Thoughtworks has identified growing challenges as AI agents scale, including securing operational environments, managing access, ensuring effective human oversight, and controlling operational costs.
In its latest Technology Radar report, the consultancy notes that as AI agents handle larger and longer-running tasks, engineering challenges increasingly extend beyond the models themselves.
“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,” said Rachel Laycock, chief technology officer at Thoughtworks. “Models will keep improving, but lasting value will go to the organisations that invest in the engineering and oversight around them.”
Building and managing systems around AI models
Volume 35 of Thoughtworks’ Technology Radar explores what organisations must build and manage as AI agent adoption grows. Key themes include:
- Engineering AI agent environments: Organisations should manage prompts, tools, memory, and security controls as engineered systems, with ongoing evaluation and maintenance.
- Scaling AI agents responsibly: Businesses are deploying multiple agents for complex tasks, but must ensure human oversight and review processes remain effective.
- Looking beyond frontier models: Open-weight models are increasingly viable, enabling organisations to choose AI models based on performance, cost, privacy, and deployment requirements.
- Managing AI operating costs: Scaling AI requires optimising not only model costs but also computing, storage, networking, and infrastructure resources to ensure sustainability.
The report also emphasises the increasing importance of safeguards for production AI systems. Its 11 recommendations in the Adopt ring include practices such as agent sandboxing and threat modelling.
The findings indicate a broader shift in enterprise AI, where organisations must invest in both advanced models and the engineering, infrastructure, and oversight required for reliable large-scale deployment.
