Mahendra Sekaran, Microsoft CVP, on AI Transformation, Copilot, Work IQ, and Multi-Model Agentic AI

Prefer audio?

Who should listen: Microsoft 365 service owners, Teams voice and contact centre leads, and partners planning agent deployments who want a first-party view on Work IQ, Agent 365, multi-model strategy and Teams Phone Agent — plus a candid look at how Microsoft's own product teams have restructured planning and development around agents.

Guest: Mahendra Sekaran — Corporate Vice President, Product Management and Science, M365 Core (communications and collaboration), Microsoft

Mahendra Sekaran leads product management and science for M365 Core communications and collaboration, the platform behind messaging, calling, audio and video across Teams, Copilot, Copilot Studio and Foundry agents. He talks to Tom Arbuthnot about how his own PM and engineering teams have rewired planning and development around agents, where Work IQ fits in the frontier transformation stack, Microsoft's multi-model position, and the early expectations for Teams Phone Agent.

Many thanks to Pure IP, sponsor of this episode.

Key insights

• Azure grew around 43% in the most recent quarter, with remaining performance obligations quoted at roughly $678 billion — the number Sekaran points to as evidence customers are committing capital to AI and cloud, not just experimenting. ▶ 4:05
• A PM in the team built a planning agent grounded on several semesters of Azure DevOps intake data plus access to the code repositories; it produces cost estimates and writes them back into ADO at roughly 80% accuracy, collapsing a planning cycle that involved tens to hundreds of people over six to eight weeks into days. ▶ 8:15
• MCP servers are now deployed across most of the team's data assets, so PMs interrogate product telemetry directly (for example, where SMB calling plan growth is coming from) instead of queuing weeks of work with data engineers and analysts. ▶ 9:46
• The comms platform is built from tens to a couple of hundred microservices on seven-to-ten-year-old code bases dating back to Skype consumer; the six-to-eight-month priority has been wiring full agentic loops — deployment, validation, monitoring and production rollout — with the goal of covering the entire fleet within months, while keeping a human on the final go decision. ▶ 11:50
• Work IQ is positioned as the layer that makes documents, email, chats, meetings and call transcripts intelligible to Copilot and agents; because the data is already semantically indexed and grounded, Sekaran reports significant reductions in token consumption and response latency versus customers extracting data and re-grounding it themselves — and avoids rebuilding governance, compliance and retention outside Microsoft's boundary. ▶ 21:30
• Work IQ is not just an internal dependency for Copilot: its retrieve and grounding APIs are available to customers for line-of-business agents, respecting existing access controls — Sekaran cites a wealth manager agent versus a loan risk-assessment agent as examples of role-scoped grounding. ▶ 22:34
• On model strategy, Microsoft's line is choice plus an intelligence and governance layer: OpenAI, Anthropic and in-house models, with automatic selection on cost, efficiency and performance — around 11,000 models accessible in Azure. Sekaran frames multi-model as insurance against regulatory or geopolitical loss of access to a single model, as well as a token-cost lever. ▶ 23:37
• Teams Phone Agent is in Frontier public preview with an expected 70–80% deflection of calls that previously needed a human; the same agent can be fine-tuned to sit in front of a contact centre, and Sekaran sees the real value in the post-call agentic workflow (CRM updates and downstream business processes) plus a partner opportunity around agent build, eval frameworks and ongoing validation. ▶ 31:46

Insights summarised by AI from the episode transcript, reviewed by the Empowering.Cloud team.

Listen: Apple Podcasts · Spotify · Other platforms