Technique

Shift Work

Distinguishing between interactive and non-interactive system growth.

Two Modes of Growth

We distinguish between interactive and non-interactive system growth.

Interactive

The familiar cadence: generate, clarify, generate, approve, correct.

• Cursor, Claude Code, Codex
• Jointly evolving shared vision
• Expressing human intent into new systems

Non-Interactive

Tasks that are already fully specified.

• Formal specifications (RFCs)
• Existing working applications
• Execution runs end-to-end once intent is complete

Functional vs Non-Functional

Historically, software engineering distinguished between functional and non-functional specification.

The Instagram Example

A plausible Instagram clone can be described in a few sentences of functional intent. The technical difficulty of a global, image-oriented social network lies primarily outside the functional specification: scalability, performance, availability, security, efficiency, and related properties.

Non-Interactive Inputs

Canonical examples of non-interactive inputs include:

Formal Specifications
RFC 9113 for HTTP/2, often accompanied by validation suites (h2spec, nghttp2)
Spec + Test Suite = Complete Intent
Existing Working Applications
A system in an older Java dialect serves as a complete behavioral specification for reimplementation in Python
Executable Specification
"
An existing application constitutes an executable specification.

These cases motivated the development of reliable non-interactive coding agents, capable of operating over fully specified intent without human-in-the-loop clarification.

Our Implementation

Attractor

Our non-interactive coding agent, designed to operate without human interaction over fully specified work.

Attractor details →
FilesystemProducts