AI Agents Conference · December 1–3, 2026
S Ravi Kumar
Solutions Architect at Noventiq
Session
Lessons from a Year of Building Software with AI Coding Agents
For a year I have built real software with AI coding agents, and for most of that year I was stuck in one loop: the agent builds, something is wrong, I point at it, it fixes, something else is wrong. This session is what I learned getting out of that loop. Why an agent agrees with a rule written as a sentence and then does not follow it, and what to write so that it does. Why the agent that builds should never be the one that says the work is done. Why the screen should be designed before the code, and how that design becomes the test. How to write a requirement an agent cannot misread. And how to choose which model does which job. Each lesson comes with the mistake that taught it to me and with what I changed. You can apply all of them with whatever coding agent you use today.
Share your speaker card
Speakers: download this card and post it on LinkedIn, X or Instagram. Links to this page preview with the same design.
Full conference agenda
See every session across the AI Agents Conference December 2026 line-up.