How do you use AI to accelerate software development—without losing human understanding, learning, and responsibility?
In this experience report, Song Tao and Terry Yin share hands-on lessons from a real AI adoption journey inside a 20-year-old legacy system with heavy cross-team dependencies.
Rather than treating AI as a replacement for thinking, they explore a human-centric, AI-augmented approach—where people stay accountable for intent, learning, and design, and AI is used deliberately where it adds leverage.
Key ideas from this talk:
Why requirements, domain understanding, and design must remain human-led
The hidden risk of AI: cognitive debt (code works, but no one truly understands it)
Human-AI partnership instead of “vibe coding”
A continuous loop: Context → Generate → Automated Test → Learn
Using autonomation (Jidoka) instead of blind automation
Why end-to-end tests provide better AI feedback than unit tests in legacy systems
Treating automated tests as both safety nets and living documentation
Why AI does not create a competitive moat—and what actually does
Shifting focus from “making progress” to seeking high-quality feedback
This talk challenges the idea that faster code is always better—and argues that shared understanding, feedback loops, and human learning are the real constraints in AI-augmented development.
📍 Recorded at LeSS Conference 2026 Amsterdam
🎤 Speakers: Song Tao & Terry Yin
Download the presentation here: https://less.works/conferenza/sessions/2025-global-less-conference-amsterdam-human-centric-software-development-augmented-by-ai-457
