From Intent to Merge: A Hands-On Talk on SDLC in the Agent Era
summary
Live demo of Entire shows an agent touching 47 files touched, inventing three abstractions, briefly considering a Rust rewrite, passing tests and disappearing; the platform binds AI checkpoints to commits in a hidden .checkpoints branch so every prompt, transcript and tool call is versioned alongside the diff. Semantic graphs cut search cost (claimed 70% token savings), seven agents are supported, and pull requests become trails (reimagined pull requests) with programmable JSON runners, gates, confidence metrics and optional automerge.
AI-written summary · about
chapters (18)
- 0:00Introduction and setup
- 1:38AI agents and developer productivity
- 3:24Introducing Entire platform
- 4:28Semantic layer demo
- 7:42Repository memory and checkpoints
- 10:40Agent agnostic and open source
- 12:29Code blame and intent tracking
- 13:34Graph-based code search
- 15:19Why-based codebase queries
- 16:46Distribution and Git hosting
- 18:30Pull request challenges in AI era
- 19:56Trails reimagined pull requests
- 23:10Runners and automation gates
- 24:57Self-evolving software demo
- 26:42Confidence and risk assessment
- 28:40Slope runner and cleanup
- 29:24Summary and future vision
- 31:35Call to action and resources
