Andrew Ng

Synthesized from job postings, expert interviews and surveys: four skills defining AI engineering — building and deploying AI applications, software fundamentals, using coding agents well, and shaping the build, meaning deciding what goes into the spec now that agents deliver against one.

A rebuttal to the advice that AI makes programming a dead skill, tracing the same prediction from Herb Simon's punchcard-era claim forward. The argument: every abstraction jump made coding easier and made more people worth teaching, and AI-assisted coding is the latest one.

Ng's breakdown of building AI applications: LLM foundations that predict failure modes, grounding beyond vector RAG, agent loop and tool/memory design, evals driven by disciplined error analysis, and production observability covering drift, cost and prompt-injection response.

Ng and Rohit Prasad release a local Mac agent aimed at finished deliverables — documents, Slack messages, calendar edits — rather than chat. Bring your own key across GPT, Claude, Gemini or open-weight models, or run Ollama locally, and it confirms before consequential actions.

Ng's version of 'loop engineering': a spec plus evals lets a coding agent write, test, and iterate for an hour without intervention, closing a new build every few minutes. The human feedback loops around it move far slower, and that mismatch shapes both how and what he builds.

A course with Anthropic, taught by Elie Schoppik, on using Claude Code past the basics: coordinating multiple subagents on different parts of a codebase, autonomous issue-to-merge GitHub workflows, notebook-to-dashboard cleanup, and Playwright over MCP for visual debugging.

Ng's skills map on the engineering half: full-stack literacy, data architecture agents can't cheaply retrofit, architecture that shifts by project phase, shift-left security, and production operation. Fundamentals are how you name the tradeoffs an agent will otherwise pick badly.

Andrew Ng on which software engineering fundamentals still matter when an agent writes the code: full-stack literacy, data modeling and lifecycle, architecture that changes by project phase, shift-left security, and production operation — framed as knowing the tradeoffs well enough to steer.

Andrew Ng breaks the emerging skill of directing coding agents into a lifecycle workflow (plan, execute, deploy/monitor) and five core competencies, including calibrating agent autonomy and reviewing output with automated and human checks.

An agentic paper reviewer trained on ICLR 2025 reviews reaches 0.42 Spearman correlation with human reviewers, against 0.41 between two humans. It grounds critiques by searching arXiv, so it works best in fields that publish there; usage has already outpaced NeurIPS submission volume.

Reading China's ban on major tech firms buying Nvidia as a self-sufficiency signal: DeepSeek-R1-Safe trained on 1000 Ascend chips, and Huawei's CloudMatrix 384 answers the GB200 with more, weaker chips orchestrated at system level. TSMC concentration is the mirror risk.

An extraction approach that treats a PDF as layout, charts and tables rather than a text stream — breaking the document into components and reasoning over each, targeting meaning-preserving output for RAG instead of OCR's flattened text.

Applying the economics of complementary goods to software teams: cheap AI-assisted building raises demand for the work that specifies what to build. Ng expects today's typical 4:1–10:1 engineer-to-PM ratios to compress as coding stops being the bottleneck.

Context Hub (npm i -g @aisuite/chub) is an open CLI that feeds coding agents curated, current API documentation on demand, aimed squarely at stale-endpoint and hallucinated-parameter failures. Agents can annotate docs with workarounds, building a shared record over time.

Ng reads the DeepSeek-R1 release and the market reaction as confirmation of three trends: the US–China capability gap has largely closed, permissively licensed open weights are commoditizing foundation models, and algorithmic innovation is cutting training costs faster than scaling alone.

The latest Context Hub release turns doc retrieval two-way: coding agents can annotate what a document got wrong or omitted, with safeguards for privacy and security, so corrections accrue across the community. Install via npm i -g @aisuite/chub.

LandingAI's Agentic Object Detection takes a natural-language prompt plus an image and reasons at length to locate the described objects, skipping training-data labeling entirely. A public demo is available for testing the approach on your own prompts.

The case that momentum, not current position, decides the AI race: China's open-weights ecosystem, hypercompetitive market and semiconductor push compound fast, while the US lead rests on advantages that policy alone won't sustain. Also a swipe at treating AGI as a finish line.

A working definition of the GenAI Application Engineer: fluency across many AI building blocks rather than one, combined with the ability to build systems dramatically faster using AI assistance. Includes the interview questions used to tell the two capabilities apart.

Replit's president and DevRel head teach agent-driven building in the cloud IDE — editor, package manager and deploy in one place. The load-bearing advice is Ng's: partition the problem, spec each module, generate and test one at a time, because one-shot prompting fails at app scale.

Agentic Document Extraction cut median processing time from 135 seconds to 8 while still pulling diagrams, charts and form fields out of PDFs as LLM-ready output — a latency shift that changes where document parsing can sit in a pipeline.

LandingAI's Agentic Document Extraction moves to a Document Pre-trained Transformer aimed at documents that break conventional parsers — large, complex tables above all — and ships an SDK that reduces a call to a few lines of code.

Ng splits the bubble question three ways: the application layer is underinvested, inference infrastructure still needs capital, and training infrastructure is where froth is plausible. He also notes VCs defaulting to infrastructure because picking application winners is harder.

OpenWorker's new release adds defensive security agents — code vulnerability scanning, dependency supply-chain checks, cloud misconfiguration review — to the open-source local agent. Model choice stays open, including fully local open-weight runs so sensitive code never leaves the machine.