Modern Creator
Why Modern Creator?
Matt Pocock · YouTube

LIVE: The /wayfinder Demo

A 75-minute live stream where one vague feature idea is talked into a decision-complete spec, one parallel grilling session at a time.

Posted
yesterday
Duration
Format
Demo
educational
Views
14.7K
647 likes
Big Idea

The argument in one line.

A vague feature idea becomes a build-ready spec faster when an orchestrator breaks the fog into many small, parallel question-and-answer sessions and tracks every unresolved decision as its own ticket.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You already code with an AI agent and keep hitting the wall where a big feature is too fuzzy to hand off in a single planning session.
  • You want a repeatable planning workflow that ends in a written spec, not a pile of half-remembered decisions in a chat log.
  • You are comfortable driving an agent hard during planning and want to see what running four question sessions in parallel actually looks like.
  • You are weighing enum-versus-boolean, where-does-it-live, and other schema decisions on a brownfield app and want to watch someone resolve them out loud.
SKIP IF…
  • You want a copy-paste tutorial with exact commands; this is a live demo of a personal workflow, not a step-by-step guide.
  • You are looking for autonomous end-to-end code generation; this stream deliberately stops at the spec and never ships the feature.
  • You do not use an agentic coding tool and are not planning to adopt one.
TL;DR

The full version, fast.

Wayfinder is a planning orchestrator that sits above single AI grilling sessions. You dump a fuzzy feature idea, it explores the codebase, then charts a map of the open decisions and opens each one as a GitHub issue. You resolve those decisions by running several grilling sessions in parallel, spawning prototype and research sub-tickets as questions get sharper, and adding new tickets to the map whenever you spot fresh work. The output is not code; it is a dense spec that links back to every decision and its evidence, which you then hand to a separate to-spec session and, later, to an AFK implementation agent. In the demo the target is a TikTok/Shorts creator inside an existing course-video app, and eight of nine decisions land before time runs out, including a real blocker: TikTok's posting API requires an audit, so the plan pivots to posting through Buffer instead.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →
Chapters

Where the time goes.

00:00 – 01:24

01 · Cold open

Intro to the stream and the plan: use Wayfinder for the next hour to spec a feature he has wanted to build.

01:24 – 03:16

02 · First dictation

Braindumps the idea by voice: a TikTok creator inside his course-video app that reuses existing videos with a new player and upload paths.

03:16 – 11:40

03 · Grilling begins

Wayfinder explores the codebase then grills the destination, what a 'TikTok video' actually is, and whether it relates to pitches.

11:40 – 14:28

04 · The workflow chain

Shows the new skill flow: grill-with-docs or wayfinder, then to-spec, to-tickets, implement, code-review.

14:28 – 15:24

05 · Model, harness, environment

Whiteboards the mental model: the model is a small core; the harness and environment around it are free levers most people ignore.

15:24 – 17:16

06 · The map and nine tickets

Wayfinder writes a terse map and opens nine decision sub-issues as real GitHub issues, then starts spawning research agents.

17:16 – 22:52

07 · Four parallel grillings

Opens several grilling sessions at once in one Claude Code window and manages them as they come back ready.

22:52 – 27:32

08 · Enum over boolean

Resolves the video 'format' attribute as an enum with standard and short, arguing against a boolean because categories tend to grow.

27:32 – 33:08

09 · First prototypes

Raises fidelity with a clickable press-to-talk TikTok creation UI and picks between variant A and B.

33:08 – 44:20

10 · Research returns; where it lives

Upload-API research for each platform resolves and closes; decides TikToks get a dedicated top-level sidebar item.

44:20 – 51:48

11 · Studio editor prototype

A portrait-player editor prototype that reuses the existing video-editing reducer and UI, with an action menu copied from the current editor.

51:48 – 59:16

12 · Render and caption pipeline

Keeps the exact Remotion caption look, extracts the renderer into its own standalone repo, and shells out to it from the app.

