The throughline from Build Day 2: @Microsoft is assembling a broader and more enterprise-centric AI stack than I have seen to date.
On hardware, the Surface Laptop Ultra with an onboard NVIDIA RTX GPU demoed well. Where it lands competitively is an open question, but the signal is clear—Microsoft wants first-party credibility for AI-heavy local workloads.
The client edge is the more unique story compared to its peers. The new developer-focused Windows configurations (some GA now, the rest arriving in the fall), combined with security features like MXC, finally give IT a defensible case for standardizing the client-edge footprint on Windows/M365. Prying Macs out of developer and creative teams remains unlikely near-term, but the standardization argument is materially stronger than it was.
On models, MAI paired with Frontier Tuning is, to me, the most interesting enterprise thread—and the part I dug into with Don Scott and Dave Citron. Microsoft reinforced a thesis I've held: with open-weight models, a smaller fine-tuned model can outperform a much larger general-purpose one on latency, accuracy, and cost. In the right conditions, the ROI on an organization-specific model is real. The gating question is execution—how well you can align your context/data estate against the total cost of building and maintaining a bespoke model.
Finally, Foundry (with Foundry IQ) as an agent control plane continues to look credible as every vendor drifts toward a more stack-centric posture.
More in my forthcoming stack research at @MoorInsStrat Insights & Strategy.
Meet Microsoft Scout.
An always-on agent that keeps work moving, taking action without needing to be prompted each time.
As Microsoft’s first Autopilot agent, Microsoft Scout works across Teams, Outlook, OneDrive, and more—taking action within the controls your organization sets.
Learn more: https://t.co/n8xTWVZA5P
The underlying theme: Microsoft made a visible infrastructure argument at a developer conference. Three AI workload classes — training, inference, agentic — with distinct investments behind each. Developers and enterprise architects should be paying closer attention to the infrastructure layer than most currently are.
4/ Some tantalizing thoughts on where this is heading. Frontier Tuning + RLE makes custom model development accessible without large infra or data science teams. m-Dash uses collaborative agents to autonomously surface and debate security vulnerabilities. And Satya's most forward-looking line: evaluations will become as strategically important as context and data.
3/ Microsoft IQ (Foundry + Fabric + M365) shows what platform-level integration can do for agents. Web IQ → Fabric IQ → Work IQ, each handling a distinct context type, wired together at the platform layer. Agentic AI has already driven a 3x increase in GitHub commits. Context quality is a real competitive variable.
2/ GitHub is being repositioned as the control plane for agentic workflows — beyond version control. The more immediate enterprise story: OpenClaw on Windows with MXC delivers process-level isolation and admin policy controls for open-weight models. That removes a real blocker for enterprise open-source AI adoption.
1/ Edge AI is getting real infrastructure behind it. Ion Instruct and Ion Plan run full agentic loops locally on Windows. Surface RTX Spark dev box targets Apple's hardware lead directly. Project Solara introduces agent-first form factors for field, health, and industrial use cases. This is infrastructure investment, not concept work.
New research, live today: plugging a general-purpose agent (Claude Cowork, Copilot 365) into your ERP is NOT just "connect an MCP server and go."
We talked to ERP vendors. They admit it's messy. Mission-critical ERP needs mission-critical agents.
The questions tech leaders should be asking 👇 https://t.co/bdQmI746iD
Good, fast or cheap, you can pick two. Easy to pick when you own the tokens factory. But, for enterprises, not so much.
Why Anthropic Dropped Markdown for HTML https://t.co/MF4jITVPnK via @YouTube
Google saved Antigravity for I/O — and it delivered. Desktop app, CLI, SDK — hooks, subagents, scheduled tasks. All the expected moves, but the shared SDK with Agent Builder is the smart differentiator: LOB teams prototype, devs take it from there. /Grill-me for stress-testing requirements is the sleeper feature. Migration window is open. Move now. #GoogleIO #Antigravity #AICoding
Customer keynotes are usually too scripted. @SAP’s Day 2 was different. Two things that worked:
1. Customers were honest about the messy parts — technology, change management, the real AI transition journey. Not a highlight reel or product endorsement.
2. Everyone stayed on stage. It became a panel, not a parade. The vibe got warmer and more credible as the hour went on.
Won’t change your software decision. But vendors take note — this is how you run a customer keynote.
#SAPSapphire #AI
🚨 Marc @Benioff is confirmed to open The Six Five Summit: AI Unleashed 2026.
We're thrilled to have the @salesforce Chair & CEO kicking off three days of executive conversations focused on the operational realities of AI. Marc will set the tone with his take on where enterprise AI is heading next and the realities of deploying AI within large organizations.
As the conversation shifts from pilots to production, AI Unleashed 2026 will explore what it actually takes to operationalize AI across infrastructure, cloud, edge, data, and enterprise workflows at scale.
No travel. No noise. Just the inside edge on your schedule, August 25-27. Registration is free. https://t.co/G5Z5HrEj2E