Descriptions:
Andrew Ng takes the stage at AI Dev 26 San Francisco to share observations from his teams about AI-native software development and preview two new DeepLearningAI initiatives. His central argument: the practical gap between 80% AI-assisted coding and 100% AI-assisted coding is far larger than it appears. Teams requiring human review of every AI-generated line create a human bottleneck that dramatically limits throughput, while fully AI-coded-and-reviewed teams compound speed advantages across the entire development lifecycle.
Ng maps the downstream effects of accelerated software development: as coding speeds up 10–100x, product management becomes the new bottleneck, followed by design, legal compliance, and marketing. PM-to-engineer ratios have shifted from a historic 1:8 toward 1:1 or even reversed in some teams, with small generalist teams using AI to cover multiple functions simultaneously. He argues the key structural shift is that deciding what to build — not the building itself — now determines how fast teams move, and that smart AI-native teams can use AI to cover functions like basic marketing and legal first-pass review that previously required dedicated specialists.
The presentation closes with two announcements: a new ‘AI Prompting for Everyone’ course aimed at broadening AI literacy beyond technical audiences, and Context Hub, a new resource for navigating the AI developer tooling landscape. Ng explicitly pushes back on job-displacement narratives, citing Federal Reserve Bank of Philadelphia research indicating most businesses are not yet reducing headcount due to AI.
📺 Source: DeepLearningAI · Published May 20, 2026
🏷️ Format: Keynote Launch
