The 34th Technology Radar: Navigating AI, Security, and Harness Engineering

Introduction

Every six months, ThoughtWorks releases its Technology Radar, a comprehensive survey of the technology landscape based on firsthand experience. The 34th edition, just published, features 118 blips covering tools, techniques, platforms, and languages that have caught the team’s attention. While AI continues to dominate the radar, this volume also highlights a surprising countertrend: a renewed focus on foundational software practices and the growing importance of security and governance in an agent-driven world.

AI Dominance and Revisiting Foundations

Unsurprisingly, many of this edition’s blips center on artificial intelligence, particularly large language models (LLMs). But instead of simply chasing the next breakthrough, the radar reveals something deeper: AI is forcing the industry to look both forward and backward. As the authors note, “part of this is revisiting familiar ground with LLM-assisted eyes.” This dual perspective is reshaping how teams approach software development.

The Resurgence of the Command Line

One of the most interesting developments is the renewed popularity of the command line. After years of graphical interfaces abstracting away the terminal, agentic tools are bringing developers back to the command line as a primary interface. This shift reflects a desire for precision and control in an era where AI can generate rapid, complex outputs.

Software Craftsmanship Revisited

The radar also emphasizes a return to core principles: clean code, deliberate design, testability, and accessibility as a first-class concern. Practices like pair programming, zero trust architecture, mutation testing, and DORA metrics are being reexamined not out of nostalgia, but as a necessary counterweight to the speed at which AI can introduce complexity. This balancing act ensures that innovation doesn’t come at the cost of maintainability and security.

Security Concerns with Permission-Hungry Agents

A major theme of this radar is the security challenges posed by “permission-hungry” agents. These are AI-driven tools that require broad access to private data, external communication, and real systems to deliver value. Examples include OpenClaw, Claude Cowork, and Gas Town—agents that supervise work tasks or coordinate entire codebases.

The Bind of Access vs. Safeguards

The problem, as the radar describes, is that “the safeguards haven’t caught up with that ambition.” Like a skier who learns to turn and immediately points at the hardest black run, these agents operate with a dangerous gap between access and protection. Issues like prompt injection mean models still cannot reliably distinguish trusted instructions from untrusted input. To address this, the radar includes a strong focus on security, with Jim Gumbley joining the writing team to bring expertise from threat modeling and other areas.

Harness Engineering: A Growing Focus

Given the security and complexity concerns, many blips in this edition are dedicated to harness engineering—the practice of designing guides and sensors that allow AI agents to operate safely and effectively. The radar meeting itself was a major source of ideas for Birgitta’s article on the subject, and the resulting blips suggest specific tools and techniques for building a well-fitting harness.

Building Guides and Sensors

These blips cover the components needed to monitor and constrain AI behavior: from guardrails to observability tools, and from policy engines to audit trails. The authors expect that when the next radar appears in six months, the list of harness-related items will have grown significantly. This forward-looking emphasis underscores the industry’s recognition that unchecked AI can create more problems than it solves.

Conclusion

The 34th Technology Radar paints a picture of an industry at a crossroads. AI is accelerating development, but it also demands a renewed commitment to tried-and-true practices, security rigor, and thoughtful governance. By balancing the new with the foundational, ThoughtWorks provides a roadmap for teams navigating this complex landscape.