Why Faster Code Doesn't Always Raise AI Developer Productivity

Sunita Verma has been building AI since before anyone called it AI, and her warning to engineering leaders today is about what happens after the code gets written. Speeding up one step of development, she says, just hands the bottleneck to the next one.

Sunita, CTO of Ironclad, spent years at Google and Character AI before leading AI strategy in legal tech, a heavily regulated field. On Keep Moving Forward, she connects AI burnout to deployment choices and makes the case for spelling out what a model shouldn't do before it does anything. She also shares how Ironclad keeps its engineers moving in the same direction while the tools change underneath them.

They cover:
- When to hand a task to AI outright, and when staying in the back-and-forth creates more work
- What agentic coding does to code review when nobody redesigns the review process
- How Ironclad decides when a human stays in the loop, using a risk-based framework that lets standard work like NDAs run autonomously
- Why Sunita ran a 20-day AI program across every location to give a distributed team a shared vocabulary
- What Ironclad saw after pairing its most AI-forward engineers with teammates further behind the curve