New interview just dropped.

I sat down with Adel El Hallak at NVIDIA GTC 2026 — VP of Product Management for NVIDIA AI — and we went straight to the question I’ve been writing about all year: who orchestrates the agent swarm?

Adel’s answer: NVIDIA’s internal deep researcher deploys agent swarms — Opus for planning, Nemotron for every specialist sub-agent. Hit #1 on both DeepResearch benchmarks. Then he confirmed the ServiceNow proof point — 90% of L1 tickets resolved autonomously. Agents that act, not suggest.

This is the thesis from “The Orchestration Layer: Why ServiceNow Spent $11.6B to Become CUDA for Workflows” playing out in real time.

We also dug into extreme codesign crossing from silicon into the model layer, Open Shell’s deny-by-default sandbox for enterprise agent trust, tokenomics as the new language of AI economics, and Jensen’s 2017 prediction that AI would write code — which is now reality inside NVIDIA.

Full interview:

If you haven’t read the pieces that frame this conversation:

→ “The Orchestration Layer” —

→ “Extreme Codesign Era” —

→ “NVIDIA Nemotron 3: Why This Changes Enterprise AI Economics” —