Agents That Get Things Done

Build AI Agents That Work—and Stay Under Control

Build AI agents that do useful work—and know where their authority ends. In Agents That Get Things Done, developer Valeria Andros offers a hands-on guide to designing, testing, and deploying tool-using AI systems that remain understandable and under human control.

Start with a real workflow and decide whether it truly needs an agent—or whether ordinary automation will do. Then turn the task into a practical design: define goals and success criteria, map the agent’s decisions, choose between a fixed workflow and a flexible loop, and set clear boundaries for what it may do.

As you build, learn to create narrow, well-defined tools; validate inputs; handle errors; and limit access with scoped credentials and safe defaults. The guide emphasizes a crucial principle: a model can propose an action, but your application must decide whether that action is valid and allowed. You’ll also learn to make assumptions, failure states, and human decision points visible, so you can assess behavior rather than trust a fluent answer.

Whether you’re a developer or a technically curious builder, you’ll finish with a working agent prototype and a repeatable method for evaluating, securing, and improving it. Build for reliable progress—not just impressive demos—and deploy with greater confidence.