Microsoft Build 2026 showed us how AI has evolved from answering questions, to doing the work, to keeping the work moving.
Microsoft’s annual developer conference gave us an insight into the tech which will evolve over the next 12–24 months. With Copilot Cowork, we moved beyond simply asking Copilot to help with tasks, towards systems that can run longer, multi‑step tasks across multiple sources.
Now, AI will be able to take ownership of work, monitor what’s happening, and act on our behalf, often without needing a prompt each time.
In this article, we’ll break down the key announcements from Microsoft Build 2026 and, more importantly, what they mean for businesses.
The shift from Copilot to Autopilot
For the past couple of years, Copilot has been positioned very deliberately as just that: a co-pilot. Something that sits alongside you, helping you draft, summarise, and find information when you ask for it. The interaction model remained the same: you initiated the work, and the AI responded.
With the introduction of Autopilots, we’re starting to see the emergence of AI agents that no longer wait to be asked. These systems are designed to remain active in the background, continuously interpreting what’s happening across your work environment and taking action where appropriate.
How much of your working day do you spend prepping and organising around your ‘core’ work? Things like chasing updates, managing your calendar, coordinating between different people, creating meeting agendas, writing up minutes and making sure everyone has seen and read them… the list goes on. These are exactly the types of activities Autopilots are being positioned to take on.
What is Microsoft Scout?
Microsoft Scout is an autonomous agent that works in the background and sticks with a problem until it finds a viable route to completion of the task. It also has access to your local files and external websites, not just data on the cloud.
Microsoft Scout is the first example of an Autopilot in practice. Scoped initially as a personal assistant, Scout is designed to operate across the core tools people already use every day: Outlook, Teams, calendars, files and communications.
What does Microsoft Scout bring to the table?
Instead of completing a task and stopping, Scout is intended to maintain context over time. It observes patterns, tracks progress, and identifies where attention is needed. In practical terms, that might mean recognising that a key piece of work has stalled, noticing gaps in preparation ahead of a meeting, or surfacing risks that would otherwise only become visible later. It can then act in the background on that understanding, whether that’s scheduling prep time for meetings, prompting a decision, or preparing supporting material, without requiring a new prompt each time.
In early hands-on demonstrations, Scout has shown the ability to work through problems. For example, when asked to carry out a relatively mundane task, reviewing a local downloads folder and identifying unnecessary files, it didn’t just return a summary. It analysed the contents, identified large installer files, and then carried out the clean-up itself once permission was given. Let’s face it, no one wants to trawl through their downloads folder to free up space.
When given a more complex task it couldn’t complete directly, such as extracting content from a source it couldn’t access, it explored alternative approaches, suggested potential solutions, and was able to take further action.
If the idea of having an AI tool run off its own autonomy in the background makes you sweat, Microsoft has been very clear that these agents operate within existing permission structures, identity controls, and governance frameworks. Actions can be constrained, audited, and require approval where necessary.
AI that understands your work (Work IQ and Web IQ)
Work IQ has been underpinning Copilot for a while, allowing it to reference your emails, meetings, documents and contacts. What’s changed is how central it now is to everything else Microsoft is building.
Rather than being a background feature, Work IQ is now the context layer that powers how agents understand your work as it’s happening, not just what files you have, but who you’re working with, what you’re involved in, and how things are progressing over time. You can see this shift in the product experience already, with the removal of the old “work vs web” toggle. Instead of you choosing where Copilot looks, it’s deciding that dynamically itself.
Alongside this, Microsoft has introduced Web IQ, which is not about giving AI access to the internet (that already existed), but about fundamentally changing how that information is retrieved and used.
Instead of returning pages or search results designed for humans, Web IQ is built specifically for agents. It pulls back structured pieces of information, usable evidence, that can be fed directly into multi-step reasoning. Crucially, it’s designed for repeated retrieval, allowing an agent to check, refine and update its understanding continuously as it works through a task.
Microsoft positions Web IQ as significantly faster (around 2.5× in their comparisons) and more efficient, with a focus on higher information density per query. In simple terms, fewer tokens are needed to reach a useful answer, which improves speed and reduces cost at scale.
Taken together, these layers are what make Autopilots viable. Work IQ gives the agent an understanding of your business, Web IQ connects that to what’s happening in the wider world, and both are designed to work in real time as part of an ongoing workflow, not just a single search and response.
