CHICAGO, Nov. 4, 2026 — Thoughtworks, a global technology consultancy, has released volume 35 of its Technology Radar, emphasizing the necessity for enhanced controls as AI agents take on larger and more complex tasks. The report notes that significant engineering challenges now extend beyond merely interacting with AI models, highlighting crucial areas such as securing operational environments, managing access, ensuring human oversight, and controlling operational costs.
The Technology Radar features 11 recommended practices and technologies placed in its Adopt ring, which include strategies like agent sandboxing and threat modeling. These recommendations indicate the growing importance of safeguards necessary for the reliable deployment of AI systems in production settings.
According to Rachel Laycock, Chief Technology Officer at Thoughtworks, “Not long ago, the question we heard from clients was how fast agents could write code. Now it’s whether anyone can trust, review and afford what they produce.” She emphasized that while AI models will continue to improve, lasting benefits will come to organizations that invest in comprehensive engineering and oversight frameworks.
The current edition shifts focus from the expanding capabilities of AI—previously noted in volume 34—to managing emerging risks as AI adoption increases. Key themes explored include the systems that guide agent behavior, the role of human judgment, and the infrastructure that impacts cost and reliability.
The report discusses several critical aspects of managing AI effectively:
- Harnessing engineering at scale: An AI agent’s operational environment, encompassing prompts, tools, memory, guardrails, and feedback loops, is increasingly viewed as a separate product that requires its own engineering strategy, including security and maintenance.
- Maintaining human judgment: As developers run multiple agent sessions simultaneously and delegate more tasks, the risk grows that human review processes could become bottlenecks. The report cautions against superficial approvals and encourages a careful assessment of team capacities before increasing agent use.
- Choosing the right models: The Radar points out that proprietary frontier models should not be automatically selected for every task. Open-weight models have become viable alternatives, offering flexibility regarding deployment based on several operational factors.
- The economics of AI: The costs associated with AI deployment extend beyond model pricing to include various operational elements like routing and storage. The report calls for innovation to ensure the economic sustainability of AI strategies.
These findings indicate a significant shift in enterprise AI. To unlock its full potential, organizations must develop engineering practices, infrastructure, and effective human oversight to manage AI agents responsibly and at scale.
For more insights, the interactive version of the Radar and a downloadable PDF are available at www.thoughtworks.com/radar.
