23–25 May 2023, Seattle Convention Center (in person, with a free livestream alongside). This was the one. If you ask us to point at the single Microsoft event where "AI" stopped being a side track and became the spine of the entire developer story, it's this one. Six months earlier ChatGPT had landed on the world like a meteor; Microsoft had wired GPT-4 into Bing in February; and now, on a grey Seattle morning in late May, the company walked on stage and reorganised its whole platform around one idea. Not "we added some AI." Rather: here is how you build a copilot of your own.
We were in the room, as we have been every year since 2012. Here is what actually happened, what was real versus what was a slide, and why it changed the conversations we have with our Mittelstand customers back in Ibbenbüren.
"Build your own copilot"
The framing came straight from Satya Nadella's opening keynote, and it was sharper than the usual conference throat-clearing. The pitch wasn't "look at our cool AI features." It was: we built all of these copilots — GitHub Copilot, Copilot in Microsoft 365, in Power Platform, in Dynamics 365, Security Copilot, Bing — on one common architecture, and now we're handing that architecture to you.
"After all, we built all these copilots with one common architectural stack. We want to make that available so that everyone here can build their own copilot for their applications. We will have everything from the AI infrastructure to the foundation models to the AI orchestration, all the way up to your copilot." — Satya Nadella, Build 2023 keynote
That is the whole conference in one paragraph. Microsoft named this the Copilot stack: AI infrastructure at the bottom (the GPUs and the Azure OpenAI Service), foundation models in the middle, an orchestration layer above that (prompts, plugins, grounding on your data), and your application on top. For a shop like ours, the important word in there is orchestration — because that's the layer where a demo turns into a product that has to behave in front of a paying customer.
Microsoft CTO Kevin Scott put the developer ambition more bluntly, and the line stuck with us:
"You may look at Bing Chat and think this is some super magical complicated thing, but Microsoft is giving developers everything they need to get started to go build a copilot of their own. I think over the coming years, this will become an expectation for how all software works." — Kevin Scott, Microsoft CTO, Build 2023
"An expectation for how all software works." Three years on, writing this from 2026, that prediction reads as merely accurate rather than bold — which is exactly the point. In May 2023 it was a genuine bet.
Plugins, and a rare moment of standards sanity
The announcement we expected to be quietly the most consequential — and we think it was — is that Microsoft adopted the same open plugin standard OpenAI had introduced for ChatGPT. One plugin definition, written once, would work across ChatGPT, Bing, Dynamics 365 Copilot, Microsoft 365 Copilot, and Windows Copilot.
In an industry that loves to ship five incompatible versions of the same idea, two of the biggest players agreeing on a single plugin contract was genuinely surprising, and genuinely useful. Kevin Scott's mental model for what a plugin is was the best one-liner of the week:
"A plugin is about how you, the copilot developer, give your copilot or an AI system the ability to have capabilities that it's not manifesting right now and to connect it to data and connect it to systems that you're building." — Kevin Scott, Build 2023
In plainer terms: the model is the reasoning engine, and the plugin is the actuator — the thing that lets it actually do something against your data, your line-of-business system, your API. That separation is what made "build a copilot" sound like engineering rather than magic. With hindsight, you can draw a straight line from this shared-plugin moment to the open agent/tool protocols that dominate the conversation in 2026. The detail nobody could have known at the time: this particular plugin format itself didn't end up as the long-term standard. The principle — one open, portable contract between models and tools — absolutely did.
Microsoft Fabric: the biggest data launch in years
If Copilot was the headline, Microsoft Fabric was the announcement with the most immediate weight for our customers' data platforms. Fabric pulled Data Factory, Synapse (data engineering, warehousing, data science, real-time analytics) and Power BI into a single SaaS product, all sitting on one storage layer called OneLake — a "OneDrive for data," in Microsoft's framing — with Copilot wired in at each layer.
Nadella called it the biggest data product launch from Microsoft since SQL Server. That's a marketing line, and we'd treat the superlative with the usual pinch of salt, but the underlying architectural move was real: one lake, one copy of the data, every analytics engine reading from the same place instead of each tool keeping its own duplicate. For German mid-market companies who'd spent a decade building a tangle of point-to-point data pipelines, "stop copying your data five times" is a pitch that lands.
Azure AI Studio and "chat with your own data"
The piece we knew we'd use first was Azure AI Studio, introduced inside the Azure OpenAI Service. The promise was deceptively simple: ground GPT-4 on your private data — point it at your documents, your knowledge base — and get a "chat with your own data" experience without hand-rolling an entire retrieval pipeline. Alongside it came Azure Machine Learning prompt flow for building and evaluating prompt chains, and vector search arriving in Azure Cognitive Search (then in preview) so concept-based retrieval was a first-class option rather than a bolt-on.
This mattered to us for one boring, decisive reason: it could stay in your own Azure tenant. The first question every German mid-market customer asks about AI isn't "how clever is it" — it's "where does my data go." Being able to answer "it stays in your tenant, under your compliance boundary" is what moved these conversations from interesting to actionable.
Microsoft also flagged the guardrails, not just the engine. Azure AI Content Safety went to preview as a moderation layer for text and images. Sarah Bird, leading responsible-AI product work, made the point that this wasn't theoretical plumbing:
"It's the safety system powering GitHub Copilot, it's part of the safety system that's powering the new Bing." — Sarah Bird, Microsoft, Build 2023
We appreciated that the safety story shipped with the capability story rather than as an apologetic footnote a year later.
Windows Copilot, Dev Home, and a faster disk
On the Windows side, Windows Copilot was the crowd-pleaser: a system-level AI sidebar built into Windows 11, able to take actions across the OS and apps — change a setting, summarise a document, kick off a task — rather than just answer questions in a box. Preview was promised for June 2023. The idea of a first-party, OS-level copilot was a real first, and the demos were slick. (Honesty compels a 2026 footnote: the Windows Copilot of 2023 was a long way from finished, and Microsoft has reshaped it several times since. The direction was right; the first cut was early.)
