Notes
Pànxing's Redesign Is Frozen
Twenty years' worth of versions, a redesign in its final stages, and a pause the project did not expect. The end —for now— of this stage, and a lesson about AI with no brakes.
2 min read
Today, Pànxing's redesign reaches a point that was not on the roadmap: the freeze. For now, there will be no more phases.
This was not my decision, nor did it come because the project had stopped. On the contrary: the structure was in its final stages. The nine-phase roadmap had been written and documented, with forty-six specifications explaining each block before it was built. The editorial design had been validated block by block, without breaking anything along the way. The new platform already worked: the old WordPress was migrating to Payload in phases, the advertising carried the advertiser's contact details with it, subscriptions implemented genuine double opt-in, and newsletters had a composer, delivery, and a public archive. We ran a full audit of the project, which found a JWT that nobody was verifying—and we closed that gap—we fixed a backup restore process that worked by accident, set up CI so that no error could reach main silently, and the production phase, the eighth, already had its foundation in place. Forty tasks on the list, the critical block closed.
And then, without it being part of the plan, the project continued on another track. Management built its own version in parallel, with the help of an AI assistant, and I found out as you did: from the result, not through a conversation. Twenty years of making every version of this website—design, code, content, updates, all by myself—weren't enough to earn so much as a knock to ask, "what would you make of it?"
I am writing this without rancor, but with a lesson I want to set down in writing because it applies to all of us: ignorance is a double-edged sword. Today's AI tools are versatile and powerful, yes. But a model does not know what you do not ask it. Without knowledge of programming, design, or SEO, a project built solely from instructions in a chat seems functional until it collides with reality: a form that filters poorly, a migration that silently loses data, a website that is quick to build and slow to live with. The bomb is not the tool: it is the distance between what the model gives you and what you can verify. I have spent twenty years learning to verify this, and it is precisely what cannot be borrowed.
The project remains frozen with the door open: well-organized repositories, readable specifications, a validated design, and a log that explains every decision. If it is ever unfrozen, there is enough material to get started in an afternoon. And if the parallel project bears fruit, so much the better for everyone: the website is not mine; it belongs to whoever cares for it.
For my part, I am taking away the method of comparing before applying changes, the specifications before the code, and proof that a single person can sustain an entire product if they know why each thing is done.
For as long as it is needed, Pànxing.