amit-bendor talks with ohad-parush, Chief R&D Officer at Gong, about deliberately embedding AI into development. Copilot drew lukewarm reactions (“autocomplete on steroids”); Claude Code was the revolution, run in a Docker container per Anthropic’s security guidance, with no token or session caps. Practices: the Developer Experience (DX) team owns AI adoption (environments, manifests, practices); mandatory AI review prompts on every PR, always with a human override (“AI for Quality”); hands-on training on developers’ own code; and a top-down mandate on one new product where an all-senior team had to use Claude Code, and got hooked. Results: unfamiliar tech (Apache Iceberg, WebSockets) in two days instead of two months; tests first, so TDD gets a revival; design docs synced back to Confluence from code. Gong still hires aggressively: “200 developers with the impact of 300.” New role “DS2” (data scientist 2.0): prompt templates (machines write the prompts), evals, judges, gold sets, and continuous calibration (CI/CC/CD). Token usage as a metric is interesting but shallow.
Show: osim-tochna
Concepts: ai-rd-rollout · ai-engineering-metrics · test-driven-development · llm-evals · ai-verification · future-of-software-engineering