Navigating the AI Era: Insights from ThoughtWorks' 34th Technology Radar
The 34th Technology Radar: Key Themes
ThoughtWorks has released the 34th edition of its Technology Radar, a biannual report that captures the collective experience of the company's technologists across the globe. This edition features 118 "blips"—brief assessments of tools, techniques, platforms, and languages that have been observed in real-world projects or have piqued interest. The radar serves as a barometer for the evolving technology landscape, and this volume is heavily influenced by the rapid integration of artificial intelligence into software development.
AI Dominance and a Return to Foundations
Unsurprisingly, AI-related topics dominate the latest radar. However, the report does not simply focus on cutting-edge LLMs and agentic systems. An interesting consequence of AI's rise is that it is prompting teams to revisit established practices. Many blips point back to core software craftsmanship principles: clean code, deliberate design, testability, and accessibility as a first-class concern. The radar also highlights a renewed interest in techniques like pair programming, zero trust architecture, mutation testing, and DORA metrics. This is not nostalgia but a necessary counterweight to the complexity that AI tools can generate at speed. After years of abstracting away the command line in favor of graphical interfaces, agentic tools are bringing developers back to the terminal as a primary interface.
Security Concerns and 'Permission Hungry' Agents
A strong security presence on the radar team is more important than ever, given the serious risks associated with large language models. One of the key themes is the challenge of securing what the report calls "permission hungry" agents. These are agents—like OpenClaw, Claude Cowork, and Gas Town—that require broad access to private data, external communication, and production systems to deliver value. The bind is that the most useful agents are those that need access to everything. Yet the safeguards have not kept up with that ambition. Issues such as prompt injection remain unsolved, meaning models still cannot reliably distinguish trusted instructions from untrusted input. This theme underscores the need for robust security frameworks as AI agents become more autonomous.
Harness Engineering Takes Center Stage
Given the security concerns and the need to control AI-driven complexity, the radar dedicates many blips to the concept of "Harness Engineering." This term refers to the set of guides, sensors, and controls that ensure AI systems operate within safe and effective boundaries. The radar meeting itself generated ideas that Birgitta expanded into an article on the subject. Several blips suggest specific techniques and tools for building a well-fitting harness—for example, methods to monitor agent behavior, enforce access controls, and log decisions. The expectation is that the next radar, six months from now, will include even more entries in this category as the industry develops better practices for governing AI agents.
For more on the background of the radar and its team, see the introduction above. The interplay between AI's promise and its perils is a thread that runs throughout this edition, making it essential reading for anyone involved in modern software development.