Daksh Gupta, co-founder and CEO of Greptile, joins Brett Gibson on High Bit and explains why human code review is essentially "security theater" and how Greptile uses AI to catch bugs by understanding entire codebases. He dives into why code generation and code review must stay separate, the surprising challenge of teaching AI what's a nitpick versus a severe issue, and how intelligence becoming "abundant and nearly free" is reshaping software practices.Plus: why some companies will be left behind if they don't adopt AI tools, what happens when human "taste" becomes the final bottleneck, and whether code legibility will matter in an AI-dominated future.
Follow Greptile and Dkash for more:
Chapters
00:00 - About Greptile, the evolution to specialized bug detection
3:00 - Humans are bad at code review. Why AI works
6:00 - Intelligence becoming abundant
7:20 - Sneaky disruption - what people are looking for vs what they need
10:50 - Code gen and verification are different problems
14:00 - Code gen and code review should be separate
17:15 - Getting LLMs to understand code
24:15 - Claude 4's tool-using capabilities changed their approach
25:00 - Architecture: from flowcharts to agent tools
27:30 - What’s hard about code review - what’s a nitpcik vs. a severe issue
31:00 - What was the “High Bit”
35:06 - Whether code legibility will matter in an AI world
37:25 - Why Terraform and infrastructure code is particularly difficult
43:25 - Re-architecting systems to be AI-friendly vs. adapting AI to messy reality
45:55 - Human "taste" as the final bottleneck
47:14 - Rick Rubin level taste in software
47:55 - Human appetite for change - kitchen exhausts for stoves
50:50 - Working with companies that use AI to generate code
