Emerging Patterns for Coding with Generative AI
summary
Maps 32 concrete patterns and shows how to make agentic coding reliable: treat LLMs as stateless context windows, extract work into a knowledge document and ground rules, use git worktrees for parallel implementations, and write deterministic scripts (Puppeteer) so agents stop hallucinating. A committer agent (Claude/Claude Code) actually flags wrong files; refinement loops raise first-pass quality to ~90–95%. Built a ricochet-robot/counter app in a 3‑day hackathon, rolling the dice three times and combining the best outputs to succeed.
AI-written summary · about
chapters (47)
- 0:00Introduction to coding with AI
- 1:15Coding genie analogy
- 3:00What we can control with AI
- 4:48Three tracks of patterns
- 5:51Context management basics
- 10:07Context rot and its impact
- 10:27Knowledge document pattern
- 12:13Ground rules for AI
- 13:37Extracting knowledge from AI
- 16:04Context window limitations
- 17:54Focused agents and specialist behavior
- 20:23Granular context control
- 23:12Knowledge composition and references
- 24:36AI noise and verbosity issues
- 26:23Semantic zoom technique
- 27:27Noise cancellation in AI responses
- 31:16Semantic zoom in tests
- 33:23High-level debugging with AI
- 33:45First Q&A break
- 42:36Nondeterminism and reliability
- 42:56Knowledge checkpoint pattern
- 45:48Parallel implementations with AI
- 46:09Offloading determinism to AI
- 49:01Low determinism pattern
- 51:07AI hallucinations and playgrounds
- 55:40Unvalidated leaps and slow validation
- 57:04Predictive TDD process
- 58:06Chains of small steps
- 1:00:54Selective hearing and stubborn instructions
- 1:03:25Using hooks to enforce rules
- 1:05:33Focusing agents with reminders
- 1:07:36Second Q&A break
- 1:08:22AI slop and code quality issues
- 1:09:04Feedback flip for better code
- 1:12:53Refinement loop process
- 1:17:27Third Q&A break
- 1:19:10Silent misalignment in AI
- 1:27:36Active partner pattern
- 1:31:30Check alignment before implementation
- 1:34:37Context markers for visibility
- 1:37:51Answer injection and its risks
- 1:43:09Weaponizing answer injection
- 1:46:38Reverse direction technique
- 1:48:44Polyglot AI and multimodality
- 1:51:33Using modalities in practice
- 1:56:25Summary and key takeaways
- 1:58:10Closing and resources
related talks
27mEmerging Patterns Agentic AI
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips
48mAI Coding Agents and How to Code Them
01h 01m10 tips to level up your ai-assisted coding
01h 03mAI Coding Agents and how to code them
30mBeyond the Hype: Using AI Coding Agents Today
14mThe 4 Patterns of AI Native Development
01h 01m10 tips to level up your ai-assisted coding