Java Annotated Monthly Q&A: May 2026 – AI Agents, TDD, and the Human in the Loop

Welcome to the Java Annotated Monthly Q&A for May 2026. This month we’re diving into the latest developments in AI coding agents, test-driven development, and the evolving role of human engineers. Featured contributor Emily Bache brings sharp insights on how AI tools like agentic assistants are reshaping our workflows. We’ll explore key articles, talks, and reflections that challenge conventional thinking about AI in software development. Below you’ll find seven questions that capture the most important takeaways from this month’s edition.

1. What is the most important new skill for developers using AI coding agents, according to the May 2026 Java Annotated Monthly?

The critical skill highlighted this month is designing a “harness” for your AI tool. Birgitta Böckeler published what Emily Bache calls the best reference so far on this topic. A harness is a structured environment or set of guidelines that help you control and evaluate the output of an AI agent. It includes defining clear prompts, providing context, and setting up automated tests to validate the AI-generated code. Without a solid harness, developers risk wasting time on incorrect or irrelevant suggestions. The mental model involves treating the AI as a junior developer who needs constant oversight and clear instructions. Building a good harness is now seen as essential for productive use of AI coding assistants, especially for experienced developers who want to mentor others or integrate AI into team workflows.

2. Who is Emily Bache and what did she contribute to the May 2026 Java Annotated Monthly?

Emily Bache is an independent consultant, YouTuber, author, and technical coach with over 25 years of experience in Java and other languages. She specializes in agile practices like refactoring and test-driven development (TDD) and founded the Samman Technical Coaching Society. This month, she joins as a featured contributor, sharing her insights on the intersection of AI agents and TDD. She highlights several notable articles from April, including Birgitta Böckeler’s work on AI agent harnesses and Chris Parsons’ comprehensive guide on using AI for coding. Emily also contributes her own initial assessment of TDD in the age of agentic AI, based on interviews with practitioners she trusts. Her long experience makes her perspective particularly valuable for understanding how established engineering practices evolve alongside new AI tools.

3. What metaphor does Drew Breunig use for AI-assisted coding, and what risk does it highlight?

Drew Breunig introduced the metaphor of building a “Winchester Mystery House” to describe the danger of AI-assisted coding. The Winchester Mystery House is a famous, sprawling mansion with no coherent architectural plan, built continuously without a master design. Breunig warns that letting AI agents make changes incrementally without a clear overall vision can lead to a software architecture that is equally chaotic and unmaintainable. This metaphor underscores the risk of accumulating technical debt at an accelerated pace when developers rely too heavily on AI suggestions without maintaining a solid design. It serves as a caution to ensure that human engineers remain in control of the architectural direction, using AI as a tool that follows a plan rather than drives it. This idea is further explored in Kevlin Henney’s talk on being the human in the loop.

4. How did Michael Taggart’s experience report reflect on the ethics of using AI tools?

Michael Taggart published an introspective experience report that wrestles with his conscience over using AI tools. He acknowledges the productivity benefits but grapples with feelings of guilt and unease. His internal conflict touches on questions of authenticity, skill atrophy, and whether relying on AI diminishes the craft of software development. He doesn’t provide a definitive answer but instead opens a dialogue about the ethical implications of AI adoption. This candid reflection is important because it humanizes the debate, reminding us that many developers share similar doubts. It also encourages readers to consider their own values and the kind of engineers they want to be. Taggart’s report is a counterpoint to the more practical, tool-focused articles, emphasizing that technology adoption is not just a technical decision but a personal and ethical one.

5. What does Kevlin Henney’s talk “Being the Human in the Loop” emphasize about engineering skills?

In his talk, Kevlin Henney examines the engineering skills that remain critical even as AI coding agents become more prevalent. He focuses on the concept of being the “human in the loop”—the person who oversees, evaluates, and guides the AI’s output. Henney argues that skills such as critical thinking, system design, code review, and testing are more important than ever. These human abilities help prevent the kind of architectural chaos that Drew Breunig warns about. Henney also highlights the need for developers to maintain a deep understanding of the codebase and the business logic, so they can effectively judge the AI’s suggestions. His talk serves as a reminder that AI is a collaborator, not a replacement, and that the best results come from a synergy of machine speed and human judgment. This is essential reading for anyone worried about losing their edge in an AI-driven world.

6. What is Emily Bache’s initial assessment of test-driven development (TDD) in the age of agentic AI?

Emily Bache, a long-time TDD advocate, conducted interviews with several practitioners who are now using agentic AI tools. Her initial assessment, published this month, finds that TDD principles remain relevant but require adaptation. The main change is that developers now write tests not only to guide their own manual coding but also to “drive” the AI agent’s output. By providing a clear test harness, the AI can generate code that passes the tests, shifting the focus from writing code to specifying behavior. Practitioners report that TDD helps maintain quality and reduces the risk of AI introducing subtle bugs. However, the discipline of writing tests first is even more critical now, as the AI will blindly follow instructions. Emily notes that the best results come when developers treat the AI as a programming partner that needs the same kind of precise requirements a human developer would require. Her full analysis is a must-read for TDD enthusiasts and skeptics alike.

7. What other topics are covered in the May 2026 Java Annotated Monthly besides AI and TDD?

While AI and testing dominate this month’s newsletter, there is plenty more to explore. Java itself remains active with fresh updates and practical tips for developers. Kotlin continues to advance alongside Java, with new features and ecosystem developments. The AI section is packed with additional resources worth your attention, including links to articles, tools, and discussions. The newsletter also lists upcoming events that developers can plan to attend, ranging from conferences to local meetups. Finally, there are a few “ideas to challenge your thinking”—provocative concepts that push beyond the usual technical topics. These might include philosophical takes on software development, career development advice, or new methodologies. The goal is to keep the Java community informed not just about tools and code, but about the broader landscape of technology and practice.