Vibe-noting

Type: kb/types/note.md

One advantage of vibe coding is that code is a stored, structured, inspectable artifact. A stateless agent can read a codebase in a new session, recover its explicit implementation state, and continue from there. Tests, types, runtimes, and interfaces do different work: they constrain and verify changes. Inspectability creates continuity; executable constraints make that continuity more reliable and make automation safer.

The linked ephemeral-software critique sharpens the analogy's scope. AI may make disposable scripts and prototypes cheaper, but important software does not become safely forgettable. State, integrations, interface expectations, and audit needs still make durable artifact stacks load-bearing. Vibe-noting makes the analogous weak claim. Inspectable knowledge artifacts — durable records of reasoning — can make augmentation cheaper and more continuous, but they do not make durable knowledge work disappear.

Much organizational knowledge work is not retained in a form a stateless agent can reuse. Decisions live in chat threads, reasoning evaporates after meetings, and analysis sits in documents without stable addresses or navigable relationships. A later session may find the text and still fail to recover the shared vocabulary, reasoning lineage, or which artifact represents the current explicit state.

A KB supplies the missing layer only when its artifacts are stable enough to cite, linked or routed by reader need, and maintained so superseded or low-quality material does not masquerade as current. Then notes can serve reasoning as code files serve implementation: explicit state an agent can inspect, parse, and use. This is the practical core of vibe-noting. A later session can continue from an accumulated argument instead of reconstructing it from the original exchange.

Inspectability makes that recovery possible, not automatic. Knowledge storage does not imply contextual activation: the system must still route relevant artifacts into context and bring them to bear on the task. Poorly written or incorrectly linked notes can instead compound negatively and make recovery worse than starting fresh. The weak oracles that make KB automation hard also make this degradation expensive to detect.

The reverse-compression failure mode

The common failure mode of vibe-noting is reverse-compression: a human offers one sentence, the agent expands it into a full article from its training knowledge, and the article carries no more epiplexity — learnable structure the intended reader can extract — than the original sentence. It looks deep while only expanding a compact signal.

Links resist this failure only when their targets add load-bearing information that the seed did not contain. Decorative links leave the expansion unchanged. The full analysis develops that condition.

The tension with verification

The boundary of automation is the boundary of verification explains why code's cheap oracles support automation, while KB work still depends on judgments such as whether a connection is right or a synthesis is useful. Vibe-noting targets the remaining augmentation opportunity.

This yields two independent axes for LLM-assisted work. The verification axis comes from the linked note; their combination is this note's contribution:

  • Inspectability — can a later session recover the relevant explicit state?
  • Verifiability — can the system check whether a new output is acceptable?

Code is strong on both axes. Unstructured knowledge work is weak on inspectability and often weak on verification. A maintained KB raises the first axis without necessarily raising the second. That can lower the cost of human-supervised augmentation without making knowledge work self-verifying or autonomous.

Relevant Notes: