AI Radar Dominates but Demands Foundations and Safeguards
Thoughtworks' 34th Tech Radar (118 blips) spotlights AI trends like agent security and harness engineering, while urging return to basics like pair programming and clean code to counter AI-generated complexity.
Revisit Foundations to Counter AI Complexity
AI tools accelerate development but generate unchecked complexity, so pair them with established practices: use pair programming, zero trust architecture, mutation testing, DORA metrics, clean code, deliberate design, testability, and accessibility as core concerns. Command lines resurge as agentic tools make terminals the primary interface, reversing years of abstraction for usability. This isn't nostalgia—it's essential to balance AI speed; without it, tools produce bloat like 50KB single files (2,000 lines) in a 100KB codebase, where even capable LLMs like Claude resort to sed edits instead of refactoring.
Secure 'permission-hungry' agents needing broad access to private data, external comms, and systems (e.g., OpenClaw, Claude Coworker, Gas Town). Safeguards lag: prompt injection lets untrusted input override instructions. Build 'harness engineering' with guides and sensors for safe delegation—expect more blips on this in six months.
Human Oversight Essential for Durable AI Code
AI-generated code can pass unit tests and handle real workloads but hides architecture mixes good design with 'incomprehensible mess'—you must read it to know. Claude Code's 500,000-line leak exemplifies this duality. For throwaway analysis scripts, let AI 'vibe away'; for maintainable tooling or durable code, enforce regular human review. Prompt the model for evaluation using hints on good code traits.
When scale discomfort hits (e.g., 'this file is too big'), AI decomposes sensibly into classes and adds tests—but won't volunteer it. Use CLAUDE.md seriously for guidance; combine with patterns like Rahul Garg's to break frustration loops in iterative edits.
Organizational Lessons from Tech Failures
Simple reforms hide deceptive complexity, blocking implementation in governments or corporations—e.g., DOGE axed DirectFile, a free IRS online tax filing tool, despite public service ethos. Contrast with DOGE's disinterest in users. U.S. IRS now down 25% staff and 40% budget vs. 2010, weakening enforcement; boosting funding pays for itself via revenue (Yale Budget Lab). Efficient taxes underpin security, as Britain's 18th-century edge over France showed—wonky systems invite revolution.