It's 7:30am in Sarandë (Albania) and my kids are giggling in an empty street: "Papa messed up!" We're standing in front of a bakery with the shutters down, and Google Maps says it's open 24 hours...
We're a family of early birds, all of us usually up before 7am. Even on vacation. Which in Sarandë means you're out while the town is still asleep, and it's been hard (not impossible) to find a place to grab breakfast and coffee open before 8:30. On my first day, gullible me told my wife "amazing! there are so many places open 24/7!" based on Google Maps info. I hadn't noticed that these were unclaimed business listings, nor that half the listings carried no times at all, while the rest all said "open 24 hrs".
I was confidently misinformed. I know, it's just coffee, but I came with the confidence of someone who has accumulated more than 9,000 notes over 10 years with 22,500 interconnections in his second brain. Hell, I should KNOW better. Yet I was the one standing in front of that taunting shuttered storefront, still wanting my morning coffee.
Fool me once. I made a mental note not to trust Google Maps listed hours in this city anymore, and adapted my habits to the city's circadian rhythm. I didn't get a chance to go in and ask for the opening hours to update the listing, so the correction lives just in my head.
Want this running in your team?
evoilabs starts with one task, the one your team repeats every week, and builds the system that handles it. Then your team builds the next one.
See how it works →Hey, even the AI guy that talks about context engineering still gets fooled by a context problem. If you're reading this from the beach in a country you've never been to, make sure you challenge your assumptions on opening hours, what really has a 5-star rating, and what information actually exists online vs talking to people face-to-face. If you're at your desk, think about what tribal knowledge you've created at home or at work that's led to a "hey where did I put that post-it with the times for the plumber to drop by?" or "whatever you do, don't change that spreadsheet. The guy who built it left years ago and we don't know how it works."
We all have some sort of "listing nobody has touched in years"-shaped skeleton in our closet. And they stay there because we as humans are relentless optimizers, so we won't even have noticed that we've routed around bad info rather than fix a problem, until it blows up in our face. So when you try to throw AI at your process, of course it doesn't give you the expected output. Until it's corrected, it just becomes an additional burden: you can't automate chaos. Given what tokens cost these days, the real price to pay is in fatigue, and in people quietly deciding the thing doesn't work.
You're handed bad context ten times a day with no model anywhere near it. It's rarely as obvious as a shuttered bakery at 7:30am.