Custom AI workers and pipelines I built to compress my own throughput: from a developer agent with an identity to a content pipeline that turns a brain dump into a finished short-form video.
Off-the-shelf AI tools are calibrated for the average user. I am not the average user. I'm running multiple SaaS products in parallel as a solo founder. I need agents that know what I'm building, how I work, and what they're allowed to decide on their own. So I built them.
A controlled AI developer with a defined identity, working style, and chain of command. Alex is not a chatbot; he's a developer on my team. He takes briefs, executes independently where he can, flags blockers, and never goes above my word on scope decisions.
The framing matters: structured identity plus explicit role boundaries plus per-project memory produces vastly better technical work than a blank assistant prompt. This is the second iteration (MK2). The first version proved the architecture; this version drops the artificial knowledge ceiling.
A Python + Claude pipeline that turns a daily brain dump into a finished short-form video plan. One day in, one lesson, one feeling, one story out.
A smaller agent that builds and maintains a reusable component CSS library across my own projects. The Questra site uses its output. Saves rebuilding the same button forty times.