Tool use, retained context, and the choice of a next step turn a model-backed feature into a more complex application. Michael Albada's Building Applications with AI Agents examines the design decisions that follow, including when a single agent is sufficient and when coordination adds value.
Designing Agent Behavior
The book covers tools, orchestration, knowledge and memory, learning, single-agent and multiagent architectures, and validation. User experience is part of the design: autonomy, communicating uncertainty, and asking for input affect how an agent can be used.
The connection between architecture and behavior gives the book its practical value. Tools need clear responsibilities; memory needs a purpose; additional agents bring state and coordination to manage. Evaluation tests whether that machinery improves the task.
Where It Fits
This is useful for developers designing applications that can take several steps, especially before committing to a framework. Alongside the broader AI Engineering, it offers a focused treatment of agent workflows and the boundary between system autonomy and user responsibility.