QUESTION · ARCHIVE

How do you start a big project like Show Hub—do you write a spec doc outside of the AI tool or upload it as a product requirements doc?

ASKED BY · Unknown · 19 SEP 2026 · ▶ 00:22:57

THE PANEL'S ANSWER
Craig

Craig uses AI to generate specs by having it interview him first rather than immediately building; this produces detailed requirements (2-10 pages) that get broken into phases, with rapid MVP iteration and adjustments based on hands-on testing.

Summarised by AI from what the panel said. Watch the moment for the full answer.

IN THEIR OWN WORDS

From the transcript.

Machine-transcribed, so it may contain errors. Every timestamp opens the episode at that second.

22:33start your projects but obviously conceptually you have an idea for a product you kind of know what it what you want it to be but I like for me and I've only done like a couple of I mean extremely rudimentary basic apps and I always started with like a simple a simple prompt but how do you start like a big project like show hub obviously many people in the community that watch

22:57this show know what it is uh but there there are many that don't so like when you're going to develop a web-based application do you write like a spec doc a product requirements doc outside of the the ai tool and then just upload and say this is the the product spec sheet or or like how are you doing that you know i i i use ai to do it and one of the easiest ways is to say

23:25hey i want to build an app to do and then it's a two or three sentence thing and then you tell it interview me for what you need to know to build the spec. Not to go ahead and do it, but to build the spec. Because often the spec can be huge, 20 or more pages with lots of detail. And one of the things you want to be able to do, especially these days, is be able to break that

23:53up into phases. Because once you start playing with it, you may have different ideas about the you know, feature or enhanced features. And so have it interview you and then it have it create it and tell it, and I'm going to have this other AI bot build it out. And that piece is actually surprisingly interesting because it'll say, oh, if you're going to have chat GPT build this,

24:21here's some things that, and it will disclose things that it thinks are deficiencies or extra things that it needs to know about. And that's really fascinating to see. But then, so then I'll have this doc, whether it's two or 10 pages, and then I'll give it to the bot that's going to be building it and say, look at this and tell me what's missing or confusing and give me how are

24:50you going to approach it. And then it goes ahead and does that. And I say, go ahead and whatnot. and it builds it. And usually I want the smallest, like show me enough so I can see that it's real, play with it because I'm human just like everybody else. My expectations of the future are going to be changed once I put my hands on it. So I want to see that MVP prototype, whatever,

25:20as quick as possible. And then I'll do a phase two or three that builds it out and adds all these things. And typically in that interview process in the beginning, I'll have it make sure it talks about, so what's the end goal? Is this like an app that you're going to put in the app store? Is it something you're going to be in production with thousands of people using?

25:44is it just for you and your friends like make sure it covers some of those sort of things as well yeah you know i gotta admit i i don't think i would have thought the fact that you asked it to interview you to uh provide what it needs to develop it i probably would have wasted hours writing a document i don't think i would have thought about that that's that's super clever

26:08I'm going to, I'm writing this down right now. I'm going to remember this. Go ahead, Guy. Yeah. So I'm probably going to get some grief in the chat for this one, but this was in service to the community. So I came at it with a good heart. So whoever's going to give me grief, yeah, I'll deal with you later. But basically I had a couple of people message me over the last

READ THE FULL TRANSCRIPT →
MENTIONED IN THE ANSWER
MCP server→
ALSO IN THIS EPISODE

28 more questions
from CoreVideo Pro Zoom App Uses GPU-Native Encoding | Vertical.

The whole episode →
▶ 00:00:03I hear you've been working on some production related tools. Do you have anything you can share with the class this morning?▶ 00:03:28Is this app living inside of OBS, or is this accessible within it, or is it outside?▶ 00:04:19How do you get Zoom sources into OBS, and where does OBS come into play with this app?▶ 00:05:02Is that a performance meter of some kind in the buttons at the bottom of the toolbar?▶ 00:05:48Do you have individual telemetry for the participants that are joining, like can you see their frame rate and the resolution that they're sending when they've joined?▶ 00:06:14If you've got someone that's joined the meeting and their audio is fluctuating a lot, if you as the engineer want to remotely turn their audio up or down, is there any way to do that on a per-user basis?▶ 00:08:06Are you charging for these apps, or is Core Video Pro still in alpha?▶ 00:09:15What can you speak to the differences in terms of the video performance, the engine, the graphic subsystems like on a Windows PC versus Mac—are you running OpenGL on a GPU? Is it GPU driven on the Mac as well?▶ 00:10:48Can you live stream from this app? What about recording? Where does it go from there?▶ 00:11:37How does the routing work for ISO recording and output options?▶ 00:12:25What record options and settings are available for per-camera bit rate control?▶ 00:13:13Are there plans for remote control surfaces like iPad apps or web interfaces?▶ 00:13:33Have you heard anything about the new Zoom talkback feature in the API that allows you to have a comms solution?▶ 00:15:24What's a low budget and high budget underwater video and audio setup?▶ 00:18:24Can you recommend additional resources for underwater video and audio?▶ 00:20:19What's your AI project of the week—anything to share?▶ 00:26:32What's your current AI project related to searching and indexing Office Hours content?▶ 00:30:07How much does a talker's distance to the wall behind them matter compared to the wall behind the mic? Which side would you treat first?▶ 00:32:41What's everyone paying for AI plans?▶ 00:34:57What's the relationship between Cursor and X/Elon's companies, and what models does Cursor offer?▶ 00:37:53Can you describe your personal and work AI spending and platforms?▶ 00:39:30How do you think about AI plan costs relative to the value and time savings they provide?▶ 00:40:50Can you describe your experience building production tools with AI using MCP and cloud instances?▶ 00:42:16How has AI enabled rapid clip generation and highlight creation for Office Hours content?▶ 00:52:52The iPhone 18 Pro has variable aperture—what can't the iPhone still do that Apple needs to add next?▶ 00:53:15What is the optical zoom and sensor size on the iPhone 18 Pro, and how much does computational photography still play a role?▶ 00:56:04Did Samsung have a phone camera with editing capability years ago, and how does that compare to iPhone features?▶ 00:57:45What role did the Newton's patents play in iPhone development and Apple's survival?