Philosophy and design commitments
WeaveMark is not only a notation. It is a proposal about the relationship between language, cognition, and computation: prompts can formulate ways of thinking, and prompting can be understood as a practice of cognitive structuring. Natural language already carries rich structure; careful abstractions can help us inspect and compose that structure without displacing language itself.
Foundational principle
Prompts are often treated as blobs of unstructured information. But natural language is already richly structured through hierarchy, sequence, scope, contrast, reference, and argument. WeaveMark’s computational abstractions help expose, compose, and more fully leverage that structure without replacing language as the primary medium.
Prompting is not merely the phrasing of a request. It is the work of formulating a way of thinking: what to notice, how to frame and decompose a problem, which perspectives and reasoning methods to apply, how to test conclusions, and what form an answer should take. In that sense, WeaveMark is a language for cognitive structuring: a way to structure cognition through readable prompt specifications.
Design commitments
These commitments keep WeaveMark from becoming either a pile of prompt tricks or a conventional programming language wearing a Markdown costume.
Language stays primary
A promplet should remain readable before compilation. Directives mark meaningful seams—reuse, transformation, branching, contracts, execution—while ordinary language carries the purpose, argument, nuance, and domain knowledge.
Cognitive structuring is composable
To structure cognition is to shape attention, decomposition, perspective, inference, criticism, and expression. Evidence standards, issue trees, teaching moves, adversarial checks, personas, and output disciplines capture parts of that structure. WeaveMark gives them names and boundaries so they can shape many promplets without being copied into each one.
Semantics before plumbing
Operations such as refining, revising, normalizing, compressing, or changing style are transformations of meaning, not merely string substitution. WeaveMark lets authors declare those moves and uses a language model to realize them in context.
Formality serves meaning
Formal structure is valuable when it makes an important boundary visible, checkable, or reusable. It should not turn a readable cognitive artifact into incidental plumbing. WeaveMark aims for the smallest computational surface that genuinely improves expression and understanding.
Power remains accountable
Semantic compilation involves judgment, and runtime interpretation can produce effects. Neither should disappear behind a magical interface. Authors and reviewers should be able to inspect what was written, what was compiled, how it was interpreted, what was produced, and what may act.
Three kinds of structure
WeaveMark connects structures that are often discussed separately, while keeping each one visible.
Hierarchy, sequence, scope, reference, contrast, argument, and examples already organize natural language.
Attention, framing, decomposition, perspective, evidence, criticism, synthesis, and expression organize thought.
Composition, variables, branching, transformation, contracts, and explicit effects make structures operable.
Power needs machinery
Its generative and interpretive capacity becomes useful when language gives it shape and computation gives it controlled paths into the world.
Source
Transforming machinery
Useful work
Electricity without machinery is only potential. Likewise, an LLM’s capability is not yet a workflow, a decision, or a trustworthy artifact. Promplets, processors, engines, tools, and output contracts transform that power into work that can be understood and inspected.
Programming after AI agents
Saying that programming is over because Copilot CLI, Claude Code, and similar agents can write most of today’s programs confuses the current medium with the enduring task. Formal programming languages are one extraordinarily successful way to communicate intent. They are not the definition of programming.
An agent still cannot build what a person needs unless it can understand that person’s intent: goals, distinctions, constraints, tradeoffs, evidence, quality bars, and acceptable effects. As agents absorb more implementation work, this act of specification becomes more central, not less.
Natural language
Formal structure
Promplets
Programming is not disappearing. It is moving upward into new languages of intent.
Specification and refinement
WeaveMark is not a formal method, but it takes substantial inspiration from formal methods—especially their disciplined notion of specification.
A precise vocabulary
Let Spec be an abstract description of required
properties, and let Imp be a more concrete description
of an implementation. Read both as predicates over possible
behaviors or outcomes.
A correct implementation or refinement must add enough detail to become useful without violating any obligation in the specification.
Imp ⇒ SpecEvery behavior that satisfies Imp also satisfies Spec.Spec ⇏ ImpA behavior can satisfy Spec without satisfying this particular Imp.⟦Imp⟧ ⊆ ⟦Spec⟧The set allowed by Imp is narrower because implementation adds detail.Specification Spec
Implementation Imp
This gives refinement a clear direction: a refinement is not merely a rewrite, an expansion, or a stylistic improvement. It strengthens a description by making choices and adding detail while preserving everything the more abstract description requires.
That idea informs WeaveMark’s @refine directive and
its broader compile-to-implementation workflows. But the relationship
is a design obligation, not a machine-checked theorem:
model-based compilation can fail to preserve intent, so the resulting
artifact remains visible for inspection, testing, and further
refinement.
Why use LLM-based compilation? Why not only templates? Is this literate programming?
Continue to the FAQFrom philosophy to practice