AI that adapts to your business (Frontier Tuning)
Alongside these broader shifts, Microsoft also introduced Frontier Tuning, which allows AI to be adapted to how your organisation actually works, rather than relying purely on general training data. Instead of repeatedly prompting Copilot to follow your processes, the AI can learn from real workflows, the steps people take, the decisions they make, and how tasks are completed within your own environment.
Importantly, this happens within your organisation’s boundaries, meaning your data and ways of working stay yours. The result is AI that produces outputs more aligned to your business, while also improving efficiency, with Microsoft’s early testing showing that tuned models can deliver comparable results at significantly lower cost.
Microsoft has built its own AI models
These sit under what Microsoft is calling the MAI model family, covering a range of core capabilities:
- MAI‑Thinking‑1 – a reasoning model designed for more complex problem-solving
- MAI‑Code‑1‑Flash – a coding model optimised for tools like GitHub Copilot and VS Code
- MAI‑Image‑2.5 – an image generation and editing model
- MAI Transcribe‑1.5 – a speech-to-text model
- MAI Voice‑2 – a text-to-speech model
- alongside more efficient “Flash” variants focused on speed and cost
Up until now, much of Microsoft’s AI capability has relied on external providers like OpenAI and Anthropic. What we’re starting to see here is a shift towards Microsoft building and owning more of that stack itself.
If you’re using Copilot, you may have already spotted the new image model in PowerPoint, and we tested this directly.
Using Copilot within PowerPoint, you can now select Microsoft’s image model and generate visuals directly into your slides. In testing, we asked it to create a concept image representing Microsoft Scout, something fairly abstract.
Rather than producing a single result, Copilot worked through multiple iterations. It recognised layout issues (in this case, generating square images for a widescreen slide), attempted to correct them, and regenerated variations to try to resolve the problem. This reflects a broader shift in how these tools are behaving, moving beyond single outputs and towards more agent-like attempts to refine and complete a task.
The end result isn’t perfect. It shows some typical AI inconsistencies, and the smiley face (which represents Microsoft Scout) is a little creepy…
If you want to experiment with these models more directly, Microsoft has also made them available through its playground environment: Try Microsoft AI Playground
While the image model is visible in tools like PowerPoint, others, such as the reasoning model, are currently positioned more behind the scenes or within developer tools.
Copilot has become multi‑model
Much of Copilot’s capability came from models provided by partners like OpenAI, and more recently Anthropic, with Claude. What’s changing now is that Microsoft is moving towards a multi-model approach, where different models can be used depending on the task.
You can already see this starting to surface in the product experience. In tools like Word, Excel and Copilot Chat, users are beginning to get options such as “quick response” or “think deeper”, as well as the ability to switch between models in some environments. Behind the scenes, this means Copilot can select (or allow you to select) the most appropriate model for what you’re trying to do.
Governance, security and cost
When you have systems that can access your emails, your files, your calendar, and actively take action on your behalf, you have to consider what information these AI agents have access to. What are they not allowed to do?
Microsoft says these agents operate within existing permission structures, meaning they can only access what the user or organisation has already allowed. They also sit behind the same identity, security and compliance frameworks that already exist within Microsoft 365, including things like sensitivity labels and data protection policies.
If your permissions are too broad, your file structure is inconsistent, or sensitive information isn’t properly labelled, then an AI system will inherit those same gaps.
The same applies when it comes to cost, because many of these newer capabilities (particularly anything involving agents, automation or even which model you choose) don’t operate as fixed features. They run on usage, often through AI credits. That includes Microsoft 365 Copilot usage, but can also extend into areas like GitHub Copilot, depending on how these tools are deployed.
As more of these systems move towards “always-on” behaviour, those costs can quickly move from occasional to ongoing.
Microsoft has acknowledged this with the introduction of a cost management dashboard within the Microsoft 365 Admin Center, designed to help organisations understand and monitor AI usage across their environment.
Conclusion
The capabilities coming out of Microsoft Build are powerful, but they’re not plug‑and‑play at scale. Without the right governance, permissions and data structure in place, tools like Copilot and Scout won’t deliver their full value – and can introduce unnecessary risk.
Strategy 365 works with organisations to make sure their Microsoft 365 environments are configured correctly, so they can adopt AI tools with confidence.
If you’re considering how these developments could work in your business, call our team on 01782 916920.