Thoughtworks Technology Radar Vol 35 highlights AI agent harness engineering and infrastructure

Details

Thoughtworks Technology Radar Volume 35 (November 2026) defines four core themes for AI development: harness engineering, moving beyond frontier models, deploying multiple agents, and constructing AI factories.
Harness engineering is recognized as a maturing field for building AI agent environments, with the emergence of plugin ecosystems and the expectation that 'enterprise harness' products will appear in the coming months.
Open-weight models are increasingly viable alternatives to exclusive frontier models, reducing vendor dependency and enabling self-hosting; the report advises caution against blanket frontier model usage in favor of model routing.
Multi-agent workflows are shifting from single coding agents to parallel sessions and 'coding agent swarms,' though developer attention and review capacity remain the primary bottlenecks.
AI infrastructure is evolving into an 'AI factory' where token economics drive infrastructure economics, with Kubernetes-based platforms like KServe, KAITO, llm-d, and LiteLLM coordinating GPUs, inference engines, and storage as programmable resources.

Key point