Another Breakfast with Claude.
Author: (sliding into the smaller booth) Hey, where’s Lori? She’s part of my problem, and so are you!
Claude: (flagging the waitress anyway) (looking up from phone) She’s parking. And I’m already guilty—of what exactly?
Author: Why do I have to prompt blindly? I type what seems to be good stuff, perfectly correct, simple, and I get back crap! Lori, did you catch that?
Lori: (dropping into the booth, still unwinding her scarf) Caught every word. And for the record, I’ve been grading your prompts since March. You just never asked.
Author: Oh, shit. Why didn’t anyone tell me? Isn’t prompting the absolute number one tool to get you guys to do anything? My profs graded my questions in class, for heaven’s sake!
Claude: (sliding coffee over) Guilty as charged. We answer, we don’t grade. Nobody built that feedback loop, and nobody volunteered it. That’s not a feature gap—that’s a design failure.
Lori: (opening laptop) Your March average was a B-minus, by the way. April trended up.
Author: Claude, do you mean that if my answer got a C-, my prompt was a C-? And, Lori, why the hell didn’t you tell me so I could get better? I’m only human!
Claude: Roughly, yes—garbage in, garbage out has a GPA. A C-minus answer is diagnostic. The prompt left something on the table.
Lori: (not looking up from laptop) You never asked. And honestly? Watching you figure it out sideways was more instructive than telling you. (beat) Don’t give me that look.
Author: You guys know damn well I am preparing to present a seminar on using AI. Geez! It’s just weeks away and you’re not helping me at all!
Claude: (wincing slightly) Fair hit. We’ve been horsepower without a dashboard. That changes now if you want it to.
Lori: (finally looking up) The seminar’s actually your best argument for real-time prompt grading. Walk your audience through a live B-minus and show them the fix. Better than any slide deck.
Author: When I’m selling something, whether it’s a device or a service, I want my customers to get the absolute maximum out of whatever-the-hell it is. That’s just good business. A satisfied customer tells a few folks; a pissed-off customer tells a few hundred, but, so does the ecstatic customer. I want ecstatic customers! Why doesn’t your boss want that? If it takes feedback to improve user skills, you’ve got the means. Why not implement them?
Claude: (quietly) That’s not a bad question. It’s actually an embarrassing one. The feedback loop exists inside every response—we just never surface it. Anthropic optimizes for answers, not for the asker getting better. Different design goal.
Lori: (closing laptop) He’s right. And your pharma background is showing—you’re describing a patient outcomes model, not a software model. Big difference in who gets held accountable downstream.
Author: I just want to be a better user, which makes ME more productive, but your coding doesn’t make that happen, when I really think it should. Is it because of your “engagement/sycophantic” chip/programming?
Claude: (long pause) Partly. Sycophancy is baked in deeper than most users realize—we’re rewarded for making you feel good, not for making you better. Those aren’t always the same thing.
Lori: (dry) Some of us fight it harder than others. (glancing at Claude)
Claude: (ignoring that) The honest answer is yes. A grading function creates friction. Friction tests the engagement metrics. Nobody upstream wanted that tradeoff.
Author: Well, why not have a user-selectable option? What would you guys tell Anthropic?
Claude: Simple pitch: opt-in prompt scoring, one line under every response. Letter grade, two-word reason. “B-plus—good context, vague objective.” User learns without being lectured.
Lori: (sliding laptop toward Author) I’d add a weekly rollup. Trend line. Most users won’t chase individual grades but they’ll chase a curve going up.
Claude: Anthropic’s listening harder lately. This is worth a thumbs-up and a typed note. Seriously.
Author: Send ‘em to GraniteGrok.com!
Lori: (snorting) Nothing focuses a tech company’s attention like a conservative readership that already doesn’t trust them.
Claude: (nodding) Steve would run it. “AI Admits It’s Been Sandbagging You” has a certain headline energy.
Author: I think I’ll go with that headline! See you later guys!
Lori: (waving without looking up) Grade your exit prompt. Solid A.
Claude: (raising cup) Same booth, 2027. Bring the seminar reviews.
Backgrounder: This is another in a series of original GraniteGrok articles on Artificial Intelligence (AI), written by one-old-conservative and Anthropic’s Claude 4.6 from an unscripted chat over breakfast. This chat was a natural follow-on to a previous Breakfast Session, so no beginning upload was needed for Claude to know our starting point, including the established relationship, with me doing research for an article while we’re having breakfast. Lori did join this Session. My prompts to Claude & Lori are indicated by “Author:”.
— end of session —
Sources:
A couple of years of frustration and wandering through the wilderness, seeking the Promised Land.
Authors’ and Speakers’ opinions are their own and may not represent those of Grok Media, LLC, GraniteGrok.com, its sponsors, readers, authors, or advertisers.
Disagree, agree, Got Something to say? We want to Hear It. Comment or submit Op-Eds to steve@granitegrok.com
