Aaron Bjork

AI promises efficiency, but as this article points out, it can also reinforce silos when implemented in a department-first way. I’ve seen this happen across industries—each function adopts its own AI tools, celebrates local wins, but the organization misses the bigger opportunity. The fix is more than technical—it’s organizational: ✅ Start with enterprise-wide outcomes, not isolated department goals ✅ Use a hub-and-spoke model (central governance + embedded domain expertise) ✅ Incentivize shared KPIs that reward collaboration instead of siloed metrics ✅ Build on common data foundations so teams aren’t training models on fragmented inputs ✅ Invest in change management and leadership sponsorship to sustain adoption ✅ Review cross-function progress regularly to prevent silos from creeping back in When AI is deployed with this kind of systems thinking, it unifies rather than fragments. And the biggest beneficiary is the customer, because every function is aligned to deliver a seamless experience. #AI #DigitalTransformation #EnterpriseTech

6

1 Comment

Most AI agent samples teach you to call a model. They rarely show what happens when five agents need to coordinate, tools need to ship independently, and the whole stack needs to deploy to the cloud. Justin Yoo's Interview Coach sample fills that gap. It puts Microsoft Agent Framework, Foundry, MCP, and Aspire into a working application — an interview simulator where specialized agents hand off control to each other through a defined workflow. The patterns are worth studying: handoff orchestration where each agent fully owns its turn, MCP tool servers that deploy independently of agent code (and can be in a different language), and Aspire tying the service graph together with discovery, health checks, and tracing. The part that stuck with me — the principle of least privilege applied to agent tooling. Each agent only gets the tools it needs. The triage agent gets none. If you're moving multi-agent work toward production, what's the gap you're trying to close? https://lnkd.in/eamzATfP

15

CAT videos are back! Thanks Srini for sharing what it actually takes to scale AI agents in a large enterprise and Mehdi for bringing it all together. Next up: multi-agent architectures with Gary Pretty.

146

9 Comments

As AI workloads get bigger and more complex, we are engineering the full stack from our custom-built silicon all the way to the datacenter. Today we launched Maia 200, our next-generation AI accelerator chip. Maia 200 is an AI inference powerhouse: the most performant first‑party silicon from any hyperscaler, with three times the FP4 performance of Amazon’s third‑generation Trainium and FP8 performance above Google’s seventh‑generation TPU. It’s also the most efficient inference system we’ve ever deployed, delivering 30% better performance per dollar than the latest hardware in our fleet. Already running in our Iowa datacenter with impressive throughput, Maia 200 is accelerating today’s multimodal, multicall AI workloads with faster inference and higher output at scale. Read more in my blog post: https://lnkd.in/gRRc2Ndv

361

16 Comments

Wednesday Wisdom based on today's session at FabCon on building AI applications: AI models are *not* executable programs...ChatGPT != GPT-5.   ChatGPT is a chatbot (an app), GPT-5 is a model.   AI models are stateless - they don't have memory or persistence. #TheMoreYouKnow ✨

26

1 Comment

This is one of those key speeches for people who work on business strategy. We all need to realize that AI will invalidate existing work processes and deliverables. The success will depend on 4 factors: - does your team understand existing processes (big and small) and their outcomes - does your team have the visionaries that understand where AI can change and improve the process - does your team have the underlying data and data processing structures that AI will need to be successful - does your team have the capability to evaluate and QA the output the new agents

2