How Businesses Should Adopt AI

AI is Smart Until It's Dumb: Why LLM Will Fail When You Least Expect It - Emmanuel Maggiori

Topic 11 of 19

Published Nov 10, 2025 · Clip 34:12–38:18 (4:06)

Emmanuel advises against starting with 'what can we do with AI'—the hammer looking for nails—which causes most failed AI projects, citing companies that created biased internal AI-promotion teams. Instead, treat AI as one tool: find problems first, acknowledge hallucinations and limitations from the start, and embed that into the product (e.g., have AI point to source paragraphs rather than interpret). Use AI where the problem is easy to describe in a prompt and easy to validate, like asking ChatGPT how to read a file from S3 with Boto3; avoid it for hard-to-describe or hard-to-validate tasks.