Andrew Ng suggests 1:1 eng:PM

Andrew Ng says AI-native software teams are changing shape as coding agents speed up engineering work and shift bottlenecks into product decisions. (deeplearning.ai) In an April 20 letter, Ng wrote that some teams are pushing engineer-to-product-manager ratios down from about 8:1 to as low as 1:1. He said that is happening because teams can now build software much faster. (deeplearning.ai) Ng argued that a one-product-manager, one-engineer pairing can still be too slow if every decision has to cross that handoff. He said the fastest teams now have engineers who can do some product work and product managers who can do some engineering. (deeplearning.ai) His core claim is that coding agents compress the time needed to turn an idea into working software. When coding speeds up by 10 times or 100 times, he wrote, deciding what to build takes a larger share of the schedule. (deeplearning.ai) That logic changes what companies hire for. Ng said strong engineers increasingly act as part product manager, part designer, and sometimes part marketer rather than staying inside a narrow coding role. (deeplearning.ai) He also tied speed to physical proximity. Small teams in the same office, he wrote, can move faster because face-to-face communication cuts delay between deciding and building. (deeplearning.ai) The bottleneck does not stop at product management. Ng wrote that some teams now build features so quickly that marketing struggles to explain them to users, while legal review can take a week for software built in a day. (deeplearning.ai) Ng published the essay on his site on April 20, and DeepLearning.AI highlighted it again in The Batch on April 24. The piece lands as he hosts AI Dev 26 in San Francisco on April 28 and 29 around the future of software engineering. (andrewng.org) (deeplearning.ai 1) (deeplearning.ai 2) His closing advice was direct: engineers should learn product management skills, and product managers should learn to build. In Ng’s version of an AI-native team, the org chart gets flatter because the handoffs get shorter. (deeplearning.ai)