Rafael: So, I read about this guy, Nevo David, he has this fitness account for TikTok. He just told his AI, you know, 'Find the latest trends, make videos, and then post them.' Like, one instruction.
Ama: Just like that? One instruction for all of that?
Rafael: Exactly. And the AI, it just… went. It filled his whole content calendar for the entire week. He said it felt illegal, you know? The ease of it.
Ama: Illegal? That's quite a word. But I get what he means. It's like, you set it free and it just… works.
Rafael: Right? We're past just, like, writing a prompt and getting one thing back. People are building entire, like, automated departments now. Stitching AIs together, saying, 'Here’s the goal, go run my marketing.' No human in the loop, initially.
Ama: No, but there is, isn't there? That’s the clever bit. With this Nevo guy, he creates an 'issue' for each video.
Rafael: Oh, yes, the feedback loop! I forgot about that. So, the AI generates a video, saves it as a draft, right? And then, he, the human, adds notes to this 'issue'—what was good, what wasn't, what music to use.
Ama: Precisely. So the AI learns. It's not just a one-off thing. It gets better. You teach it what works for your audience. Very smart.
Rafael: Yeah, like, imagine you have some workflow, something you do every day, with maybe two, three different tools. Like, moving data from here to there, sending an email, posting something. If you can explain it to, like, a new intern, that’s your first 'goal' for an AI agent system.
Ama: Hmm. An intern. That's a good way to frame it. Makes you break it down into simple, actionable steps.
Rafael: It’s like the next generation of Zapier, you know? My friend who has an e-commerce store, he uses Zapier to connect Shopify to Mailchimp and then to Slack, just for notifications. But this is like that, but instead of just moving data, it's intelligent agents doing tasks. It's Zapier with a brain.
Ama: Zapier with a brain. I like that. So for this TikTok example, the AI makes the slideshow videos, but it saves them as drafts. So the human comes in and adds the trending music before it posts. That's the human-in-the-loop, you were talking about.
Rafael: Exactly. And it’s not just content creation. He set up retention marketing too. When new code gets merged on GitHub, the AI automatically drafts announcements for Discord, for Slack, and even for a Mailchimp newsletter.
Ama: Wow. So it's monitoring his development process, then translating that into communications. That’s a whole different level of integration.
Rafael: And SEO! Same GitHub trigger. The AI sends the commit info to another service, and it writes a full, SEO-optimized blog post about the new feature. With images, quotes, everything.
Ama: So one action from a developer, and it ripples through marketing, through SEO, through social media. That is… efficient.
Rafael: It feels a bit like magic, no?
Ama: Magic, or maybe just very well-engineered. But it makes me wonder about the other side. Like, the coding article. That was also about AI, but a different approach.
Rafael: Ah, yes, Josh Pigford. He was using Claude for coding, and he found that its plans were… not thorough. Like, it missed recent information, documentation.
Ama: He said it was like a smart, but naive, junior developer. You can't just give it a goal, tell it to code something, and expect perfection. You have to give it a process to follow before it even writes a line of code.
Rafael: Yes, this is very important. You can't just say, 'Hey, AI, write me some code that does X.' You have to tell it, 'Hey, AI, what are the three to five research questions you need to answer before you can even propose a plan?'
Ama: Right. The most powerful step he found was making the AI distinguish between the 'problem' and the 'proposed solution.' So if a user says, 'Add a config option for X,' the AI first asks, 'Is adding a config option the real problem, or is the real problem that X should just work by default?'
Rafael: This is exactly how you manage junior creatives! Like, a friend of mine, she runs a content team. She makes her writers fill out this 'briefing doc'—user quotes, competitor examples, internal data—before they’re allowed to write anything. It’s like, 'Don't just write; first, understand.'
Ama: Yes, but this is for the AI, not for managing people, right? It's about building that research process into the AI, so it doesn't just blindly implement what you told it. It questions it, finds the core problem.
Rafael: Oh, right, right. For the AI itself. My bad. But the principle is similar, no? You need to make sure the foundation is solid before building.
Ama: Exactly. So the AI, after parsing the intent, it launches parallel research. It spawns sub-agents to simultaneously search the codebase, the documentation for libraries, recent Stack Overflow posts.
Rafael: And it does all this to solve the core problem, not just what the user suggested.
Ama: Yes. And then it synthesizes everything. It evaluates the user's proposed solution, saying, 'Actually, this is too complex,' and recommends a simpler, better path it found in the official docs.
Rafael: And the stress-testing! Before it finishes, it actively searches for what could go wrong. 'What maintenance burden? What edge cases does it miss?' It's like, a built-in critic.
Ama: That's the part that really got me. It's not just about finding a solution, it's about anticipating the problems that solution might create.
Rafael: So, we have AI generating content and AI writing code with rigorous research. It’s a lot, no?
Ama: It is. And honestly, this marketing automation, the first one we talked about? It feels like a recipe for generic, soulless content. Just polluting the internet with more noise. The real value of AI, for me, is augmenting a skilled human's process, like in the coding article, not replacing creativity entirely.
Rafael: But it's not replacing it, though! Nevo David, he's still reviewing, he's still adding the trending music, he's still teaching the AI. He's just not doing the tedious part, the initial video assembly, the trend research.
Ama: Yes, but the creative part is so minimal, isn't it? It's just adding music. The core 'idea' for the video, the aesthetic, the flow… that’s all AI generated. What if it gets so good that the human input just becomes, 'Looks good'?
Rafael: But isn't that the point? To make the human more efficient? If the AI can do 90% of the work, and the human just polishes the last 10%, that’s a huge win for productivity!
Ama: Productivity at the cost of genuine connection? I don't know, Rafael. These TikTok videos are meant to feel authentic. Can an AI create authenticity?
Rafael: People scroll so fast on TikTok, Ama. Most won't even know it's AI-generated. If it works, it works, no?
Ama: But you know. The creator knows. And if everyone starts doing this, then everything looks the same. And then what was the point of human creativity? It just becomes a content mill.
Rafael: I think it lets humans focus on the higher-level strategy, not the busywork. We're getting past boilerplate code, but now we have to learn 'prompt boilerplate' to get the AI to do what we want.
Ama: That's exactly it! Are we just trading one kind of boilerplate for another? Is this just the new framework we all have to learn?
Rafael: It feels that way sometimes. It's like, you master one thing, and then the next thing comes along and you have to learn a whole new way of telling the computer what to do.
Ama: And I'm Ama.
Rafael: And I'm Rafael. This has been Manish Chiniwalar's Station.