AI Coding Assistance: The State of Play
summary
Survey of current AI coding tools names GitHub Copilot, Cursor, Windsurf, Ader, Klein and shows agentic modes that read/change multiple files, run tests, and do web research via MCP servers (Playwright example). Measured effects are modest: typical cycle‑time wins cluster around ~8–13% from autocomplete-era heuristics though agentic tasks can push task speed higher; GitHub’s 55% claim is misleading. Real outcomes include faster fixes but new bugs, security risks (custom‑rules backdoor), and rising churn in git history; beware vibe coding hype.
AI-written summary · about
chapters (10)
- 0:00Introduction and role
- 1:03History of AI coding assistance
- 7:48Agentic modes in coding
- 14:54Model Context Protocol
- 18:00Custom instructions and rules
- 21:50Emerging ways of working
- 26:07Vibe coding and its limits
- 33:11Developer skills still matter
- 40:04Long-term maintainability risks
- 44:03Final thoughts and conclusion
related talks
49mFrom autocomplete to agents: AI coding assistance state of play
26mFrom autocomplete to agents: AI coding assistance state of play
26mCoding with AI
27mAI for software development: A reality check
30mThe AI Revolution in Coding
33mAI assisted software delivery
51mAI-Assisted Coding: The Future; Between Challenges And Benefits
26mAI Agents as Your Co-Developer