W02 - Differentiation

This week I read Xiang Shuai’s 2020 Wealth Report, whose theme is divergence and choice. Looking back at this turbulent year from 2021, using the word “divergence” to understand the world’s evolution feels particularly apt. Alongside the pandemic, the globe experienced the highest unemployment rates since World War II, while at the same time the capital markets celebrated and top billionaires saw their fortunes multiply.

Something personal that also gave me an important insight: the trend of “divergence” doesn’t only apply to the world at large; it also deeply reflects the development of the engineer profession. The more creative an industry is, the more pronounced the 80/20 split becomes.

Today’s R&D, especially in large tech companies, is shifting from traditional labor‑intensive development toward innovation‑driven, platformized development. This process resembles industrial evolution: from smallholder production, to assembly lines, to large‑scale automation, and now to unmanned factories like Foxconn’s. It’s hard to say whether the barrier to becoming a programmer has risen or fallen, but the trend of divergence is clear. Either move up the path of cost reduction and efficiency gains, or accept a level playing field and become a “glue engineer.” “Glue engineer” is a term I imagine — as the name implies, it means doing the adhesive work between platforms: feeding inputs to systems or checking their outputs. There will always be tasks where running machines is actually more costly than human work.

A concrete example: in front‑end intelligence, some projects have already achieved notable results. Generating front‑end code from design mockups for secondary development (D2C) has seen promising progress; from this year’s D2 presentations, Alibaba has made impressive strides. The productivity gains in product and engineering are beginning to gain consensus. Alibaba’s D2C flagship product, Imgcook, covered over 90% of modules for the 2020 Double 11 venue, with a usable output rate of 79.26%, delivering a 1.5–2x increase in requirement throughput. They claim to have optimized usable output rate in code generation to its limits on D2C, and are now working on P2C by introducing PRD logic as annotations to push usable output toward 100% and aim to fully replace manual work. Similarly, Deco, a D2C project incubated by Aotu Lab, also claims it can soon reach the level of a T3 engineer.

Gates once said, “An excellent lathe operator earns several times the salary of an ordinary lathe operator, but the value of an excellent software engineer is 10,000 times that of an ordinary software engineer.”