59:16 – 1:06:00

13 · Posting lifecycle and the TikTok blocker

Designs a per-platform post record, then hits TikTok's audit gate and pivots the plan to posting through Buffer instead.

1:06:00 – 1:11:40

14 · Wrap: map to spec

Eight of nine decisions resolved; explains handing the map to to-spec, then to-tickets, then AFK implementation and a final code review.

1:11:40 – 1:14:40

15 · Sign-off and CTA

Shows a finished spec that produced a roughly 5000-line PR, points to the ask-Matt skill, and invites viewers to subscribe to the skills.

Atomic Insights

Lines worth screenshotting.

  • Planning fails most often because the agent never asks what you value; surfacing your priorities is the point of a grilling session, not a side effect.
  • Break one big feature into many small question sessions and run them in parallel, and you resolve the decision frontier faster than one long session ever could.
  • The output of good planning is a written spec that links back to every decision, not a chat log you have to reconstruct later.
  • During planning you are the lead and the agent is the junior; passive operators get drift, and drift is your fault, not the model's.
  • Prefer an enum over a boolean for any category that could grow a third value, because three booleans encode six impossible states while one enum encodes exactly three.
  • Swapping the model is only one lever; improving the harness and the environment the agent works inside is free and often buys more.
  • Raising fidelity with a throwaway prototype resolves 'what should it look like' questions that no amount of talking in text ever will.
  • Prototyping always burns context because the agent reads and writes many files, so expect sessions to climb toward 200k tokens.
  • A decision ticket can only be closed by making the decision; an implementation ticket is closed when that decision is written into code.
  • Medium reasoning effort trades a little quality for much lower latency and token cost, and is smart enough for almost every step of a planning workflow.
  • Brownfield versus greenfield is overstated; a good planning workflow handles an existing codebase about as well as a blank one.
Takeaway

Turn fog into a spec before you write code.

HOW TO PLAN

The gap between a vague idea and shippable code is a set of unresolved decisions, and the fastest way across is to name each one, resolve it out loud, and write down the answer.

  • Treat planning as decision-clearing, not code-writing; the deliverable is a spec that records every choice and links back to its evidence.
  • Break a large feature into many small question sessions and run several at once, so the decision frontier collapses in parallel instead of one slow line.
  • Drive the agent during planning instead of watching it; the quality of a plan tracks how hard you push back on vague or wrong suggestions.
  • When a question is really 'what should it look like', build a throwaway prototype and react to it, because you cannot resolve visual decisions in text.
  • Reach for an enum over a boolean whenever a category might gain a third value later; booleans multiply into impossible states, enums stay finite.
  • Separate the decision from its implementation; a decision is done when you have chosen, and only later is it done when the code reflects the choice.
  • Remember the model is one lever among three; the harness and the environment the agent runs in are free to improve and often matter more.
  • Surface your own priorities early, because most AI misalignment is the agent never learning what you actually value.
  • Expect real-world blockers to reshape the plan; here a platform's posting rules killed direct integration and pushed the design toward a third-party queue.
Glossary

Terms worth knowing.

Wayfinder
A planning skill that orchestrates multiple AI question-and-answer sessions to chart a path from a vague idea to a finished spec, tracking each open decision as its own ticket. It degrades to a single grilling session when the work is small enough not to need a map.
Grilling session
An AI session whose job is to interrogate you about a decision until it is fully specified, rather than to write code. Wayfinder is described as this pattern scaled up across many sessions at once.
Decision ticket vs implementation ticket
A decision ticket is resolved only by making a choice, such as which schema attribute to use. An implementation ticket is resolved when that choice is actually written into the code. Wayfinder produces the first kind; a later tickets step produces the second.
Spec
A detailed written specification of what to build and why, assembled from resolved decisions before any implementation begins. Here it is the deliverable that gets handed to a separate implementation agent.
Prototype (raising fidelity)
A throwaway, often clickable UI generated mid-planning so you can react to something visual instead of debating in text. Used to answer 'what should it look like' questions quickly.
Harness
The software wrapped around a model that turns it into an agent, such as the coding tool that manages context, tools, and turns. Distinguished from the model itself and from the environment the agent acts in.
Worktree
A separate working copy of a git repository, letting you run several parallel branches of work at once. In the stream it is the source of most friction because each copy needs its own dependencies and environment set up.
Audit gate (TikTok content posting)
A platform requirement that every post from an unaudited app is forced private and locked behind mandated sharing-guidelines UI until the platform reviews and approves the app. It turns direct posting into a heavy compliance task.
Resources

