When planners begin to think like programmers, real transformation begins—but not the superficial kind you might expect, like “using AI to draw urban planning diagrams.”

The most exciting aspect of this article from Huxiu is that it reveals the essence of AI empowerment: not tool replacement, but a fundamental restructuring of work logic. Vibe Coding started as programmers describing requirements in natural language to generate code, but its spread to urban planning (Vibe Planning) exposes a fatal flaw in traditional planning: those manual operations relying on experience and intuition are essentially unstructured “tacit knowledge.”

Many people’s first reaction might be, “AI can automatically generate urban planning solutions now,” but this interpretation is wildly off the mark. What’s truly noteworthy are three signals that disrupt traditional workflows: when planners start describing spatial needs through conversational interfaces (instead of manually drafting in CAD), when no-code platforms allow them to assemble functional modules like building blocks (instead of writing technical specifications), and when engineering environments automatically generate compliance reports (instead of manually cross-checking hundreds of pages of standards)—at this point, “planning” has transformed into a new form of knowledge engineering.

The first step isn’t to watch tech demos, but to examine planners’ original task lists
I dug into public reports from the Beijing Municipal Institute of City Planning and found that in a traditional planning process, at least 37% of time is spent on data cleaning and format conversion. Meanwhile, an AI planning platform piloted in a new district shows planners can now simply tell the system, “Mark all plots within 500 meters of subway stations with a floor area ratio exceeding 5.” What once required three professionals working for two days is now a five-minute natural language interaction. This isn’t about “drawing faster”—it’s about skipping the entire intermediate step of “translating business needs into machine instructions.”

The second step is bridging the gap in constraint comprehension
An experiment by a leading planning team in Xiong’an New Area last year was revealing: when they tasked AI with generating a road network plan, the first 10 versions were rejected for overlooking the implicit rule that “roads near schools must meet noise reduction standards.” The team realized these scattered constraints from various regulations had to be converted into quantifiable parameter labels. This suggests the core of future planning will shift from “drafting skills” to “rule structuring skills”—much like programmers no longer writing code but needing deeper expertise in business logic decomposition.

The most prone-to-failure stage is validation
A smart park project in Shenzhen highlighted a classic issue: the AI-generated plan was technically flawless, but on-site visits revealed the commercial flow design violated local merchants’ custom of “counters must face east.” This confirms my view—Vibe Planning isn’t about replacing planners but forcing them to articulate previously intangible tacit knowledge. Just as programmers must learn to communicate with AI via PRD documents, planners now need to master “describing spatial sociology in machine-understandable terms.”

A common misconception is that “AI will make planning easier.” The reality is the opposite: when you can generate 100 plans with natural language, the real challenge becomes defining evaluation criteria. As one planning bureau official put it: “The headache isn’t having no options, but not knowing how to tell AI what makes a plan ‘good.’” This requires translating abstract concepts like “urban character” or “spirit of place” into computable parameters—a task arguably harder than drafting.

The most disruptive insight here is how it exposes a shared dilemma across all “master craftsman experience”-reliant industries. Those ineffable pockets of tacit knowledge are becoming the biggest roadblocks to AI transformation. I’ve observed a new trend where top planning firms are creating “rule engineer” roles specifically to translate regulations, local customs, and even leadership preferences into machine-executable logic trees. This professional fragmentation mirrors software development’s split from “full-stack engineers” to “product managers + algorithm engineers.”

The next area to watch is “failure cases.” Current success stories are limited to new district planning with clear constraints, but what about old-city renewal projects involving complex stakeholder negotiations? Is AI simplifying problems or creating new ones? A red flag emerged when a tier-two city’s AI-generated urban renewal plan, overly reliant on Ministry of Housing data, completely ignored local “unauthorized construction replacement” unwritten rules, rendering the plan unexecutable. This reminds us: the more AI appears to “solve with one click,” the more human oversight is needed to set safeguards.

The real danger isn’t AI taking jobs, but reducing planning to a “parameter game” under the guise of AI. A lesson from Hangzhou’s industrial park last year is telling: every metric in the plan was “scientifically sound,” yet businesses moved out en masse because AI quantified “industrial synergy” simplistically as “three supporting enterprises within 500 meters,” ignoring real-world business dynamics. The future value of planners may lie in becoming “translators between machines and reality”—a skill ten times more valuable than drafting pretty diagrams.

(The End)