Andrew Ng¶
Summary¶
Co-founder of Coursera and founder of deeplearning.ai; influential educator and AI thought leader. As of May 2026, he publishes the weekly "Batch" newsletter and articulates frameworks for understanding how coding agents accelerate different software work types.
Details¶
Career¶
- Co-founded Coursera, the massive open online course (MOOC) platform
- Founded deeplearning.ai, a leading AI education company
- Published the Batch newsletter (weekly); Issue 350 (Apr 2026) featured a letter on coding agent acceleration
Frameworks and Thought Leadership¶
Ng has developed several influential frameworks in the AI education space. His most recently articulated operational framework (May 2026) maps coding agent acceleration to a four-tier ranking of software work types, from frontend most to research least accelerated. He notes he is "fascinated by how to organize software teams to use coding agents to achieve speed" and plans to continue publishing findings through this channel.
Views on Coding Agents¶
Unlike many AI leaders who speak in generalities about AI's potential, Ng provides a specific, pragmatic acceleration curve that distinguishes:
- Frontend — dramatically sped up (fluency in TypeScript/React, browser-loop closing)
- Backend — moderately accelerated (corner cases, debugging complexity)
- Infrastructure — minimally accelerated (complex tradeoffs, testing bottlenecks)
- Research — still very limited (hypothesis formulation, experiment interpretation)
His team management approach has shifted accordingly: he asks frontend teams to implement dramatically faster than a year ago, while research team expectations have not shifted.
Related Articles¶
- concepts/ai-inflection-point — The broader context of coding agent capability shifts
- concepts/harness-engineering — How teams reallocate capacity when agents accelerate different work types
- concepts/agentic-engineering — Professional engineering practices with coding agents
Sources¶
- The Batch Issue 350 — deeplearning.ai, Apr 24, 2026; Ng's letter on coding agent acceleration curve