Post by Thoughtworks

Chris Ford, Market Tech Director at Thoughtworks, draws a sharp parallel: manually reviewing code from coding agents is as awkward as Level 3 self-driving cars that expect drivers to step in when the situation on the road gets tricky. Why? Because attention fades when a person is tasked with monitoring a process they aren't actively participating in. Relying on manual inspection as use of coding agents scales results in "accountability laundering", approving code without truly understanding it, then we point at the person when something breaks. So, what’s the alternative? 🗣️Interrogate the harness, not the code. Ask the agent: "How do you know this works?" 🎯Focus on the interface points. Specify the architecture and let AI fill in the details. ✅Invest in your pipeline. Continuous delivery is your ultimate safety net. If your pipeline can’t validate and deploy without human intervention, you’re not ready for AI. Want to dive deeper into this topic? 🎙️ Catch Thoughtworks’ Chris Ford unpacking these insights as a guest on the full episode of Engineering with AI: https://bit.ly/4movYEY