Subash Canapathy
Partner Director of Engineering, Microsoft
Engineering leader with 20+ years building enterprise-scale distributed systems, the last three on applied AI: people and organizational intelligence platforms, built on knowledge graphs, that power personalization and analytics in M365 Copilot. I lead large-scale organizational AI transformations, raising group-wide AI fluency and bringing an AI-driven, human-led SDLC to complex legacy systems.
As a change agent and servant leader, I foster psychologically safe, customer-obsessed teams where trust, curiosity, and ownership of outcomes lead to impactful, high-performing products.
Work highlights
Showcase of three v1 product launches, with details on the problem, the architecture, the calls that mattered and what they taught me.
How I work
Working style
Each block is 5% of a typical week
- Strategist35%Product strategy, Architecture, AI transformation
- Builder30%Innovation & AI upskilling, Agentic SDLC
- Coach20%People Ops, RoBs, 1:1s
- Operator15%Engineering excellence, S/W Factories & Customer relations
Executive competencies
AI
Applied AIAgentic SDLCInferencing modelsKnowledge graphsAI ROIAgentic budgeting
Leadership
Product strategyAI transformationSquads ModelMentoringChange Management
Engineering
Serverless computingLegacy-Agentic migrationAI-native architectureGenerative UI
Experience
Microsoft
Jan 2022 – now · 4 yrs 10 mosPartner Director of EngineeringSep 2024 – nowPWI: People & Work Intelligence, and AI Upskilling for Microsoft 365 Copilot · VancouverLead the people and work intelligence inside Microsoft 365 Copilot and Work IQ, and AI upskilling through Learning Agent.
What I built
I lead two groups. PWI: People & Work Intelligence is one of the producers of intelligence that goes into Work IQ, Copilot, Copilot memory and Copilot and FinOps analytics. AI Upskilling builds Learning Agent and runs Viva Learning.
What PWI powers, as a customer would notice it:
- Expertise matching. When Copilot works out the best person to ask about a project, topic or system, our understanding of people and their work is at the core of that match.
- People cards. Hover over anyone’s people card in a Microsoft 365 app, and the skill tags you see come from People Skills.
- Skills landscape. Leaders can see their organization’s skills profile and where to invest in upskilling.
- Task and workflow intelligence. Copilot can answer questions like “what is the most common type of work my team does”.
- What people use AI for. We produce the categorization and the taxonomy behind analytics on the kinds of tasks people bring to AI, for a company or an org.
- AI value and consumption. Task and task-completion intelligence, mapped to token cost, drives the usage page in the Copilot app and the org-level FinOps analytics in Agent 365 Business Analytics: AI measured by the work it gets done, not the tokens it burns.
Scale & architecture
- Org
- Three product teams, about 140 engineers and applied/data scientists
- Regions
- North America, India, LATAM
- Groups
- PWI: People & Work Intelligence · AI Upskilling (Learning Agent, Viva Learning)
- Feeds
- Work IQ, Copilot, Copilot memory, Copilot and FinOps analytics
- Platform
- AI inferencing and knowledge graphs, pre-computed intelligence
- Launched
- Learning Agent · people and org context in Work IQ · the work intelligence behind Agent 365 Business Analytics
Artifacts
Principal Group Engineering ManagerJan 2022 – Sep 2024Founded and led Viva EXP Foundations (horizontals) · VancouverReduced mean time to value for Viva suite deployments by 40%.
What I built
- Incubated and launched three platform products covering the shared service pillars of third-party data ingestion, data attribute governance and fast-track deployment of the Viva suite.
- Microsoft 365 People Skills. Started as a skunkworks hack with three engineers, came out of the Viva stable with an M365-scale intent, and shipped as the skills layer of Microsoft 365 Copilot in 2025. It grew into the People & Work Intelligence charter I lead today.
- Feature Access Management. There were feature-flag services and identity-bound access control, but nothing in between: a high-throughput, inline policy evaluator. We needed it for Viva and built it for all of Microsoft 365. Viva, Copilot Analytics, Viva Insights and many privacy-sensitive Copilot features use it today.
- Employee Self-Service Agent. Incubated in my group, which built the v0 shown in preview at Ignite 2024, before it moved to a dedicated team.
Scale & architecture
- Built
- Microsoft 365 People Skills · Feature Access Management · three Viva platform services
- Patent
- Context briefs for grounded LLM responses (application US 2025/0138910 A1)
Artifacts

Amazon
Jul 2016 – Nov 2021 · 5 yrs 4 mosEngineering Leader, Amazon MWAAOct 2019 – Nov 2021Amazon Managed Workflows for Apache Airflow · VancouverBuilt and launched a new AWS service in eight months; $40M ARR within six months of launch.
What I built
- Took a stalled idea to a launched AWS service: wrote the PR-FAQ, the business plan and the engineering strategy, then hired a team of 15 in under a month. The full story.
- Designed for margin from the start: a serverless control plane, a Fargate data plane with our own autoscaling controller, and a per-environment database so one customer’s outage stays theirs.
- Launched in every tier-1 region with region builds and DR automated before GA; then paused feature work for two months to stabilize, and the first year was uneventful after that, in the best possible way.
Scale & architecture
- Team
- 15 at launch (13 engineers, a PM, a designer); 30 when I left
- Launch
- GA November 2020 in 9 regions; 15 regions by August 2021
- Revenue
- $40M ARR within six months of launch
- Architecture
- Serverless: API Gateway and Lambda control plane, Fargate data plane, DynamoDB; an Aurora PostgreSQL per customer environment
- Open source
- Two engineers became Apache Airflow committers; one is now on the Airflow PMC
Artifacts
Software Engineering ManagerApr 2018 – Oct 2019AWS Marketplace Publishing, Catalog Management · VancouverImproved catalog security from 82% to 98% across all product types.
What I built
- Drove business strategy, product roadmap and delivery of self-service product ingestion for seller customers.
- Was responsible for overall catalog security and building the pipeline for vulnerability detection and patching of SaaS software stacks on AWS Marketplace
Software Development ManagerJul 2016 – Apr 2018Prime Video, Device and Lab Infrastructure Management · SeattleImproved streaming quality for millions of customers by validating QoS across all Amazon Fire device families.
What I built
- Led a two-pizza team that built and launched device infrastructure and management service for Prime Video live streaming. Responsible for validation and QoS improvements on Prime Video streaming across all device types.
Before 2016
Comcast, Amazon, EMCEarlier roles · 2006 – 2016
Software ArchitectthePlatform (a Comcast company) · SeattleLed the transition from bare metal to OpenStack and Kubernetes through a new deployment and continuous-delivery architecture.Dec 2014 – Apr 2016
Software Engineer / SDETAmazon · Chennai and SeattleInfrastructure engineering, Android/Webkit OSS development, and backend engineering roles over seven years.Jul 2008 – Dec 2014
Software EngineerEMC Documentum (Dell) · BangaloreDeveloper on the WDK test harness for EMC Documentum Webtop.Jul 2006 – Jul 2008
