aam.
Menu

Lab

Agents that do real work.

Every entry here is a system that runs, with a write-up of what the agents decide, what they do not, and where a person stays in the loop. No demos, no slides.

  1. scheduletriggeragentactemaildataweb · browserobserveasks when unsure · 48h vetoyou

    In production since Aug 2026

    New Me: what the agents get to decide, and what they do not

    A job hunt run by scheduled agents on a Next.js and Supabase system. The rules: a scoring rubric before any application, explicit approval for anything using a fact the person has not supplied, and a person who can stop everything.

    Full write-up: in the writing queue.

  2. weeklyClaudedrafts48h vetosite schema≤ now · 10 minaamurad.comanalytics, signups, broken links: weekly report

    Live on this page

    This site runs itself, with a 48-hour veto

    Posts are drafted by an agent from an approved backlog, wait 48 hours for a veto, and publish on schedule from a Supabase table the site reads every ten minutes. A weekly agent reports traffic, signups and broken links. Nothing is posted to social accounts by a machine.

    Full write-up: in the writing queue.

Want one of these in your operations?

We pick one workflow, build the agent with a human veto in the loop, measure it, and only then widen it.

Applied AI, the lane →