Harness Engineering and AI Factories Let Developers Step Back From Constant Oversight
As autonomous AI agents become more capable, Thoughtworks' Vanitha Kumar discusses how harness engineering and AI factories enable organizations to scale—while security concerns mount.
The evolution of AI agent technology has reached a turning point. Where early systems demanded relentless human supervision to catch errors and hallucinations, newer approaches are enabling agents to identify and fix their own mistakes before developers ever see the work.
Vanitha Kumar, Head of Technology for Retail, CPG, Travel & Hospitality Market for Thoughtworks Americas, describes this shift as a fundamental change in how teams can deploy AI at scale. Through what the industry calls harness engineering, organizations equip their agents with automated testing, feedback mechanisms, and safety constraints that allow self-correction to happen automatically.
Kumar explains the dual nature of this approach: "Do it right the first time and also self-correct when you have done it for the first time." The result is that developers no longer need to "babysit and watch your agent do the work."
https://www.youtube.com/embed/EaNOeUq56SA?si=03y3Wh6j32ztymEh
Harness engineering accomplishes several things at once. Agents can execute longer-running tasks with minimal human intervention. Multiple agents can operate in parallel without interfering with each other's operations. And the entire process becomes more predictable and manageable.
From Single Agents to Organizational Scale
Managing one autonomous agent differs vastly from orchestrating hundreds across an enterprise. This is where AI factories enter the picture. Kumar draws a parallel to Kubernetes, the container orchestration platform that abstracted away infrastructure complexity for microservices. AI factories serve a similar function for agent-based systems.
https://player.simplecast.com/a2f41f5e-454d-40f8-8f86-7326dfbe7a3e?dark=true
These platforms establish standardized infrastructure for model selection, workload distribution, and cost management. Kumar frames the philosophy behind them simply: "The best way to build a platform is [to] make it easy to do the right thing."
Security Emerges as a Central Challenge
Greater autonomy for agents comes with a tradeoff. Agents require access to more data sources, systems, and tools to function effectively. This expanded access surface creates new security risks that some organizations find concerning enough to delay their AI adoption entirely.
Harness guardrails offer one path forward. These mechanisms can limit agent access to specific tools, confine operations to test environments, and require human approval before agents make sensitive changes to production systems.
When asked about emerging trends, Kumar is direct about what lies ahead: "Security is going to be a big topic. Not even one year out."
Old Practices Return With New Urgency
Thoughtworks released its Technology Radar, Volume 35, on November 4, 2026. The report highlights AI, harness engineering, and agent practices as key areas of focus. But it also flags something unexpected: established engineering disciplines that are suddenly more critical.
Penetration testing and understanding the path-to-production both appear prominently in the report. Kumar explains the reasoning: "… old things that are making a comeback because they now have more added weight because of AI."