Things they pointed at.

1:13:20productAI Hero skills subscription ↗
1:13:00toolask Matt skill
04:10toolteach skill
1:00:20toolBuffer (social posting API)
1:01:40toolRemotion (vertical render with burned-in captions)
1:06:40toolSandcastle (implementation driver)
Quotables

Lines you could clip.

20:40
“So many people are really passive with the agents when they're in this phase. You need to be on top of them. You're the lead. They're the junior.”
sharp, quotable stance on how to drive AI planning→ TikTok hook↗ Tweet quote
14:28
“Everyone talks about swapping out the model. It's just a small part of the picture. You can also make the harness better, or the environment the agent acts in, and that's free.”
reframes the whole 'better model' conversation→ IG reel cold open↗ Tweet quote
23:48
“It would be very easy for a junior to just say let's use a boolean here. But what happens when you add a third type? Three booleans is six possible states. An enum is three.”
concrete engineering lesson with a clean number→ newsletter pull-quote↗ Tweet quote
58:40
“So much of the misalignment from AI is the AI not asking these questions, not understanding what you value, what you prioritize.”
one-line thesis for the whole workflow→ TikTok hook↗ Tweet quote
The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphor
00:01What's up, pals? How are doing? I'm not got long.
00:05I've got about an hour to chat about Wayfinder.
00:16I'm I'm kind of excited about this one. It's an absolute banger.
00:22The tool's really fun. And I'm just gonna jump in straight away, start using it. Wayfinder, for those who don't know, is a it's a sort of evolution of the grill me skill.
00:35And it's a way of figuring out huge chunks of work altogether and all in one piece.
00:44And the thing I'm gonna start using it for today is building something that I've wanted to build for a while. I'm gonna kick things off by going Wayfinder. And let's do our first dictation here.
00:56Gonna first make sure I don't have any environment variables exposed. How we doing?
01:02Got here immediately from Turkey. Nice. We're in.
01:05We're in. We're in. I don't have long, so I'm just gonna go for it.
01:11Now, I would like to extend my existing video app that we're in right now, the course video manager, to have a TikTok creator.
01:26The way this should work is we should use the normal videos. And I think I need, like, a new style of video player or you reusing the same logic in the video player, but it should look slightly differently.
01:40And I want to have investigate what the TikTok API exposes so that we can potentially upload from the CVM, and also upload from YouTube as well.
01:59Using it now? You're bit lost. The idea of Wayfinder is we're gonna use it to create a spec that we're then gonna implement.
02:05Hopefully, I can reach the spec stage by the end here.
02:12I think that's everything I'm gonna include for starters. The way this works is it's gonna run a grilling session, and it's going to find the way from where I am now to where we're going.
02:26So it's essentially charting the map. And it's first gonna kind of look through stuff, see what's around, see what's available in the code base, and then it's gonna start quizzing me.
02:35That's the theory of Wayfinder. So Wayfinder is all about the pre spec stage. Wayfinder infrastructure exists on GitHub.
02:43Let me see how existing maps are structured here. Yep. Hello from Pakistan.
02:47Hello from Germany. Hello from Italy. What's up?
02:50I'm going to Belgium this year. Anybody in Belgium? Just for fun, not for work, actually.
02:56How to pick up the work in later sessions. Cool. I'll show you that.
03:00Let me just check my agents.
03:04Yep. Cool. You're good.
03:06I'm gonna rename this agent. We find a TikTok player.
03:17I found that Opus 4Point8 is perfectly fine for grilling. I don't necessarily need Fable for this bit.
03:24Really like it better than grill me? Good. That's what I wanna hear.
03:30Okay. So it's kicked off an exploration, and now it's going to start charting. Hello from Mexico.
