Signal

From the archive

Honest writing about AI tools and what they mean. No hype, no sponsored takes. Just signal.

Subscribe by emailFor people who'd rather subscribe than scroll.

From the editor

Apr 22, 2025

Why context windows changed everything (and nobody talks about it)

The benchmark wars are loud. Context window size barely gets a headline. But the jump from 4k to 128k tokens is the single change that made AI assistants feel like colleagues instead of calculators — and it quietly rewrote what's possible with RAG, agents, and long-form reasoning.

Apr 10, 2025

The honest case for fine-tuning (and when it's the wrong answer)

Everyone recommends RAG first. That's usually right. But there's a specific class of problem — consistent tone, domain-specific jargon, structured output formats — where fine-tuning is the only tool that actually works. Here's how to tell which situation you're in before spending the budget.

Mar 25, 2025

MCP is the most underrated thing to happen to AI tooling this year

Model Context Protocol quietly solved the integration problem that was making AI agents painful to build. Not by being clever — by being boring. A single standard that any LLM and any tool can speak. The implications for how we build software are still unfolding.

From Medium

Aug 18, 2025

From the archive

Why context windows changed everything (and nobody talks about it)

The honest case for fine-tuning (and when it's the wrong answer)

MCP is the most underrated thing to happen to AI tooling this year

Happening

Mist-ache

Scents

Trust

Humility

Cadence is a strong word.

From the archive

Why context windows changed everything (and nobody talks about it)

The honest case for fine-tuning (and when it's the wrong answer)

MCP is the most underrated thing to happen to AI tooling this year

Happening

Mist-ache

Scents

Trust

Humility

Cadence is a strong word.