Sharad Bajaj, VP Engineering, Microsoft | Agentic AI & Data Platforms | Building Systems that Make Decisions, Not Predictions | Ex-AWS | Author

VP Engineering, Microsoft | Agentic AI & Data Platforms | Building Systems that Make Decisions, Not Predictions | Ex-AWS | Author

  • Microsoft
  • Computer Software
  • United States
Rank
#1,746of 20,024
  • 30.3K followers
  • 17 posts / 60d
  • 0.3% avg engagement
  • 1.8K follower growth / 30d

Sharad Bajaj ranks #1,746 of 20,024 creators on the LinkedIn Ranking, working in Computer Software and based in United States. They have 30.3K followers and published 17 posts in the last 60 days at a 0.3% average engagement rate.

The roast

Sharad claims he builds AI systems that make decisions rather than predictions, which explains why his own profile’s algorithm decided 28,000 people didn't need to see a single one of his posts this month. He is the human embodiment of a RAG pipeline: he pulls information from everyone else’s success and fails to ground it in a single personality.

About Sharad

I am a senior engineering executive with 25+ years of experience building and scaling enterprise software, cloud platforms, and AI-native systems used by millions of customers globally. I was a founding member of Microsoft Teams and have since led global engineering organizations across Microsoft and AWS Amazon Connect (CcaaS), spanning large-scale distributed systems, data platforms, and generative AI.Today, my focus is on building AI-native enterprise platforms where LLMs, data systems, and agent workflows come together to deliver real business outcomes. My work includes:• Agentic AI systems and AI agents that can reason, act, and integrate with enterprise workflows• LLM platform architecture, including Retrieval-Augmented Generation (RAG) and grounding systems• Evaluation-driven development (LLM evals) to improve reliability, safety, and quality over time• Data + AI platforms that unify analytics, applications, and decision systems• Responsible and trustworthy AI, including governance, compliance, and explainabilityI believe enterprise AI fails when treated as a model problem. It succeeds when designed as a system grounded in business semantics, constrained by real-world data, and measured by decision quality, not just model accuracy. I focus on operationalizing this through:• Business ontology and semantic layers for enterprise AI• Decision-grade platforms connecting data, models, and actions• Scalable AI infrastructure designed for production, not demos• Alignment across engineering, product, and business teams Across my career, I’ve learned the hardest problems are not model performance, but:• Designing for real-world failure modes• Scaling systems under production constraints• Building AI systems that earn long-term trustI am particularly interested in connecting with leaders working on enterprise AI platforms, agent ecosystems, LLM infrastructure, and evaluation-driven systems.

Highlights

  • Top 10% Audience

    : 30,307 followers

  • Top 10% in Computer Software

    : Ranked #377 of 5,292 creators

  • Top 25% in United States

    : Ranked #743 of 7,167 creators

  • Top 25% Creator

    : 17 posts in 30 days

Recent posts

  • Building on Dataverse used to mean long architecture cycles, too many tools, and too many handoffs. Now there's the Microsoft Dataverse plugin for Copilot Cowork, and it changes the flow. Video demo where we simply described an event registration app: venues, capacity, our existing contacts, and a waitlist when events fill up. Here's what happened next: → Cowork designed the tables, relationships, security and logic, the way an experienced maker would → It linked registrations to the Contact table we already had instead of creating a new list → Every change came to me as an approval first

    52 reactions · 4 comments · 5 reposts

  • Looking forward to being at PPCC26 and talking about a problem I think many of us are running into: Building an agent is getting easier. Getting one to actually work reliably in the enterprise is the harder problem. Join me , Robert Bruckner , James Oleinik , Julie Koesmarno as we dig into why enterprise agents struggle in production, and how Microsoft IQ, Work IQ and Dataverse can bring together business context, data, processes and governance to help change that. Lots of great announcements as well. Would love to see you there! 👉 Sign up for the session through the Whova app: 
https://l

    45 reactions · 2 comments · 3 reposts

  • Enterprise AI’s hardest problem isn’t generating an answer. It’s understanding how the business works well enough to change the outcome. Consider a customer renewal. Dynamics 365 Sales shows a healthy opportunity. Dynamics 365 Customer Service has an unresolved equipment issue. A custom application links the equipment to the customer’s contract. Emails and Teams conversations reveal that resolving the issue is a condition of renewal. Each system has part of the story. The business outcome depends on connecting all of it. This is the problem Business Applications in Work IQ is designed to s

    17 reactions · 0 comments · 3 reposts

Last updated 2026-10-23