For developers specifically, the more durable Windows news was the unglamorous stuff:
- Dev Home — a new, open-source Windows 11 control centre for setting up and managing a dev environment, hooking into GitHub and cloud dev boxes, with WinGet Configuration files so a whole machine setup becomes reproducible from a single file. (We've used machine-config-as-code for years; seeing it become first-party on Windows was overdue and welcome.)
- Dev Drive — a storage volume tuned for developer workloads, built on ReFS and paired with a Microsoft Defender "performance mode," claiming up to ~30% better build times versus stock Windows 11 22H2. A faster repo disk is the kind of small thing that compounds across a team every single day.
- Hybrid Loop — a pattern for splitting AI inference between Azure and the local device, built on ONNX Runtime with the Olive toolchain for model optimisation, across CPU, GPU and NPU. This was the early signal that "on-device AI" was going to be a serious axis, not just a cloud-only story.
- Dev Box — general availability slated for July 2023, with config-as-code customisation and starter images, for teams who want a managed cloud workstation.
.NET 8 and Semantic Kernel: the bits we'd actually type
Underneath the AI spectacle, the runtime we live in kept moving. .NET 8 was deep in its preview cycle around Build, carrying the work that mattered to us day to day: Native AOT maturing, ongoing performance gains, and Blazor's full-stack story coming together — the foundations that, a release or two later, would grow into things like .NET Aspire. .NET 8 itself shipped that November as an LTS release; Build was where the shape of it came into focus.
The other developer-facing piece worth naming is Semantic Kernel, Microsoft's open-source SDK (C# and Python) for orchestrating prompts, plugins and memory in your own apps. For a .NET-first shop, this was the practical on-ramp to the Copilot stack: a way to build the orchestration layer in the language we already ship in, rather than detouring into a separate Python-only world.
Why it mattered for us
We came home from Seattle with a changed sales pitch, and we mean that literally.
Before Build 2023, talking to a German mid-market manufacturer or distributor about "AI" meant managing expectations downward — yes, it's interesting; no, you probably can't build on it responsibly yet; no, your data shouldn't go where you'd have to send it. After Build 2023, the honest answer flipped. We started telling those same customers, in plain German: the stack is finally something you can build on — and it can stay in your tenant. That second clause did most of the work. Azure AI Studio plus "on your data" plus the tenant boundary turned a hand-wavy demo into a project we could scope, price, and stand behind.
Microsoft Fabric changed our data-platform conversations in a different way. Instead of pitching yet another bespoke pipeline, we could sketch a single lake with analytics and Copilot on top — fewer moving parts to break, fewer copies of the data to reconcile, less of the integration tax that quietly eats mid-market IT budgets.
We'll also be honest about the parts we were wary of, because warmth without candour is just marketing. The "build a copilot in an afternoon" energy on stage undersold how much careful engineering — grounding, evaluation, cost control, guardrails — separates a keynote demo from something you'd let a real user touch. Plenty of the announcements were preview, not shipping, and a few were closer to direction than product. We told customers the truth: the platform shift is real, the foundations are solid, and you should start — but budget for the unglamorous orchestration work, because that's where the project actually lives.
Three years on, almost everything from Build 2023 either shipped, matured, or evolved into something bigger. The framing held up best of all. "Build your own copilot" was the right sentence at the right moment — and for once, the keynote and the next three years agreed with each other.
Sources & further reading
- The Official Microsoft Blog — Microsoft Build brings AI tools to the forefront for developers (23 May 2023): https://blogs.microsoft.com/blog/2023/05/23/microsoft-build-brings-ai-tools-to-the-forefront-for-developers/
- Microsoft Source — Microsoft outlines framework for building AI apps and copilots; expands AI plugin ecosystem: https://news.microsoft.com/source/features/ai/microsoft-outlines-framework-for-building-ai-apps-and-copilots-expands-ai-plugin-ecosystem/
- Microsoft — Build 2023 Book of News: https://news.microsoft.com/build-2023-book-of-news/
- Windows Developer Blog — Bringing the power of AI to Windows 11 — Windows Copilot and Dev Home: https://blogs.windows.com/windowsdeveloper/2023/05/23/bringing-the-power-of-ai-to-windows-11-unlocking-a-new-era-of-productivity-for-customers-and-developers-with-windows-copilot-and-dev-home/
- Windows Developer Blog — Dev Drive: Performance, Security and Control for Developers: https://blogs.windows.com/windowsdeveloper/2023/06/01/dev-drive-performance-security-and-control-for-developers/
- InfoQ — Microsoft Build 2023: Bing AI and Copilot Plugins, Microsoft Fabric and More: https://www.infoq.com/news/2023/06/microsoft-build-2023/
- Satya Nadella, Build 2023 keynote (official transcript, PDF): https://msftstories.thesourcemediaassets.com/2023/05/KEY01-Build-Satya-Nadella-Keynote.pdf
Where a claim rests on Microsoft's own framing — the "biggest data launch since SQL Server" line, preview-stage features, and forward-looking ship dates — we've flagged it as such. Quotes are reproduced verbatim with attribution from the sources above.
Image credits
All photos are used under their respective Creative Commons licences; we are grateful to the photographers.
- Seattle from Space Needle June 2018 001 — © King of Hearts, CC BY-SA 4.0, via Wikimedia Commons (source).
- San Antonio data center — © Robert Scoble, CC BY 2.0, via Flickr (source).
- PROJECT BOOT Dev Mode codes — © Deliveramable, CC BY-SA 4.0, via Wikimedia Commons (source).