Really appreciate this addition, Mateo! The agentic systems point is spot on — and honestly something I didn't fully explore in the post. A hallucinated answer you can fact-check before acting on it. A hallucinated action (wrong API call, bad data write) might already be done before anyone notices. This might actually be worth a follow-up post — "Why hallucination gets more dangerous as AI gets more autonomous." Curious if you've seen specific failure patterns in agentic setups you're consulting on?