Andrew Ng on the PM bottleneck

Andrew Ng said faster AI-assisted coding is shifting the choke point in software from writing code to deciding what to build. (deeplearning.ai) Ng laid out the argument in a July 16, 2025 letter for DeepLearning.AI, after a June 17, 2025 talk at Y Combinator’s AI Startup School in San Francisco. He wrote that “product management is the art and science of deciding what to build” and said that task is becoming the new bottleneck in early-stage projects. (deeplearning.ai) (ycombinator.com) His example was speed: AI Fund, the venture studio he founded, builds about one startup per month, and he said agentic coding tools let teams move fast enough that product decisions must keep pace. In a later “No Priors” interview published August 21, 2025, he said work that once took six engineers three months can now be built “on a weekend.” (andrewng.org) (podscripts.co) The underlying idea is simple: if software gets cheaper to produce, the scarce input becomes judgment. Ng compared it to writer’s block after the typewriter made writing easier: the tool removed one constraint and exposed another. (deeplearning.ai) He argues that this changes team economics, not just workflow. In a January 15, 2025 letter, Ng wrote that many companies run engineer-to-product-manager ratios around 6:1, with a typical range of 4:1 to 10:1, and said product work and design work should take a larger share of software teams as coding gets more efficient. (deeplearning.ai) Ng has also tied that shift to a broader rise in custom software. In an April 10, 2026 letter, he wrote that as AI makes coding easier, “a lot more people will be doing it,” which points to more teams building software for narrower internal and external uses that would have been too expensive before. (deeplearning.ai) That same logic cuts the cost of revisiting old code. Ng’s recent writing says the future of software engineering is moving toward higher-level interaction with code through large language models, which reduces the penalty for changing or regenerating software and weakens one of the old arguments for living with technical debt. (deeplearning.ai) He has been less definitive on what the new org chart looks like. In the April 10, 2026 letter, Ng said the implications for job markets and “how software teams will be organized” are still being sorted out, even as he called the product-management bottleneck one of the clearer trends. (deeplearning.ai) Ng has made a similar point about training. His January 2025 note said companies already struggle to find product managers who both understand users and understand artificial intelligence systems, including feasibility, data, iteration, and model maintenance. (deeplearning.ai) The thread running through all of it is that AI coding tools may not eliminate the need for human product judgment. They may make that judgment the part of software work that gets harder to scale. (deeplearning.ai)