TDD

Evidence (11)

• 5Hacker News·9/8/2026research
  Study compares testing techniques for coding agents

  This Hacker News post describes an evaluation of how coding agents perform when explicitly instructed to use different testing and verification techniques while implementing Zstd in Rust. It compares 26 prompting conditions, including TDD, property-based testing, fuzzing, formal methods, and several Rust testing tools, plus a few skill-based variants. The author also reports additional smaller evaluations, such as one based on the IMAP RFC.

• 5The New Stack·8/31/2026framework_update
  Why coding agents need a context development lifecycle

  The article argues that skills, prompt instructions, agent configurations, and rules files should be treated as software artifacts with their own development lifecycle. It proposes a Context Development Lifecycle (CDLC) with four phases: generate, evaluate, distribute, and observe, to improve reliability and reduce regressions in coding agents.

• 5Hacker News·8/15/2026research
  Software engineering fundamentals still matter in the age of coding agents

  The article argues that software engineering fundamentals remain essential even as agentic coding tools become more capable. It emphasizes that building software requires careful decisions about architecture, seams, maintainability, debuggability, and long-term fit, areas where current LLM-based agents still fall short. It also notes that harness-based workflows, TDD, and good validation tooling can make these tools more effective, while open-weight models are making local use more practical.

• 4Martin Fowler·8/11/2026research
  Agentic TDD shows little clear benefit in small eval

  An exploratory evaluation compared AI coding workflows that used test-driven development inside an agent loop with workflows that did not. The results showed no clear quality advantage for the TDD approach, and in some cases the non-TDD solutions ranked slightly higher in design and test quality.

• 5The New Stack·7/25/2026framework_update
  Build AI Context Systems Instead of Fixing Code

  The article argues that software teams should stop trying to correct AI-generated code line by line and instead improve the surrounding system, context, and workflows. It frames this as a broader shift from deterministic software development toward context engineering, where organizations package and scale the information agents need to work effectively.

• 5The New Stack·7/19/2026research
  AI Is Pushing Code Review Upstream

  The article argues that traditional pull-request-based code review is becoming a poor fit for AI-assisted software development, where large amounts of code are generated quickly. It suggests teams should review intent, specs, and acceptance criteria earlier in the workflow rather than waiting until merge time.

• 5InfoQ·7/14/2026framework_update
  Building a Context Store for AI-Aware Software Evolution

  The article argues that AI-assisted development can accelerate the first part of coding while obscuring architectural complexity until later stages. It proposes a repo-bound Context Store that combines spec-anchored SDD, TDD, and automated fitness functions to help human reviewers and AI agents evolve systems more safely.

• 6InfoQ·7/10/2026product_launch
  Datadog Uses Claude and Cursor to Migrate a Production System

  Datadog engineer Arnold Wakim described how the team used Claude and Cursor to help migrate a critical production system while addressing storage backend limits and improving performance. The article focuses on what worked, what failed, and the lessons learned from using AI-assisted development in a test-driven migration process.

• 4Hacker News·6/28/2026research
  Kent Beck Reframes YAGNI as a Timing Principle

  Kent Beck argues that YAGNI is not mainly about saving coding effort, but about avoiding speculative structure built before a feature is actually needed. He frames the principle around two costs: lost optionality from premature commitment and negative net present value from paying structure costs too early.

• 4Martin Fowler·4/14/2026research
  Fowler on AI, abstraction, and programmer laziness

  Martin Fowler reflects on a Pragmatic Summit discussion with Kent Beck and Gergely Orosz about AI, earlier technology shifts, agile methods, TDD, and performance metrics. He argues that human constraints and “lazy” abstraction still matter, and that LLMs can encourage overcomplicated systems unless developers apply judgment such as YAGNI and TDD-style checks.

• 4Martin Fowler·4/2/2026research
  LLMs, cognitive debt, and verification in software development

  Martin Fowler highlights a framework that separates system health into technical debt, cognitive debt, and intent debt, arguing that each limits a different aspect of software evolution. He also discusses research framing AI as a “System 3” in cognition, and notes that as coding agents reduce the cost of writing code, verification becomes the more valuable and difficult work.