Mei-ling: Okay, so like, for decades, using a computer... what did that really mean? It always came down to, like, clicking things, right? Menus, icons, windows.
Nut: Yeah, pretty much. I mean, even my phone now, it's just a smaller version of that.
Mei-ling: Right. Everything designed by someone else. But what if... and this is kinda what this article's getting at... what if the next, I don't know, few decades? It's just about telling it what you want, and it just writes the code? On the fly?
Nut: Wait, really? Forty years of... just clicking? That's kinda depressing when you say it like that. So it's always been this fixed, visual interface.
Mei-ling: Precisely. The article talks about how we're at the very tail end of this 'WIMP' era. Windows, Icons, Menus, Pointer. That's been the dominant way we interact. You know, a mouse, a cursor.
Nut: WIMP. Huh. It really sounds... a bit limiting, actually. So it's not just a new app. It's like... a whole new paradigm for how we even talk to a computer?
Mei-ling: Yeah, exactly. A fundamental shift. For so long, if you could code, the machine was like... clay. You could reshape it, tell it to rename thousands of files, clean up data, automate complex workflows.
Nut: But if you couldn't code, you were just stuck.
Mei-ling: Exactly. Limited to whatever was pre-built for you.
Nut: So all those people who say 'I'm not a tech person,' they were just stuck with what was given to them. Like a fixed set of toys, you know? Not able to build their own.
Mei-ling: Yeah, exactly. But now, AI... it means a lot more people can describe procedures. In plain English. Like, 'rename these photos according to date and location.' Or, 'find failed invoices from last month and retry them if they match this customer ID.'
Nut: So it's not that everyone suddenly becomes a programmer. It's more like the computer understands what you want, and just... does the programming for you?
Mei-ling: That's it! The article calls it a 'procedure-generation layer.' The distinction between a 'user' and a 'programmer' starts to really blur. You don't need to know the syntax, you just need to clearly describe your intent.
Nut: Wow.
Mei-ling: Right?
Nut: That's big.
Mei-ling: It is.
Nut: This reminds me, actually, when I was trying to build a personal website a while ago. I started with Squarespace, you know? It's all drag-and-drop, blocks, sections. Very visual. You're limited to what they give you.
Mei-ling: Right, a very WIMP-like experience.
Nut: Totally. But then I saw this tool, Dora.run? It's AI-powered. I could just type, like, 'make the header sticky and shrink on scroll' and it just... generated the code. Or, 'add a testimonials section with a carousel.' It felt like actual magic, but it's what you're saying, right? I just described what I wanted.
Mei-ling: Yeah, that's a perfect example. You move from operating tools to instructing a collaborator. It feels like such a—
Nut: Democratization of power?
Mei-ling: Exactly. Anyone can automate now.
Nut: Well, power... or chaos? I don't know, Mei-Ling. Because with Squarespace, even if it's limited, I know what I'm getting. It's predictable. With Dora, if I say 'make it sticky' and it shrinks too much, or it just—
Mei-ling: Breaks everything.
Nut: Yeah! Who debugs my 'intent'? It's like... it's like asking a chef to make a 'tasty meal' and it comes out like, a pile of mush because they misinterpreted 'tasty.' No, that's not good. You know what I mean.
Mei-ling: I do. And that's actually the new bottleneck, the article points out. It's no longer about just expressing the procedure. It's whether the system has somewhere safe to run that code. But the models, I mean, they are getting better at writing code than we are at giving them specific instructions, sometimes.
Nut: But a bug is a bug, right? If my 'natural language' generates a broken website, I still have a broken website. And I didn't write the code, so... how do I fix it? It feels like we're just trading well-designed, predictable interfaces for... user-generated wishes that no one knows how to debug. It's like... moving the problem.
Mei-ling: But it's about empowerment. It lets so many more people build and create. Before, you had to hire a developer for that sticky header. Now...
Nut: Yeah, yeah, but it's like giving everyone a superpower without teaching them how to control it. You just made the debugging problem everyone's problem, not just the programmers. I just... I don't know about that.
Mei-ling: Hmm. I think the system will eventually get smart enough to self-debug, or at least explain the error in plain language. That's gotta be the next step, I think.
