Little Worker

Give tasks to AI agents, review the results, and keep track of the work.

Overview

Little Worker lets me give tasks to AI models, set limits, review the results, and return to the work later. I use it for client work and share what I learn in videos and writing.

Problem

AI work gets split across models, tabs, prompts, approvals, and notes. It becomes hard to see what happened, what it cost, who approved it, or what should happen next.

Solution

Little Worker keeps the screen simple and puts the hard parts behind it: model choice, permissions, approvals, spend limits, evidence, training, reusable workers and plugins, and state that survives each session.

Decisions

Keep the work surface simple

A user should be able to hand off work and review it without learning the model and infrastructure stack underneath.

Put people at the approval boundary

The system records a clear decision before approved work moves forward.

Build with real work

Client work tests the system under real deadlines, permissions, budgets, and quality standards.

Lessons learned

The best work surface can stay simple even when the system underneath it is deep.
Approval, evidence, and spend controls have to be part of the work, not reports added later.
Using the product on real client work finds gaps that demos miss.
Durable state matters because serious work lasts longer than one chat session.