The SOMspec Journal

Writing

Analysis, research, and practical guides on the Semantic Object Model, agent-native publishing, and the economics of AI web fetching. Updated as the field evolves.

Featured
Benchmarks

What Three Weeks of Public Benchmarks Reveal About the Web’s Token Bill

Three weeks. Two Plasmate releases. Thirty-eight production sites. The WebTaskBench public benchmark gives the first defensible empirical answer to a question publishers and agent authors have been asking each other since 2024: how much of the web’s token cost is actually load-bearing?

Read the article →9 min read

Specification

SOM/1.0 format mechanics, schema design, conformance.

1 piece
Specification

SOM vs MCP: How Publishers and Agents Are Different Problems

MCP is the protocol an agent uses to talk to its tools. SOM is the format a publisher uses to describe a web page. They are routinely confused, frequently positioned as alternatives, and in fact occupy different layers of the stack — designed to compose, not compete.

8 min read

Benchmarks

Empirical token efficiency data from WebTaskBench and the field.

2 pieces
Benchmarks

TechCrunch Was Blocked. Now It's 77×. What Changed?

Plasmate v0.5.0 fixed something important: major news sites blocked by anti-bot protection now fetch cleanly. TechCrunch went from the failure list to 77× compression — the highest in the news vertical.

5 min read
Benchmarks

How to Read the WebTaskBench Leaderboard

What does a 43× compression ratio actually mean for an AI agent? A practical guide to interpreting token efficiency data and why it matters for your publishing economics.

4 min read

Compliance

robots.txt SOM Directives, agent discovery, framework adoption.

1 piece
Compliance

The Discovery Gap: Why AI Agents Miss Your SOM Directives

Even when publishers correctly implement SOM Directives, most AI agents never find them. The research explains why — and what framework authors can do about it.

6 min read

Publishers

Adoption guidance, publisher economics, implementation patterns.

1 piece
Publishers

The Publisher's Third Option

Publishers face a binary: block AI agents or serve them raw HTML at full cost. The robots.txt SOM Directives proposal offers a third path — cooperative, economically rational, and five minutes to implement.

5 min read

Industry

Standards landscape, regulation, and the agent-native web.

2 pieces
Industry

The Web’s Second Reader

The web had one reader for thirty years. Tim Berners-Lee saw the second reader coming, but his Semantic Web asked humans to do the writing. AI agents are the readers he predicted, and the format they need is the one that finally serves both audiences without forcing either to compromise.

11 min read
Industry

SOM vs llms.txt: When to Use Which

llms.txt tells an agent what your site is. SOM tells an agent what your page contains. They are different layers of the same problem, and publishers should ship both.

7 min read

Subscribe via JSON Feed or XML sitemap. Agents can also fetch llms.txt for a structured site index.