From Copilots to Agentic SE Teammates: Engineering the SE 3.0 Era with AI-Native...
summary
Argues a shift to Agentic SE 3.0 with concrete stack pieces: Teammate.next, IDE.next, Compiler.next, Runtime.next. Presents AIDev — a 61K+ repo dataset showing millions of agent-generated PRs, many accepted but ~50% of agent PRs go unacted, and specific failure modes like oversized diffs and obsolete changes. Describes an FM-native IDE (FMRS) with conversational, versioned intent, a layered local+cluster model runtime that teaches small models, and Ray-optimized scheduling for heterogeneous GPU/CPU workloads to meet SLA-aware agent execution.
AI-written summary · about
chapters (12)
- 0:00AI-native software engineering intro
- 1:28Codeware vs AI-native software
- 6:21SE 3.0's four pillars
- 17:43Rethinking SE processes
- 23:20Human-AI collaboration benefits
- 27:41Programming 3.0 and IDEs
- 35:37Agentic dev demo
- 39:33Agents in open-source
- 43:35Runtime for AI-native apps
- 47:52Intent-driven development
- 48:58AI-native education challenges
- 50:04Q&A: Maintenance in AI-native
related talks
18mCollaborative AI Engineering: One Dev, Two Dozen Agents, Zero Alignment
27mAgentic Engineering: Working With AI, Not Just Using It
16mBeyond the IDE: The Future of Developing Software With Agents
29mPairing with the machine: How AI Agents are rewriting the rules of software development
14mThe 4 Patterns of AI Native Development
28mAgentic AI
14mAgent-Driven Software Development
35mThe Agentic AI Engineer