AI-assisted coding: practical lessons from small startups...
summary
Company runs Stonehenge Rails monoliths serving 80,000 churches with a 100-person engineering team and uses Claude with per-user and per-project Claude MD files to steer behavior. Practical lessons: use plan-and-iterate for messy legacy work and plan-and-oneshot for well-known APIs (Stripe); build agents for BugSnag triage and a Rails-specialist writer + Stripe-reviewer combo; prototype an AWS Lambda via LLMs that runs but isn’t commit-ready; watch context decay around ~125k–200k tokens; use Graphify to save tokens.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 5:16Setting up with Claude
- 8:25Defining workflows skills and agents
- 9:28Leash length and SDLC
- 13:01Plan and oneshot vs iterate
- 16:11Choosing the right approach
- 18:56Managing context and PR reviews
- 21:51Agents for efficiency
- 24:17Real-world AI workflows
- 26:43Skills atrophy and coding sharp
- 27:46Saving tokens and costs
- 28:51Final thoughts
related talks
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips
01h 03m10 tips to level up your ai-assisted coding
01h 42mAI Powered Application Modernization using Claude Flow
30mBeyond the Hype: Using AI Coding Agents Today
01h 00m10 tips to level up your ai-assisted coding
01h 02m10 Tips To Level Up Your AI-Assisted Coding
15mUsing AI effectively in large Java codebases
01h 01m10 tips to level up your ai-assisted coding