The era of developers constantly reviewing AI-generated code for errors is fading. Advances in how teams architect AI agents are now enabling machines to catch and fix their own mistakes before human eyes ever see the work, reducing the need for constant supervision.
Vanitha Kumar, Head of Technology for Retail, CPG, Travel & Hospitality Market at Thoughtworks Americas, describes this shift in a recent podcast appearance. The mechanism behind it is harness engineering—a practice that wraps agents with automated tests, feedback mechanisms, and safety constraints that let them self-correct. According to Kumar, the approach has two dimensions: "Do it right the first time and also self-correct when you have done it for the first time."
https://www.youtube.com/embed/EaNOeUq56SA?si=03y3Wh6j32ztymEh
The result is that teams no longer need to "babysit and watch your agent do the work," Kumar explains. Agents can handle extended operations with minimal intervention, and multiple agents can operate in parallel without interfering with one another.
From Single Agents to Factory Scale
Running one autonomous agent is straightforward. Deploying hundreds across an organization introduces a different set of challenges. This is where AI factories enter the picture. Kumar draws a parallel to Kubernetes, which abstracted away the operational complexity of managing microservices at scale. AI factories serve a similar role for agent-based systems, establishing standardized infrastructure for model selection, workload distribution, and expense management. As Kumar frames it, "The best way to build a platform is [to] make it easy to do the right thing."
https://player.simplecast.com/a2f41f5e-454d-40f8-8f86-7326dfbe7a3e?dark=true
Security Becomes the Trade-off
Granting agents greater autonomy means granting them access to more data sources, systems, and tools—each a potential attack surface. Some organizations are hesitating to adopt AI at scale specifically to avoid expanding their security footprint.
Harness guardrails can help mitigate these risks by limiting which tools agents can reach, isolating changes to test environments, and requiring human sign-off before agents modify sensitive systems.
When asked about the technology landscape ahead, Kumar is direct: "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, with a focus on AI, harness engineering, and emerging agent practices. The report also highlights established engineering disciplines that are suddenly critical again. Penetration testing and path-to-production understanding are examples Kumar points to as "old things that are making a comeback because they now have more added weight because of AI."