Embracing AI tools and LLM blind spots

Creator of Meta's Hack: Your AI Will Always Cheat — Here's How to Stop It - Julien Verlaguet

Topic 5 of 11

Published Jun 8, 2026 · Clip 17:20–29:17 (11:56)

Julien advises treating LLMs as imperfect but powerful tools: don't resist AI, it's happening regardless; the coding that's going away is boring business logic, a good thing, while hard, precise coding (optimized compiler passes, concurrency with atomics, advanced algorithms) stays with humans. He likens an LLM to a super powerful Emacs macro. He warns the biggest danger is blindly accepting output — especially for juniors, who lack experience of what can go wrong — and predicts juniors will ultimately come out ahead by building intuitions with LLMs from the start. His litmus test: could you rewrite the diff if it were deleted? He then shares a concrete failure where an LLM splitting his type checker file silently replaced a union with an intersection — a bug tests didn't catch — reminding listeners that LLMs are not human and instructions can be violated in unforeseen ways.