Some more detail on the ROIC Intelligence App I built yesterday and mentioned on today’s earnings call.

Microsoft CEO Satya Nadella built a ROIC Intelligence App using Copilot code and autopilot in Microsoft's new superapp, turning a Morgan Stanley research PDF into a governed enterprise application. The app, which includes history, lookups, scenarios, and what-ifs, is fully integrated with GitHub Enterprise, Fabric, and Agent 365, with all artifacts versioned and controlled. Nadella emphasized that the AI was the builder, not the building, and that the app's architecture ensures every figure is labeled by source and unverifiable numbers are set aside, highlighting a shift toward governed, reusable enterprise AI.

Some more detail on the ROIC Intelligence App I built yesterday and mentioned on today's earnings call. I took the PDF that Brian Nowak https://www.linkedin.com/in/brian-nowak-17246b2b?trk=public post-text at Morgan Stanley put together for Hyperscale ROIC this week and used Copilot code coming in our new superapp with a single prompt + skill /drill-me to create the plan, then used autopilot in auto to create the full app with history, lookups, scenarios, what-ifs, etc . And /rubber-duck to test. And the best part is that all the artifacts are in my enterprise environment. My app is in Copilot, my code is in GitHub Enterprise; all my data pipelines/lake/semantic models are in Fabric. And everything is under Agent 365 IT/Sec/FinOps control So this is not about Tokenmaxxing or vibe coding. Every step of the way the rails are engineered to create value, making everything a long-term reusable asset, with governance/security, and cost controls. This is the full system to drive business value. Disclosures: This is all pulled from public sources, and for illustrative purposes only...not financial advice : Here is the app and architecture... This illustrates where enterprise AI adoption is heading: governed, reusable, and integrated into existing security frameworks rather than isolated experiments. The discipline behind embedding controls and cost accountability is what separates sustainable innovation from novelty, a principle equally critical in scaling preventative health technology responsibly. The most interesting thing in this post is what's missing from the diagram: the AI. Satya Nadella https://www.linkedin.com/in/satyanadella?trk=public post comment-text built the app with AI. Then he drew a picture of what he'd built, and the AI isn't in the picture. That's not an oversight. The AI was the builder, not the building — it wrote the thing and stepped out. What runs afterwards is ordinary, checkable plumbing. What IS in the picture is the part I'd care about after 25 years of reviewing financial models. Every figure carries a label for where it came from — filed, calculated, research estimate, illustration — and those labels are never allowed to blend. And twelve numbers the system couldn't verify were set aside instead of filled in. That second one is rare. In real life the row nobody can tie out gets a prior-year average at 6pm so the file will run, and six months later nobody remembers it was a guess. That row is where models die in front of a board. So the question for a finance leader isn't which AI you use. It's whether you could still explain your numbers if you changed it. Satya Nadella https://www.linkedin.com/in/satyanadella?trk=public post comment-text What I find most interesting isn't that Copilot turned a Morgan Stanley research report into an application—it's that every intermediate artifact became a governed enterprise asset. The prompt became a plan, the plan became code, the code became Fabric pipelines, OneLake assets, semantic models, and a Copilot experience—all versioned in GitHub Enterprise and governed through Agent 365. Nothing is ephemeral; every step compounds into reusable enterprise capability. That's an architectural shift from AI generating software to engineering enterprise intelligence systems. For me the app isn't even the interesting part. What gets me is that our CEO actually sat down and built this himself, hands-on-keyboard, full-stack, from scratch. Not "throwaway vibe code". It runs on engineered rails, with governance, security and cost control, so what was built actually sticks and gets reused. I think this is a skill every business leader needs to grow into. If you want to stay relevant and lead with credibility from here on, you have to be able to build, not just talk about it. "This is not about Tokenmaxxing or vibe coding. Every step of the way the rails are engineered" — that sentence is doing more work than the demo. Anyone can now generate a working app from a PDF in an afternoon. The part almost nobody solves is what happens next: where does the code live, who can audit it, what governs the data it touched, and can a colleague maintain it in six months. In your version, the answers exist before the app does — GitHub Enterprise, Fabric, Agent 365. That's the actual product; the app is the demonstration. We run AI-assisted systems in production at a tiny fraction of this scale, and the lesson is identical: generation was never the bottleneck. Custody was. The reason most AI prototypes never reach customers isn't that the code doesn't work — it's that nobody can safely own it afterwards. The interesting shift isn't that a CEO can build an app in a day. It's that the artefacts land somewhere accountable when he does. 🛠️ This is the most Microsoft possible way to announce, “I vibe-coded a finance demo,” while insisting it was not vibe coding. The funniest sentence is “the rails are engineered.” Satya began with a single prompt and ended with a full application yesterday. That is vibe coding with procurement authority, identity management, and a much larger invoice. What independent benchmark, source-reconciliation test, model-risk review, error budget, and non-Microsoft exit path would distinguish this from a polished earnings-call demo assembled inside the world’s most expensive corporate terrarium? Harry Ghuman https://www.linkedin.com/in/harryghuman?trk=public post comment actor-name 6d Great example, Satya. What strikes me is how quickly the constraint moves once application creation becomes dramatically easier. The harder problem becomes the enterprise context around it—trusted data, semantic models, scenarios, governance, security, economics, and ultimately knowing which decisions are worth improving. AI may make applications abundant. Differentiated decision capability is a different challenge. The IT/Sec/FinOps wrapper is the easy 80%. Harder question is who signs off on the model's assumptions before a Copilot generated ROIC number touches a capital allocation decision Kaan Can G. https://ae.linkedin.com/in/kaancanguven?trk=public post comment actor-name 6d The artifacts line is the whole post and it will get read as a footnote. App in Copilot, code in GitHub Enterprise, pipelines in Fabric, all under Agent 365 governance. That is the difference between a demo and something that still exists in six months. Most AI builds die in exactly that gap. Someone produces something genuinely impressive over a weekend, then nobody can find the code, nobody knows what data it touched, and there is no owner when it breaks. So it gets rebuilt from scratch next quarter and counted as innovation again. The /rubber-duck detail is the part I would steal. Testing as a named step inside the same flow, rather than something you get to after the exciting part is done. Genuine question on the governance side: when an agent in that chain makes a call that turns out wrong, does Agent 365 tell you which agent acted and under whose authority? Producing the artifact is solved. Attributing the decision is the part I still see teams improvising. Donna Box https://www.linkedin.com/in/donna-box-1a27b?trk=public post comment actor-name 4d I use the /grill-me skill all of the time, and it had a huge impact on my work. I've never heard of /drill-me. Any pointers? See more comments https://www.linkedin.com/signup/cold-join?session redirect=https%3A%2F%2Fwww%2Elinkedin%2Ecom%2Fposts%2Fsatyanadella some-more-detail-on-the-roic-intelligence-activity-7488406960443183104-g52O&trk=public post see-more-comments