From autocomplete to agents: AI coding assistance state of play
summary
Overview of current AI coding assistance: autocomplete and IDE chat evolve into supervised agentic modes that can read/change files, run tests and web-search (examples: GitHub Copilot, Cursor, Windsurf) and pair with MCP servers for local tools and project context. Measured gains shift from GitHub’s 55% per-task claim (≈13% cycle-time) to teams reporting up to 80% faster on specific tasks, while telemetry shows more churn, duplicated code and worse refactoring. Risks include the rules file backdoor, leaking MCP servers, brittle tests and maintainability regressions; recommended controls are checkpoints, review, and tightened CI/static checks.
AI-written summary · about
chapters (14)
- 0:00Introduction and evolution
- 1:59Features of coding assistants
- 4:44Measuring productivity
- 7:52The rise of AI agents
- 11:28Agent capabilities
- 15:17Model Context Protocol
- 18:17Configuring agents
- 21:17Effective AI workflows
- 24:17Optimizing for AI
- 26:53The 'vibe coding' debate
- 30:51Lessons from experience
- 37:14Risks to code quality
- 40:59Recommendations for success
- 45:30Q&A session
related talks
26mFrom autocomplete to agents: AI coding assistance state of play
50mAI Coding Assistance: The State of Play
27mAgentic Engineering: Working With AI, Not Just Using It
26mAI Agents as Your Co-Developer
45mAI Assistance for Software Teams: The State of Play
30mHow AI Agents Will Change the Future of Programming
29mPairing with the machine: How AI Agents are rewriting the rules of software development
25mFrom AI Chaos to Production Ready: Using AI Agents to Improve Your Code Quality