The release of Thoughtworks’ 34th Technology Radar has sparked fresh conversations across the software development community. This biannual report distills our direct experience with hundreds of tools, techniques, platforms, and languages—118 specific “blips” this time. Rather than a dry inventory, the radar captures what we’ve actually found useful, risky, or worth watching. While artificial intelligence dominates the headlines, the real story is how AI is forcing us to reexamine the foundations of software craftsmanship, security, and team dynamics. Below are ten essential takeaways from this edition, each offering a practical lens for navigating today’s fast-moving tech landscape.
1. AI Takes Center Stage—But Not As You’d Expect
Unsurprisingly, the radar is packed with AI-oriented blips. However, the most striking pattern isn’t about new models or flashy demos; it’s how AI is driving us back to core engineering principles. Large language models (LLMs) are everywhere, but their real impact lies in the way they challenge us to think more carefully about clean code, deliberate design, and testability. The speed at which AI can generate complexity demands a counterbalance grounded in timeless practices. This edition’s AI coverage isn’t a celebration of novelty—it’s a call to maturity.
2. Revisiting the Foundations of Our Craft
A fascinating consequence of AI adoption is that it compels us to look backward as much as forward. While assembling this volume, the team found themselves returning to established techniques such as pair programming, zero trust architecture, mutation testing, and DORA metrics. These aren’t relics—they’re essential tools for taming AI-induced complexity. The radar highlights that accessibility as a first-class concern and software craftsmanship are more relevant than ever. This isn’t nostalgia; it’s a strategic recalibration to ensure quality keeps pace with innovation.
3. The Command Line Makes a Comeback
After years of being abstracted away by graphical interfaces and IDEs, the command line is experiencing a renaissance—driven by agentic AI tools. These “permission-hungry” agents prefer the terminal as their primary interface because it offers direct, programmable control. For developers, this means dusting off shell scripting skills and embracing terminal-based workflows again. The radar notes that this shift brings both efficiency and new security considerations, as agents executing commands need careful oversight.
4. Strengthening Security Expertise on the Radar Team
I’m particularly pleased that Jim Gumbley has joined the radar writing team. A long-time colleague and a trusted voice on security, Jim has contributed to this site’s Threat Modeling Guide and consistently brings a rigorous security perspective. Given the serious concerns around LLM usage—such as data leakage and prompt injection—having a strong security presence is critical. Jim’s insights help ensure that every blip addressing AI also accounts for the associated risks.
5. The “Permission-Hungry” Agent Dilemma
One of the radar’s central themes is the tension at the heart of agent-based AI: the agents worth building are exactly the ones that demand expansive access. Tools like OpenClaw and Claude Cowork supervise real work tasks; Gas Town coordinates agent swarms across entire codebases. These agents need access to private data, external communication systems, and production environments. Each use case argues that the payoff justifies the risk, but safeguards haven’t caught up. This is the “permission-hungry” bind—ambition outpacing protective measures.
6. Prompt Injection: An Unsolved Security Challenge
A direct consequence of giving agents broad access is the renewed urgency around prompt injection. The radar highlights that models still cannot reliably distinguish trusted instructions from untrusted input. This vulnerability opens the door to malicious manipulation, especially when agents have permission to execute commands or access sensitive data. Addressing prompt injection requires not just better models, but stronger architectural boundaries—something the radar flags as a critical area for further research.
7. Harness Engineering: The Framework for Safe AI
Given the security concerns, many of this edition’s blips focus on harness engineering—the discipline of designing the “guides and sensors” that keep AI systems in check. The radar meeting itself was a rich source of ideas for Birgitta’s excellent article on the subject. Harness engineering encompasses monitoring, feedback loops, and governance structures that ensure agents operate within safe parameters. The current list of recommended practices is substantial, and we expect it to grow significantly by the next edition.
8. Safeguards Must Catch Up with Ambition
The radar likens the current state of AI agents to “a skier who’s just learned to turn and confidently points themselves at the hardest black run.” The ambition is admirable, but the safeguards are lagging. Many organizations deploy agents with broad access while their security postures remain immature. The radar calls for deliberate investment in guardrails—such as rate limiting, behavioral monitoring, and least-privilege principles—before scaling agent usage in production environments.
9. The Role of Telemetry and Observability
To build effective harnesses, teams need robust telemetry. The radar includes several blips promoting observability as a non-negotiable component of any AI-integrated system. Metrics like latency, error rates, and user feedback must be collected and analyzed in real time. The resurgence of the command line also ties into this: shell-based workflows often produce cleaner logs that are easier to aggregate. Without visibility into agent behavior, you cannot govern it.
10. Looking Ahead: Six Months to a Richer Radar
The radar concludes with a forward-looking note: many of this edition’s blips are starting points, not final answers. Harness engineering, prompt injection defense, and agent governance are all areas where we expect rapid evolution. By the time the next radar appears in six months, we anticipate a longer list of proven techniques and tools. For now, the key message is to experiment carefully, invest in foundational skills, and never lose sight of security.
In summary, Thoughtworks’ 34th Technology Radar is not just a snapshot of current trends—it’s a strategic guide for balancing innovation with discipline. Whether you’re adopting AI agents, revisiting pair programming, or tightening your security posture, these ten insights provide a practical roadmap. The radar reminds us that the best way to navigate a future shaped by AI is to anchor ourselves in the principles that have always made software great.