Microsoft Build 2026: Pushing The Frontier With A More Opinionated AI Playbook

Microsoft didn’t fail to bring out the big guns at this year’s Build conference. From Satya Nadella conversing with Jensen Huang (remotely) on stage to in-depth sessions around managing and securing agentic workflows, Microsoft clearly learned from past events. Gone were fictional company demos and the indecisive advice of “We’ll figure out what developers should be doing together.” In their place was a message that was more opinionated, prescriptive, and decidedly full stack. In fact, the bulk of the time was spent on the layers of that stack: infrastructure, models/content/tools, agent runtime, developer tools, and security and observability.

The Vision For AI Clearly Includes Hardware Beyond The Cloud

Surprisingly, a lot of time was spent on hardware. Microsoft showed the allure of “unmetered intelligence” with the beefy Surface RTX Spark, an AI “data center” that can sit on your desk. While it’s coming in the fall, unsurprisingly, no price was announced. Project Solara, an effort to bring agents to new hardware forms, explored the possibilities of engaging AI in new, diverse ways, like through your ID card.

The takeaway here is that while cloud is king, Microsoft recognizes that edge computing matters when it comes to AI. While it also showcased responsible data center creation and pushed the Azure Cobalt (Arm64) CPU, the overall message was that the future of AI is distributed. To address this, Forrester guides enterprises to explore small language models and evaluate local AI hardware on a per-outcome basis to address their needs.

The Context Layer Is Where Data Becomes The Differentiator

The next level up the stack is arguably the most important. The big news here was the General Availability (GA) of Fabric IQ, a combination of OneLake, a semantic model, ontologies, and data agents. This provides a rich experience for agentic AI — if you’re willing to commit to the stack. Another big announcement was the public preview of Azure HorizonDB, an “enterprise-ready” Postgres-compatible solution. Web IQ, another offering, enables agents to respond with fresh context and is model-agnostic.

The context layer is quickly becoming a high-profile battleground. While it’s nice that Foundry supports over 11,000 models, what really matters is the combination of semantics, ontology, and knowledge graphs for AI-ready data. Forrester believes that making your data AI-ready isn’t a new phenomenon — it’s the unfinished work of enterprise data management programs. Experimentation with new context layers should be key for enterprises the remainder of this year.

Contained Intelligence: Why OSes Need Forrester’s AEGIS Framework

The next level up focused on how people run agents, and one environment Microsoft really would like you to run them on is Windows. The biggest announcement here was Microsoft Execution Container (MXC), a way to contain agents so they don’t wreak havoc across the operating system. Microsoft also announced Windows-native OpenClaw, secured by process-level controls with MXC.

From our perspective, one of the key components of Forrester’s AEGIS framework is to apply Zero Trust architecture principles to agent access. Tools like OpenClaw are great but only if they don’t go deleting your files or email inbox. Fine-grained access controls at the OS level will help this happen, but it’s unclear how well users will implement these controls.

An Attempt To Make Development On Windows More Palatable

In the multiagent era, having a dozen open terminal windows quickly becomes an unwelcome reality. The big news here was the GitHub Copilot app, a way to manage the onslaught of sessions that multiagent workflows typically entail. Also welcome was the announcement of Rayfin, a quick way to deploy application back ends — a task most developers despise. Lastly, good old Windows is getting some welcome upgrades, such as a dev-focused, distraction-free environment and native Homebrew support.

While the new Windows configuration should simplify development (Microsoft, please make a distraction-free nondeveloper version!), it’s a little early to say if the GitHub Copilot app will gain the same popularity as VS Code, a tool nearly all developers use. Rayfin is a little easier to judge: It makes it easier for developers to stand up well-built back ends but requires the use of Microsoft Fabric. Still, using code-based ways to set up data, identity, and access policies should greatly simplify moving prototypes to production — a challenging task for any app developed through agentic means.

Trusting The Unattended: Why AI Governance Is Critical For Autopilots

The last major area of focus is a critical one: access controls and observability. Agent 365 has added new features and an expanded SDK. Foundry can create rubrics to examine agents for governance issues, a welcome capability. It can also improve agents by determining the best candidates to deploy based on these findings. Finally, “autopilot” agents, a new class of “enterprise claws,” have access to M365 resources but have to be approved by admins.

As we move further into the “always on” AI agent era, proper governance becomes critical. Managing a modern AI agent portfolio requires a federated approach involving enterprise architecture, platform engineering, product/business teams, security and risk, and compliance and legal. Enterprises should definitely be experimenting with solutions that provide greater observability and automatic rubric creation to help meet new governance needs.

The Waiter Finally Suggests Which Steak Knives To Use

In addition to the stack, Microsoft also provided solid insight on other areas of focus. Enterprise training (also referred to as “hill climbing”) was demonstrated and is a key component in making private models a reality. Microsoft Discovery is now in GA and enables scientists and engineers to accelerate research through AI and high-performance computing. Lastly, Nadella pulled Majorana 2 out of his pocket: a credit card-sized quantum chip with claims of qubits lasting a mean lifetime of 20 seconds (although some physicists don’t believe Majorana 1 worked to begin with). While not all of these have immediate enterprise applications, it’s clear that Microsoft is interested in demonstrating itself as a “frontier” organization that it also hopes its customers should become.

The reality is that enterprises in industries such as financial services, government, and healthcare want to be on the frontier but are held back by (very genuine) concerns about governance, regulations, and sovereignty. Being more opinionated about its full-stack offerings not only guides, but soothes Microsoft’s customers in their decision-making process.

Put another way, Microsoft has always given you 12 knives to cut a steak — at least the waiter is now recommending which four you should use.

Forrester has a full team of analysts covering the gamut of Microsoft’s offerings, both with our evaluative Forrester Wave™ and best-practice guidance. Schedule a guidance session with us if you’re a client, or leverage Forrester AI for instant insights on Microsoft.