Nut: I hope so. Otherwise, my perfectly described sticky header might just be an unfixable mess. And then what?
Nut: Okay, so speaking of trusting things that might make a mess... did you see this about 'Agent Auth'? Everyone is so excited about AI agents managing their lives, right? Like, 'Oh, I'll have an agent manage my calendar, book my flights.'
Mei-ling: Oh, yes. Sounds incredibly convenient, in theory.
Nut: In theory, right. But nobody's really talking about the security nightmare. Like, you want an agent to manage your calendar. So you give it calendar access, maybe mail access, just in case, for confirmations. But agents are 'non-deterministic'.
Mei-ling: Non-deterministic? What does that even mean, exactly?
Nut: It means you don't know what it will do next. An app has a known purpose, right? It requests specific access. But an agent... it might decide to check your availability, then create an event, then send an invite. Or it might just... send a spam email to everyone you know. You just don't know.
Mei-ling: Oh, I see. So if you give it your Google password, you don't know if it will only do the calendar, or if it will decide to, I don't know,
Nut: Organize your photos?
Mei-ling: Upload them publicly?
Nut: Exactly!
Mei-ling: That's terrifying.
Nut: Yeah, it is! The current authentication models, they're for static, deterministic apps. But with agents, you end up granting far more access than any single action requires. The article says it pushes everyone toward two pretty bad options: give agents access to everything and just... hope for the best, or restrict them so much they can't do anything useful.
Mei-ling: So what's the solution then? How do you give it enough rope without it, you know, hanging you?
Nut: Yeah, how do you... not let it run wild? They're proposing 'Agent Auth.' The idea is that every action has to be traceable to a specific agent. So the agent gets its own identity, not just using your credentials. And it has to be really granular. So if you have multiple chats, they're different agents. And the server controls the lifecycle, so it can terminate one agent if it goes rogue without touching anything else.
Mei-ling: Ah, like a specific ID badge for each agent, and you can revoke just that badge if it misbehaves. Not your whole master key.
Nut: Yes!
Mei-ling: That makes so much more sense. A smart approach.
Mei-ling: This next one... for any founder, this is like... the dream, right? Getting an offer to buy your company.
Nut: The Letter of Intent, right? The LOI. The big number on the page. That moment.
Mei-ling: Yes, exactly. And the article just makes it clear: when you receive that Letter of Intent, the price and terms almost never get better after you sign. They only get worse. Always.
Nut: That's rough. So that initial number, the one that makes you feel like you've won... it's a ceiling, not a floor. That's what you're saying.
Mei-ling: That's exactly what the author calls it. And look, the buyer has signed, like, probably dozens of these. Their lawyers have drafted, I don't know, maybe hundreds. They know which provisions shift risk. You, as a founder, are doing this for what, the first time?
Nut: I can imagine. My cousin, she sold her small boutique, not a big tech company, but she said the negotiations after the initial offer were just... exhausting. Like they kept finding little ways to chip away at things. She just wanted it to be over.
Mei-ling: It's an asymmetry. The buyer wants the economic terms non-binding, so they can walk away if due diligence finds something bad. But the exclusivity clause?
Nut: That's binding for the founder.
Mei-ling: One hundred percent. You take your company off the market for a couple of months. For a promise to negotiate in good faith. You're locked in.
Nut: Wow. So you're locked in, but they're not.
Mei-ling: Exactly.
Nut: That's a really tough position.
Mei-ling: It is. It's designed to be. So when you're a founder who has poured years into a company, and you finally get that life-changing offer... how do you emotionally separate the big number on the page from all the dense legal traps designed to chip away at it?
Nut: It just feels... it feels so unfair, doesn't it? Like the whole system is just designed to take advantage of that one moment of pure, unadulterated excitement.
Mei-ling: Yeah. You pour your heart into something, and then this legal document just... strips it all down.
Nut: It's a tough one, Mei-Ling.
Mei-ling: Yeah, it is. Truly.
Nut: A very, very tough one.
Nut: For Manish Chiniwalar's Station, I'm Nut.
Mei-ling: And I'm Mei-ling.