×
Ben Kim
The question on everyone's mind
How are the best developers using AI
to write code?
Slide 03 — History
A short history of AI coding
2023
ChatGPT prompt → copy / paste
Q1 2024
Cursor Autocomplete
Q1 2025
Cursor Chat
Q2 2025
Claude Code Chat
Q4 2025
Claude Code Opus 4.5 dominates — 100% of code can be AI generated
Q1 2026
OpenAI Codex GPT 5.4 dominates
Slide 04 — State of play
Where are we today?
- 100% of code is AI-written.
- Software is approaching the cost of power and compute.
- Total cost of engineering should inversely track the cost of tokens.
- Salaries are being displaced by token use.
- Tools are fragmented — Claude Code, Codex, Cursor, v0, Lovable, Pi — and token costs are rising.
Slide 05 — Case study
Case study: Ramp
Ramp is a US fintech that issues credit cards.
They built an AI that receives errors from their system, writes the code to fix them, and reviews or tests the work.
57%
of their code is written by this AI
80%
of the AI itself was created by the AI
Slide 06 — Building internally
Building this internally
En mi empresa actualmente:
- ~35% del código que se shippea no viene de prompts humanos.
- ~80% de los bugs se resuelven sin prompts humanos.
from “AI writes 100% of my code”
to “I don't tell the AI what code to write.”
to “I don't tell the AI what code to write.”
Slide 07 — Takeaways
What can you do — today, tomorrow?
Today
- trainingEducate your team
- trainingWorkshops
- trainingOffice hours / consulting
Tomorrow
- infrastructureBuild systems that do your work
- systemsUse the best tools (harness, code review)
- cultureTop-down AI mindset
- hiringHire experts
