newsmock.

The email is the joke.
The system underneath is the point.

On the surface: a useless email. One headline, one dry observation, delivered by an AI impersonating a 54-year-old tech veteran who has seen every hype cycle since before the dot-com bust and really just wants to have a nice lawn.

✓You're in. Expect a dispatch soon. Check your inbox. Or your spam folder, where all good things are initially filed.

No tracking pixels. No upsell funnel. No course on how to be an AI thought leader. Well, maybe someday, but not today.

What's actually happening

Underneath: a working multi-agent AI system, built by a curious non-developer who wanted to see one with his own eyes. The email is the deliverable. The system is the project.

Agent 1

Orchestrator

Given a goal — "ship today's dispatch" — and a set of tools. Decides what to do next, in what order, with what arguments. Owns control flow.

Agent 2

Scout

Reads 50 headlines across five news sources. Returns a curated shortlist worth mocking. Invoked by the Orchestrator on demand.

Agent 3

Critic

Scores every draft, returns feedback, demands revisions. In one run it forced 14 redrafts across two abandoned stories before approving.

Three AIs. One email. None of them following a script. Each one making real-time decisions about what to do next. AI researchers call this the orchestrator-worker-evaluator pattern. The output looks trivial. The architecture is not.

The output can be small.
The architecture is not.

Most of what's marketed as an "AI agent" right now is a saved prompt with a slide deck. A real agent is an orchestrated system where an LLM owns control flow, makes runtime decisions, and delegates to other LLMs that report back. Newsmock was built to demonstrate the difference at the smallest possible scale.

The email is a byproduct of the lesson. If you find it amusing, subscribe above. If you find the system underneath more interesting than the email itself — good. That was the point.