Ray Kao | Applied Research

About

Ray Kao

Ray Kao

I’m a Principal Solutions Engineer at Microsoft, part of the Cloud & AI Platforms Global Black Belt (GBB) team. The GBB team has historically been a specialized field engineering and product incubation team that works directly with enterprise customers to help solve complex business needs via novel technical solutions.

I’ve dedicated 10 years of my career to this role. Most of this time has been spent working alongside large engineering organizations internally at Microsoft and within some of the worlds top enterprises, working through difficult real world problems with the dedicated people and teams responsible for the outcomes.

This site is where I document the experiments, research, frameworks, and thinking that come out of that work.

Background

My technical roots are in open source software development, infrastructure, and cloud native computing. Before Microsoft, I co-founded and served as Chief Technologist at a digital consultancy in Toronto, where I worked with organizations building and shipping software during an earlier wave of platform change. That work included web and mobile applications, containers, distributed systems, and early container orchestration platforms such as Kubernetes, Mesos, and Docker Swarm.

That experience shaped how I think. I am less interested in technology for its own sake than in the organizational and architectural conditions that determine whether systems can work at scale.

My first chapter at Microsoft focused on open source solutions and Linux on Azure. Over time, the role evolved to include working with Azure customers on Kubernetes, developer productivity, infrastructure, DevOps, GitOps, and platform engineering. Much of that work involved closing the gap between what cloud native promised and what it actually took to operate reliably in real environments, especially the parts that do not show up in documentation.

More recently, my focus has shifted toward AI developer tooling and agentic platform engineering within Microsoft’s Cloud & AI Platforms organization, on the Developer Global Black Belt team. This is a specialized group that works across GitHub and Microsoft to help customers adopt modern developer tools and AI-assisted engineering practices.

The questions I am working on now center on what it means to build engineering systems that are ready for AI, both as a capability teams are trying to deliver and as an active participant in the engineering process itself.

This Site

What gets published here comes directly out of practice. I work with real organizations on real problems, and writing things down helps clarify the patterns, experiments, and frameworks that emerge from that work.

This is closer to a public working notebook than a traditional blog. It contains experiments, frameworks, and observations from the field, along with attempts to turn those into something more structured over time.

Focus Areas

AI Native SDLC

The software development lifecycle is changing as AI systems move beyond autocomplete into planning, coding, testing, and deployment. I am interested in what that looks like in practice, especially in organizations that are further along, and how it changes team structure, tooling, and process.

A central question for me is how human and AI agent interaction will evolve as implementation time decreases. If teams can move from intent to working software faster than before, then the bottleneck shifts from writing code to framing the right problems, making better decisions, and managing the systems around that work.

Platform Engineering

Platform engineering is how organizations build internal developer infrastructure, tooling, and experience so teams can move quickly without constantly rebuilding foundations or navigating unnecessary process friction.

A key question now is how these platforms evolve when AI systems become users and operators of platform capabilities, not just human developers. How do we move at AI speed while keeping the system understandable, governable, and useful at human scale?

Agentic Systems

AI agents that take sequences of actions in software engineering contexts are already being deployed. I focus on what works in practice, including the architectural patterns that hold up, the failure modes that recur, and the operational challenges that only become visible in production.

I am especially interested in how observability, telemetry, and tooling outputs can be connected to agents in ways that help humans make better decisions in event driven and automated systems.

Developer Productivity

In an AI assisted environment, productivity is harder to measure than it appears. I think of it as a property of the overall engineering system rather than an individual metric, and I am skeptical of approaches that optimize what is easy to measure instead of what actually improves outcomes.

The more important questions are often contextual. What actually matters to the organization? Which measures of success are broadly useful, and which are specific to a team, workflow, or company?

Working Thesis

Across much of the work I have been involved in, a pattern has been emerging.

AI is starting to act less like a simple augmentation tool for humans and more like a coordination and execution layer across the software development lifecycle.

As this continues, the shift is not only about developers asking AI better questions. It is about AI becoming part of an asynchronous, intent based engineering process that coordinates tools, workflows, and decisions.

From that perspective, the most important architectural decisions are not only about which models to use. Models will continue to improve and, in many cases, become increasingly interchangeable for common tasks. The more important question is how to build systems around those models so humans can focus on creativity, synthesis, judgment, and better decision making.

That means asking questions like:

  • what interfaces should be exposed
  • what state needs to persist across workflows
  • how actions should be audited and constrained
  • how intent should be captured and translated into execution
  • and where human judgment remains essential

That is the area I am most interested in exploring.

Connect