Skip to main content

SPECIFICATIONS ARE THE NEW CODE

What if the future of programming isn’t code — but clearly written intent?

At Genum, we believe in a fundamental shift: prompts are no longer temporary text instructions — they are AI specifications, the new source of truth. These specifications define how AI systems behave, what they value, and what they produce. They are not just assistants — they are replacing traditional code and orchestration logic.

In this new model:

  • Prompts become structured, persistent, testable artifacts.
  • They act as the logic layer of intelligent systems.
  • All downstream outputs — code, UI, documentation, decisions — are derived from a shared source-of-truth spec.

Genum helps teams clearly articulate intent and turn it into structured, validated, versioned AI behavior.

Specification eats the code

Prompt Types​

Based on how they process data and define behavior, prompts in Genum generally fall into the following categories:

Structured-to-Structured Transformers​

Convert one structured format into another. Common in workflows like:

  • Schema migration
  • API adapter design
  • Internal format conversions

Example: STRUCT_TO_STRUCT_TRANSFORMER

Unstructured-to-Structured Parsers​

Extract structured representations from raw or freeform data.

  • Form field extraction
  • Email/message parsing
  • Metadata capture from open text

Example: UNSTRUCT_TO_STRUCT_PARSER

AI Agents with Tools​

Combine prompt instructions with tool-execution capabilities.

  • Use tools (e.g., search, math, memory) during reasoning
  • Chain calls based on goal decomposition

Example: AI_CHAT_ASSISTANT_WITH_TOOLS

Contextual Categorization Prompts​

Classify inputs or assign semantic labels based on intent or context.

  • Email or ticket routing
  • Sentiment or topic classification

Example: MAIL_CATEGORIZER

Custom and Hybrid Prompts​

  • Routing logic across models
  • DSL-to-instruction converters
  • Validation logic for AI decisions

These prompt types can be chained, versioned, and tested within Genum.

How Genum Handles Prompts​

Genum enables specification creation through multiple entry points:

  • Manual input: Users can write their own structured specifications.
  • Natural language input: Simply describe what you want — Genum’s multi-agent system will generate a structured, versionable specification from it.
  • Voice input: Speak your intent — Genum transcribes, interprets, and builds the specification.

Each prompt in Genum becomes:

  • A tracked specification (with commit history and test coverage)
  • An API endpoint (for processing inputs externally)
  • A source of alignment (between teams, systems, and outcomes)

Prompt Creation

In Genum, prompts are not text — they are operational, testable, production-grade AI interfaces.