THE OPEN-SOURCE LANDSCAPE

CHECK YOUR UNDERSTANDING • What does a license do? ◦ Answer: It states what you may and may not do with the software. • Why read a minimal reference implementation end to end? ◦ Answer: You see how the loop, tools, context handling, and checks connect, which teaches the whole design. • What is the danger of over-tuning to a benchmark? ◦ Answer: The harness learns to pass that exam instead of doing real work, so the score overstates real performance. • Name two things to check before adopting an open-source template. ◦ Answer: Any two of maintenance activity, security history, community size, licensing, and safe permission defaults. • Why try open-source code in a sandbox first? ◦ Answer: It may contain insecure defaults or harmful code, and a sandbox limits the damage while you review it. PART THIRTEEN: PERFORMANCE, COST, AND RELIABILITY ENGINEERING In the earlier parts of this book you learned how to build an agent harness that works. Recall from Chapter 5 that the harness is the software around a model that gives it tools, memory, rules, and a loop. An agent is a program that uses an AI model to pursue a goal by deciding on steps, using tools, and checking its own progress. Working is only the first requirement. A real product must also be affordable, fast, and dependable. Part Thirteen teaches you those three disciplines: cost, speed, and reliability. Think of a delivery company. It is not enough that the truck can reach the customer. The company also needs to know what each delivery costs, how long it takes, and how often packages arrive damaged or lost. In the same way, a harness engineer needs to know what each agent run costs, how long it takes, and how often it fails.