Blog

The Letterwrite Blog

Notes on writing kernels, adversarial drafting, and prose worth reading.

Cover illustration for “Prompt Management as a Software Engineering Problem”

October 17, 2026

Prompt Management as a Software Engineering Problem

Versioning, testing, and governance turn prompts into managed production artifacts.

Cover illustration for “Content Pipeline Design for Engineering-Led Teams”

October 16, 2026

Content Pipeline Design for Engineering-Led Teams

Build content pipelines with specialized agents and validation gates, not prompt refinement alone.

Cover illustration for “Writing Kernels Explained for Engineers”

October 13, 2026

Writing Kernels Explained for Engineers

A four-step loop of draft, critique, vote, and recurse produces verifiable content at scale.

Cover illustration for “AI Detection in Publishing and Journalism Contexts”

October 12, 2026

AI Detection in Publishing and Journalism Contexts

Current AI detection tools cannot reliably catch machine-generated text in newsrooms.

Cover illustration for “Shredding LLM Output Across Multiple Models”

October 10, 2026

Shredding LLM Output Across Multiple Models

Splitting AI text across models defeats watermark detection by dispersing statistical fingerprints.

Cover illustration for “Substack AI Detection and What It Catches”

October 9, 2026

Substack AI Detection and What It Catches

Substack's AI detector protects human authorship claims that power its business model.

Cover illustration for “False Positive Rates in AI Detection Tools”

October 8, 2026

False Positive Rates in AI Detection Tools

Detectors flag honest AI use more often than evasion attempts, punishing compliant writers.

Cover illustration for “AI Watermarking Standards and How They Fail”

October 5, 2026

AI Watermarking Standards and How They Fail

Watermarking fails courtroom scrutiny and contradicts competing detection systems.

Cover illustration for “How AI Text Detectors Actually Work”

October 1, 2026

How AI Text Detectors Actually Work

These tools measure statistical patterns in text, not whether a human or machine actually wrote it.

Cover illustration for “Why Single-Model Output Is Structurally Detectable”

September 30, 2026

Why Single-Model Output Is Structurally Detectable

Statistical fingerprints in AI output survive attempts to hide them.

Cover illustration for “Model Fingerprinting in LLM-Generated Text”

September 30, 2026

Model Fingerprinting in LLM-Generated Text

Every language model leaves a detectable stylistic fingerprint in its text.

Cover illustration for “Humanizer Tools vs. Architectural Approaches to Detection”

September 27, 2026

Humanizer Tools vs. Architectural Approaches to Detection

Humanizers fool weak detectors, but architectural defenses are catching up.

Cover illustration for “Building an Internal Writing Quality Benchmark”

September 26, 2026

Building an Internal Writing Quality Benchmark

Establish test cases for AI content before publication, not after.

Cover illustration for “Content Operations Tooling for Technical Marketing Teams”

September 26, 2026

Content Operations Tooling for Technical Marketing Teams

Design content workflows before deploying AI, not the other way around.