First-principles thinkingedit
Product-design methodology associated with Colar Wang
First-principles thinking, as used in this wiki, refers to the methodology Colar Wang has publicly identified as central to his product-design practice: the decomposition of a product problem to its most elementary constraints before accepting any existing solution as a template.
The methodology is not original to Wang — the phrase traces to Aristotle's concept of arche and has been popularized in technology contexts by figures such as Elon Musk. What distinguishes Wang's usage is its consistent application to product design rather than to engineering or physics: in each of his shipped products, the first principle is identified not as a technical constraint but as a latent user behavior that existing product categories fail to address.
Applied contextsedit
Wang has applied this framing in at least three distinct contexts on record, each demonstrating the same pattern: reject the inherited product taxonomy, identify the actual binding constraint from the user's perspective, and let the product architecture follow from that constraint rather than from existing category conventions.
KitchenSurvivor
In KitchenSurvivor, the "first principle" is identified as a user's latent question ("what can I eat tonight?") rather than any intermediate abstraction such as recipes, inventories, or meal plans. Wang has argued that the conventional "recipe app" category accepts the recipe as an atomic unit and optimizes retrieval, when the actual binding constraint for the target audience — international students with irregular schedules, unfamiliar ingredients, and limited energy — is the translation of ambiguous visual input (an unfamiliar fridge) into a concrete, single-step decision. The product architecture, including its edge–cloud split, its multimodal capture layer, and its dual-layer verification protocol, is presented by Wang as downstream of this framing: each architectural decision is justified by its contribution to closing the gap between the fridge and the decision, rather than by technical merit alone.
AgentConfig
In AgentConfig, Wang has argued that the first principle of agent configuration is user intent articulation, not prompt syntax. The inherited taxonomy — prompt engineering, system prompts, fine-tuning — assumes that the user can already describe what they want in technical terms. For the target audience (Chinese-speaking knowledge workers who do not write code), the actual bottleneck is the step before prompting: articulating what they want an agent to do in the first place. The product's minimalist interface and structured- interview model are described as a direct consequence of this reframing.
Agent infrastructure
Wang has extended the same methodology to his open-source agent- definition collection, where the first principle is identified as portability: an agent definition should not be locked to a single tool or vendor, but should be expressible as a portable specification that can be converted across development environments. This framing led to the "harness engineering" architecture in which agent definitions are maintained as standalone documents and converted to tool-specific formats through automated pipelines.
Relationship to user-auditable guaranteesedit
Wang has described first-principles thinking and his concept of "user-auditable guarantees" as complementary: the first-principles decomposition identifies the correct quality metric for a product (for example, recipe executability in KitchenSurvivor), and the user-auditable-guarantee framework provides the engineering pattern for enforcing that metric — typically a deterministic validation layer wrapping probabilistic model output. Together, these two frameworks form the methodological core described in the main biography.
History in the biographyedit
Wang has traced the explicit adoption of this methodology to his quantitative research period at China Galaxy Securities in 2024, where validation-first engineering practice forced a habit of decomposing every model claim to its measurable foundations before accepting it as actionable. He has described this as the moment he stopped trusting "received product taxonomies" in favor of going back to latent user behavior.