Accelerating Linktree Engineering with AI agents
Jiayao Yu, Eng Director Linktree
Our AI Tool Ecosystem
Prototyping
Replit / V0 / Lovable for 0→1
Devin for prototyping within Linktree codebase
Bug fixes & repetitive tasks
Devin to avoid context-switching local environment
Faster to tag Devin in Slack than creating a bug report.
Main Development
Cursor with Devin scaffolding
20% chance Devin just one-shot entire feature.
Code Review
Cursor Bugbot review every PR
Complex Challenges
"Cyborg" approach: Human + All the best agents.
Common Engineering Pushback
"Devin is dumb."
"It takes much longer to prod Devin into building the right thing than doing it myself."
"It hallucinates, introduces bugs, and has caused outages."
"I feel my coding skills are atrophying."
Understanding AI failures
Use the best model / agent
Demand GPT-5 from all vendors
Nudge the engineers stuck on defaults
Hallucination rate as a code health metric
When AI makes mistakes, view it as a signal for code quality issues
Agent competency
Doom loop is often a skill issue.
Using agents effectively is an important and high ceiling skill.
Preparing for Second-Order Challenges
Code Quality
Push back on AI Slop in code review
Test coverage more important than ever
Org / Responsibility Design
Give engineers larger scope and taken out of their comfort zone.
Convergence of Product / Eng / Design
The AI-Powered Software Factory
Humans supervising teams of specialized AI agents
  • Provide knowledge, training and guidance to AI systems
  • Spot problems and bottlenecks in development workflows
  • Address second-order problems arising from AI-enhanced development