When Code Became Cheap: Deep Tech Night Panel Materials (Category #AI4SDLC)
I’ve collected the recording and recap of the panel “When Code Became Cheap: Where Are Value, Responsibility, and Expertise Now?”, which I joined at Yandex Deep Tech Night on September 5, 2026, alongside Alexander Lukyanchenko from Avito, Alexander Mazko from Sber, and Oleg Smolyakov from Yandex. If an agent writes code faster, why don’t useful changes reach users at the same pace? That question led us to the work surrounding the code.
We discussed
- The value of faster development. More briefs, requirements, and code can still leave the whole process waiting on approvals and context handoffs.
- Permissions and accountability. Which actions to entrust to an agent, where isolation and checks are needed, and who will investigate an incident after the tests pass.
- Junior education. Should training tasks restrict agents, or teach people to work with them from the start? Opinions differed. I suggested checking whether the learner understands the architecture and the reasons behind decisions.
- The cost of attention. Several parallel agents demand switching and oversight. We discussed how to assess outcomes while accounting for the engineer’s workload.
Panel materials:
📌 Panel page 📖 My positions before the panel — the preparation longread. 🎬 YouTube recording — 1 hour 6 minutes. 📝 Edited text recap
What became your main bottleneck once writing code got faster?
#AI4SDLC #AI #Agents #Engineering #Management