A developer used three languages to develop a WordPress AI image generation plugin, hitting 12 pitfalls — the bottleneck of AI application development lies in the engineering toolchain, not the models.
What this is
This article from Juejin records a developer's entire process of building a WordPress AI image generation plugin from scratch. The requirement sounds simple: users input a prompt in the backend, call a self-built Stable Diffusion (an open-source AI image generation model) API to generate an image, and automatically save it to the media library. But the implementation spans three languages (TypeScript + PHP + Python), two frameworks (React + WordPress REST API), and one AI inference engine.
The 12 pitfalls cover the entire chain: Webpack not bundling un-imported CSS, WordPress npm packages lacking TypeScript type declarations, API paths being double-concatenated, Python 3.13 incompatibility with AI libraries, version conflicts between diffusers and transformers, and outdated CUDA drivers causing PyTorch to fail to recognize the GPU... Almost every step involves wrestling with the toolchain, not with AI.
Industry view
We note a judgment being repeatedly verified: the biggest friction in AI application development right now is not "the model can't," but "the toolchain won't cooperate." This developer's experience is a typical sample — none of the 12 pitfalls originated from insufficient model capability; all were engineering integration issues.
However, opposing voices are equally worth our attention. Some practitioners point out that such articles easily lead people to overestimate the difficulty of AI development. WordPress + self-built inference service is inherently a high-difficulty combination; using ready-made APIs (like OpenAI, Replicate) would drastically reduce engineering complexity. Self-building Stable Diffusion is a geek route and does not represent the average developer's real experience.
Another risk: such pitfall articles have a shelf life. Python 3.13 compatibility issues might disappear in six months, and diffusers version conflicts will be fixed. But the overall update speed of the AI ecosystem far exceeds that of the traditional Web ecosystem — as soon as old pits are filled, new ones may already be dug.
Impact on regular people
For enterprise IT: The hidden costs of AI feature integration are underestimated. A "simple" AI feature might require collaboration among frontend, backend, and AI engineering teams; an immature toolchain means troubleshooting time far exceeds expectations.
For individual careers: Full-stack AI developers capable of cross-end troubleshooting are currently scarce. But this window won't stay open long — once toolchains mature, this capability will downgrade from a core skill to a basic skill.
For the consumer market: In the short term, the experience of AI features in traditional software will be uneven, with massive differences in underlying integration quality. When users encounter bugs, the problem might not be the product itself, but the AI toolchain.