AI Productivity Addiction

**Brian** (0:00)
Hey, what's going on, everybody? Welcome to The Daily AI Show. Today is July 14th, 2026
And with me today are Andy, Anne, Beth, and I'm Brian. And yeah, we had a good pick off yesterday. Great discussion in the comments section.
So as everybody's coming in today, we already appreciate that very much. Today with us, we have Anne, yesterday we had Gareth popping in for a little bit, so it's always appreciated. Thank you, Anne, for being here on Tuesdays. And yeah, I think there's been a few kind of interesting stories coming up, but I'm curious what you guys thought was interesting over the last 24 or so hours. Beth, what kind of stood out to you?

**Beth** (0:44)
What stood out to me is the continuing seeming disappointment with 5.6.
Sol, I see that more than I see talking about the other ones, but it just seems like there is significant difference with this model, and in a way that may be better, but that is hard for people to grasp. So I think what we're moving into is more goal kind of prompting in a way that there may be people who have not gotten to that place in terms of the clarity of their prompting. So one of the things that I was reading this morning was that if you let it, it will over engineer anything. It will just find tiny bug after tiny bug after tiny bug, and that is not something that this programmer was experiencing with 5.5, but it is something that goes away if you write a really clear goal prompt and narrow the constraints in which it can function. So some disappointment, still like a token eater definitely uses more tokens, but I do believe we're moving into places where that kind of like you can get away with it is less. That's what I would say.

**Brian** (2:21)
And I was just looking just because you happen to bring it up. I was like, I think I just read about this somewhere and it was it was the neuron put out an email. I'm sorry, not neuron. I apologize. Ben, Ben's Bites put out a newsletter today literally called How to Use GPT 5.6, and I hadn't read it. So I was just scrolling really quickly to see if he was he was saying similar things, some general patterns I've noticed. This is Ben from Ben's Bites. I assume Sol is pretty good at UI, but it's even better once you give it some references. Sol at max thinking has really good writing and it's fun to chat with. Terra feels like a replacement for 5.5 with minor improvements in UI and writing skills. It also feels more steerable, so skills would be useful with it. Luna has a bit of a mini model smell like something it doesn't get. Sometimes it doesn't get what you mean in ambiguous prompts, but it doesn't fail the tasks you clearly defined.
There might be more here to that, but anyway, those are some quick highlights from it, which I suppose it's not exactly what you were saying, Beth, but it is in line with perhaps if people aren't doing enough with their prompts, then the models are at a point now where they're far enough to pick up and run really fast before it necessarily knows what the direction is. It just runs really fast, which doesn't really help if you're running relays around a track.
If the AI doesn't understand the goal, but it only knows run fast and go hard and produce results, that's not exactly usable.

**Beth** (3:59)
Right. One of the suggestions was, if you are asking it to perform a security review, include in that instruction the level at which it matters. There will always be a cleaner way to have written this.
Is the thing that is insecure, is there a live API key in the thing? Fix that. Absolutely. Tell me to regenerate my key. Is there a better way to program something that would be a tighter, more narrow, whatever, that may not be something that is relevant, particularly based on the use, who is using the system? So mostly what I'm seeing is, the programmers from the labs have come out and been doing talks and education about like, hey, we're now into full programming, right? Like we don't program anymore, or we don't prompt anymore, we get a goal, we give it a parameter, we give it a success story, and then we let it go.
And I think that's now becoming a painful learning experience for some.

**Andy** (5:23)
I've got a couple of quotes here about a fellow, Zvi Maushevitz, who did a, he's apparently a highly regarded commentator on these things, and you'll hear from his commentary about a strong comparison between Fable and Sol.