Taking frontier models from flawless demo to dependable system: memory, orchestration, evaluation, and the governance that lets high-autonomy agents be trusted. I lead a regional AI engineering practice at Microsoft Services and build in the open as copilot-agents-dojo.
Memory, state, and orchestration: the architecture that makes an agent dependable, not just impressive in a demo.
Read more → 02The discipline that separates a demo from a system. Benchmarks, eval harnesses, traces, and the numbers that tell you it actually works.
Read more → 03Opt-in execution, verifiable guardrails, soul files, live controls. Engineering controls, not compliance theatre.
Read more → 04Hiring, coaching, and holding a uniformly high bar; executive advisory as a consequence of credibility, not the headline.
Read more →I run two second brains — a local Obsidian vault and a Cosmos DB grounding brain for my work fleet. Two earlier posts called them twins with one contract: grounded, governed, provenanced. Then an agent cited a half-finished note I'd written to myself, with total confidence, and I learned where the twin metaphor breaks. A second brain has two jobs that fight each other — capture and grounding — and one store can't do both without a membrane between them. Here's the membrane, on both sides, and the routing rule that keeps your own scratchpad from poisoning the answer.
Read →Two months ago I described the Mixture-of-Agents 'chair' as the smartest model in the room — the one that reads three drafts and writes the answer. Then I actually shipped it into my agent fleet. The first thing I deleted was the smartest model. Here's the reorder that made a panel worth running on ordinary work, the two ideas my own blog post had blurred together, and the brutal self-review of the piece that got it wrong.
Read →Two of 2026's strongest agent stacks, the same repetitive workload, thirty days of real runs. One of them quietly rewired itself and pulled ahead — and the reason it won is the reason most agent comparisons ask the wrong question.
Read →I lead the architects and engineers who turn the AI frontier into enterprise reality.
The work isn't only technical. It's translating ambition into platforms that are secure, resilient, and accountable — and building the world-class, diverse teams that can deliver them at the highest levels of customer impact.
Accelerated time-to-innovation, with AI moved safely from pilot to production.
Strengthened security postures across cloud and AI estates.
Platforms built to stay reliable under enterprise-scale load and scrutiny.
Cost optimization that holds up as platforms grow.
New business and revenue models unlocked by AI and cloud.
World-class, diverse teams capable of sustaining all of the above.
What I'm building in the open — pulled live from github.com/andreaswasita.
A behavioral governance framework for GitHub Copilot agents — skills.md and instructions to make AI coding agents think like senior engineers.
View on GitHub →A skills & discipline framework for Microsoft 365 Copilot Cowork — for knowledge workers using AI as a coworker.
View on GitHub →Mid-program feedback session. Reviewing student progress, unblocking technical decisions, and sharpening the path to the final showcase.
The capstone presentations — students share what they built, the design decisions behind it, and the lessons they took from the program.
I'm Regional CTO and Director of the Azure Cloud + AI Architecture practice at Microsoft Services, leading high-caliber architects, engineers, and cross-functional partners across APAC and beyond.
I'm an avid reader and a lifelong student of both technology and the financial industry, and I build in the open. copilot-agents-dojo and copilot-cowork-dojo are working frameworks for making AI coding agents behave with the discipline of senior engineers and collaborate as a team — where I prototype the agent patterns I bring to enterprise scale.
Always architecting the future — and deeply grateful for the teams and customers who make it possible.