# AI Wiki - PKM - Vibe-Coding **Vibe-coding** is exploratory prototyping done with coding agents when the idea is still inchoate — you don't have a specification, you have a *vibe*. The term surfaces in Andy Matuschak's 2026 MIT HCI talk (attributed to designer Nio Ono), capturing a new working mode enabled by 2025-era coding agents: describe a feel, have the agent scaffold a prototype, iterate through dialogue rather than through ahead-of-time planning. The significance for PKM is that the same move applied to one's *vault* — vault as artifact being vibe-coded — is the practical form of the contemporary agentic-vault movement. ## What Makes It New Traditional prototyping required: 1. A coherent-enough idea to specify 2. Technical fluency to implement it 3. Hours of patient assembly of the surrounding scaffolding Vibe-coding relaxes all three: - **Idea can be inchoate.** "Make reading feel more like..." or "I want something that surfaces..." — if a designer can articulate a vibe, the agent can produce a first approximation - **Technical fluency is optional.** The agent writes code; the human evaluates feel - **Scaffolding is automated.** The agent generates the surrounding boilerplate, file structure, dependency setup, even the UI chrome The loop is: describe → scaffold → feel → refine. What used to take a day now takes fifteen minutes, at the cost of finished-ness (the output is usually prototype-grade, not production-grade). ## Matuschak's Framing In the 2026 talk, Matuschak uses vibe-coding to name the dissolution of the "programming as gatekeeping" dynamic. Previously: - A designer with a novel interface idea had to *pitch* the idea to a programmer who would implement it - The programmer's priorities, constraints, and taste inevitably shaped what got built - Iteration was slow because it required renegotiation with a human collaborator - Many genuinely novel ideas were never tried because the pitch overhead exceeded their apparent value With vibe-coding: - The designer can prototype directly - Iteration is *minutes*, not days - The designer can try ideas that wouldn't have justified a pitch - The designer's taste is preserved through the cycle Quoting Nio Ono from the talk: "I'm trying all kinds of ideas I just wouldn't have tried a year ago." And: "I feel unleashed in a way that changes my sense of self." ## The PKM Application: Vault Vibe-Coding The exact same move applied to one's vault is the practical basis of the contemporary agentic-vault movement: - A user with an inchoate workflow idea ("I want the vault to surface X when I do Y") can now ask an agent to scaffold it - The agent produces a skill, a script, a plugin, or a workflow that approximates the idea - The user tries it; refines the description; the agent iterates - What used to require a developer on-call (or giving up) now happens in a working session [[AI Wiki - PKM - Knowledge Work PRs|Knowledge Work PRs]] is the *review* discipline paired with vibe-coding. Vibe-coding generates the proposed changes; Knowledge Work PRs governs how they get integrated. ## When It Works - **Inchoate ideas benefit most.** If you know exactly what you want, traditional scoped development is still faster. Vibe-coding shines when the idea is *feel-shaped*, not *spec-shaped*. - **Disposable prototypes are acceptable.** Most vibe-coded artifacts are sketches, not products. If you expect them to ship, you'll be disappointed; if you expect them to illuminate the next iteration, they're invaluable. - **Agent capability matches task scope.** Vibe-coding a 50-line skill works better than vibe-coding a 5,000-line feature, in current tooling. ## When It Fails - **When production quality is required.** Vibe-coded artifacts have quality variance and lack the invariants a scoped engineering process enforces. - **When the human doesn't refine.** If the first output is accepted uncritically, the result is a pile of poorly-aligned artifacts that exist because the agent wrote them, not because they serve the user's actual workflow. - **At scale.** Vibe-coding ten skills is fine; vibe-coding a hundred with no deletion discipline is a chaos engine. ## Risks Specific to PKM - **Vault clutter.** Vibe-coding encourages creation. Without a deletion discipline, the vault accumulates half-working skills, partial plugins, and abandoned scripts. - **Context mismatch.** A skill vibe-coded against your current context may not survive a context shift (new project, new focus, new tools). Short half-life is normal; forgetting this creates disappointment. - **Over-automation.** Not every workflow benefits from automation. Vibe-coding makes it easy to automate things that didn't need automating, turning a clear human practice into an opaque agent pipeline. - **[[AI Wiki - PKM - Cognitive Debt|Cognitive debt]].** The ease of vibe-coding can substitute for actually thinking through whether a workflow is good. The agent implements what you described; it does not tell you you described the wrong thing. ## Relationship to Other Concepts - **[[AI Wiki - PKM - Knowledge Work PRs|Knowledge Work PRs]]** pairs vibe-coding with review discipline — vibe-coding generates, Knowledge Work PRs adjudicate - **[[AI Wiki - PKM - Agentic Constitution|Agentic constitution]]** provides the standing context that makes vibe-coded outputs more aligned with user goals - **[[AI Wiki - PKM - Research Purgatory|Research purgatory]]** — vibe-coding is one mechanism by which stuck ideas escape purgatory, because the reimplementation cost drops - **[[AI Wiki - PKM - Receptive Creativity|Receptive creativity]]** — vibe-coding operates across the receptive/active tension; you *receive* the agent's output while *actively* refining it - **[[AI Wiki - PKM - PKM Automation|PKM automation]]** — vibe-coding is one style of automation, distinct from deliberate engineered automation in risk profile and velocity ## Key Points - Vibe-coding = exploratory prototyping with coding agents when ideas are inchoate - Matuschak attributes the term to Nio Ono (2026) - Enabled by 2025-era coding agents; lowers three traditional prototyping barriers (coherent spec, technical skill, scaffolding time) - In PKM: the same move applied to the vault — skills, subroutines, workflows prototyped through agent dialogue - Works best for feel-shaped ideas, disposable prototypes, task-matched scope - Risks: clutter, context mismatch, over-automation, cognitive debt - Complements rather than replaces traditional engineering ## Open Questions - What disciplines prevent vibe-coded artifact accumulation (quarterly cleanup? TTL metadata? usage-based retention?) - Is there a crossover point where vibe-coding becomes real engineering, and how is that transition managed? - Does vibe-coding shift who can build PKM tools — toward designers and domain experts, as Matuschak argues, or just toward practitioners who already had some programming literacy? ## References - Andy Matuschak, "Apps and Programming: Two Accidental Tyrannies" (2026) — coined in context, attributed to Nio Ono - Contemporary practitioner writing across Engineering Agency Substack and other agentic-vault sources ## Related - [[AI Wiki - PKM - Source - Matuschak 2026 - Apps and Programming Two Accidental Tyrannies]] - [[AI Wiki - PKM - Knowledge Work PRs]] - [[AI Wiki - PKM - Agentic Constitution]] - [[AI Wiki - PKM - Research Purgatory]] - [[AI Wiki - PKM - Composability as PKM Architecture]] - [[AI Wiki - PKM - PKM Automation]] - [[AI Wiki - PKM - Agentic Knowledge Management]] - [[AI Wiki - PKM - Receptive Creativity]] - [[AI Wiki - PKM - Cognitive Debt]] - [[AI Wiki - PKM - AI Sycophancy and PKM]]