How I lost my (old) job to AI

September 18, 2024

78 min episode · 2 min read

Kyle Carberry,Sharon Diario,Steve Paul

Episode

78 min

Read time

2 min

Topics

Artificial Intelligence

AI-Generated Summary

Published Dec 19, 2025

Key Takeaways

✓AI Performance Threshold: AI delivers average results, helping below-average developers significantly while potentially limiting those performing above average, creating uneven productivity impacts across skill levels.
✓Practical AI Limitations: ChatGPT and similar models frequently provide incorrect technical answers, particularly for infrastructure tasks like Terraform and AWS configurations, requiring manual verification for production systems.
✓Investment Bubble Indicators: Safe Superintelligence raised $1 billion after three months with no product, representing peak AI hype similar to cryptocurrency bubbles, suggesting market correction ahead.
✓Code Generation Reality: AI excels at mundane tasks like test writing and tab completion but fails at complex application architecture, requiring detailed prompts equivalent to pseudocode programming.
✓Bias and Training Concerns: AI models encode societal biases and potentially favor expensive enterprise products in code generation, creating hidden costs and discriminatory outcomes in automated systems.

What It Covers

Software engineers Johnny, Sharon, Kent, and Steve reunite to discuss AI's evolving impact on their profession, examining hype versus reality in development tools.

Key Questions Answered

•AI Performance Threshold: AI delivers average results, helping below-average developers significantly while potentially limiting those performing above average, creating uneven productivity impacts across skill levels.
•Practical AI Limitations: ChatGPT and similar models frequently provide incorrect technical answers, particularly for infrastructure tasks like Terraform and AWS configurations, requiring manual verification for production systems.
•Investment Bubble Indicators: Safe Superintelligence raised $1 billion after three months with no product, representing peak AI hype similar to cryptocurrency bubbles, suggesting market correction ahead.
•Code Generation Reality: AI excels at mundane tasks like test writing and tab completion but fails at complex application architecture, requiring detailed prompts equivalent to pseudocode programming.
•Bias and Training Concerns: AI models encode societal biases and potentially favor expensive enterprise products in code generation, creating hidden costs and discriminatory outcomes in automated systems.

Notable Moment

Kent describes AI perfectly completing function names in his code comments by analyzing context, demonstrating genuine utility while simultaneously expressing discomfort about copyright infringement in training data.

Know someone who'd find this useful?