data sources, bringing trusted enterprise context into each customized and built-in agent experiences. Additionally in preview, Cloth IQ is now built-in with Microsoft Agent 365 as a first-party mannequin context protocol (MCP) software, enabling organizations to floor brokers in shared that means and guarantee constant habits throughout their agent property.
Microsoft 365 Copilot: Cowork and Copilot Chat
Cloth IQ can be extending into Microsoft 365 Copilot, together with Cowork and Copilot Chat. This allows brokers to entry ruled Cloth information, beginning with Energy BI studies and semantic fashions, to show insights into motion. As a substitute of static dashboards, brokers can detect modifications in key metrics, generate updates, set off follow-ups, and schedule subsequent steps, all grounded in trusted, ruled information. The result’s sooner, extra constant execution throughout the group. These experiences are at present out there prospects in Frontier with a Microsoft 365 Copilot license. Be a part of this system at this time.
GitHub Copilot CLI
Utilizing Agent Expertise for Cloth, Cloth IQ instruments and abilities for information insights are accessible by GitHub Copilot CLI, bringing ruled semantic context immediately into the terminal. Now you too can question Energy BI studies and semantic fashions from the command line, grounded in ruled semantic context in Cloth IQ. Groups can ask pure language questions on utilization, metrics, or buyer habits and get solutions grounded in Cloth information immediately inside their workflow, decreasing back-and-forth and accelerating data-driven selections.
Be taught extra about these enhancements within the Construct on‑demand session “OD812 – Bringing Enterprise Ontology Instantly into the Developer Workflow.”
4. Empowering brokers to behave with operations brokers and Copilot in Cloth
Enterprises are more and more transferring to multi-agent programs made up of specialised brokers grounded in particular information or area experience. These brokers might be reused throughout a number of programs, making it simpler to scale and ship extra constant outcomes.
Asserting the overall availability of operations brokers
Microsoft Cloth helps this shift with native agent capabilities, together with Cloth information brokers and operations brokers. I’m excited to share operations brokers at the moment are usually out there. These brokers are designed to constantly monitor real-time information, detect patterns or anomalies, and act primarily based on predefined enterprise logic.
Apply agentic analytics in Energy BI
As well as, we lately launched open-source Agent Expertise for Cloth, designed to make it simpler for AI instruments and brokers to work together immediately with Cloth. These capabilities now lengthen to Energy BI, enabling builders and analysts to create studies utilizing pure language or perhaps a screenshot of the specified consequence, considerably accelerating how insights are constructed and shared.
Copilot in Energy BI can now additionally modify semantic fashions, with built-in suggestions that enhance efficiency and make fashions extra AI-ready. This helps groups iterate sooner and ship insights extra rapidly with out leaving the instruments they already use.
Be taught extra by studying the Energy BI weblog or by watching the Microsoft Construct on-demand session “OD817 – Agentic analytics with Energy BI and Microsoft Cloth.”
Watch these bulletins from Microsoft Construct
In the event you’re thinking about studying extra and seeing dwell demos on these bulletins, I encourage you to look at the Microsoft Database and Microsoft Cloth classes at Microsoft Construct, out there within the Microsoft Construct catalog.
Tuesday, June 2, 2026
Wednesday, June 3, 2026
On-demand classes (out there now)
Click on the hyperlinks under to look at instantly.
Microsoft Databases
Microsoft Cloth
