Software Development Agents: What Works and What Doesn't
summary
Demonstrates concrete workflows using OpenHands (fka OpenDevin): agents run inside a Docker container, use a diff-based editor and accessibility trees/screenshots for browsing, and can resolve merge conflicts about 99% of the time. Claims 90% of routine code flows through agents for power users, prototypes succeed ~50% and the team doubled web-browsing accuracy with a recent contribution. Best early tasks are single-commit chores: PR feedback, failing tests, infra tweaks, DB migrations and small Greenfield internal apps.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 0:44Software development is changing
- 1:50What is a coding agent
- 3:37How coding agents work
- 7:26Best practices
- 12:29Use cases
related talks
30mBuilding Tools with Tools: Modern AI-Driven Development
26mAI Agents as Your Co-Developer
45mAI agents in practice - beyond the prototype
01h 16mAutomating Large Scale Refactors with Parallel Agents
23mManaging fleets of coding agents with OpenHands
14mAgent-Driven Software Development
27mAgentic Engineering: Working With AI, Not Just Using It
50mPanel: Beyond The Vibe: Code Quality First