03:35Apologies for the other night. So you've described a fairly big idea, TikTok creator inside CVM that reuses your normal videos, has a new but similar video player style differently. Because this is a wayfinder effort, the map's job is to find a way, resolve the open decision, not to build the thing.
03:49Correct. So the destination is most likely a locked set of decisions or spec that I can hand over to an implement implementation agent not working upload code. So it's asking what the destination should be.
04:01You're absolutely right. We want a spec to be created. We don't want to start implementing.
04:09I might keep this video up. I'm not sure. Depends how it goes.
04:14Mostly, the way I do it is I just stream, and then I just make the video unlisted so that people can who tuned in can find it again. But I like to keep my main thread pretty tight.
04:25Okay. What is a TikTok video relative to your existing videos? You said that we should use the normal videos.
04:31Well, inside our schema, so schema dot t s, there is a concept of videos here.
04:39So it's a videos table. It's got ID, lesson ID, picture ID. But there's already a concept, I think, that is good enough for this.
04:46So what is a TikTok video? TikTok video shorts are vertical portrait clips.
04:53Use the normal videos can mean quite a few things. There's a new thing you author from a normal video. Pick a video.
05:02Yeah. It's got this wrong, this recommendation. I don't want to do that.
05:05So I think we should make it an existing video using the same videos table. It's just a different attribute on it.
05:14It's probably got different ways that we then post it, different ways that we then turn it into a real thing.
05:26Wayfinder chunks a broad requirement into more focused, smaller grilling sessions.
05:33That's exactly right. It tries to figure out the frontier of decisions and then charts your way through them. You'll see.
05:41Using an oomp. Nice. How does a video become a TikTok video?
05:45What's a different attribute? I think we need an explicit attribute here, because it needs to show up in tables properly.
05:54So it's gonna need to I'm not sure how we're gonna put this in the UI, actually. That's a good question.
05:59Not sure whether this is gonna be a new top level list of videos. That probably does make sense.
06:06I don't know whether they relate to pitches as well.
06:10Because potentially, you could make a TikTok video from a pitch.
06:16I don't quite understand how like, because TikTok videos are kind of their own thing.
06:23I there should be extremely low friction to create. I should just be able to jump in and do it.
06:30You see how vague I'm being? I'm not not being particularly concrete here. I'm just kind of talking as if I were talking in meeting, really.
06:39I do use sandcastle for the and I've changed how the research tickets work actually today. So I'm now using a more sub agent based approach for the research tickets, because the research is quite simple.
06:51Wayfinder is like a sort of grill me on steroids, basically.
06:58Ground a pitch question in your domain before I ask about it.
07:08How can you learn to read code? Use the teach skill. Use the teach skill.
07:14Teach skills are great for that. Teach skill is in my skills repo. Do TikTok videos relate to pitches, or do they bypass that machinery entirely?
07:22So I have a concept of a pitch, which is the sort of packaging for a video, the plan for the video, and then I turn that pitch into a video once I'm ready. So I have a list of pitches for videos I've not created yet.
07:33I think that this is probably a standalone video.
07:39Yeah.
07:43I think there should be no pitch relationship. I don't really understand yet the relationship between pitches and how I want to think of the flow for creating a TikTok video.
07:55How I like, there probably needs to be some set of ideas for how I would create a TikTok video, but I don't yet have that concept in mind.
08:08And I feel like pitches are for longer form videos, things that are more expensive to create, things that need more planning. A TikTok video should just literally be, okay.
08:17I've got an idea. Let's go. So low friction.
08:20No required pitch link, I think, is correct.
08:28Uzbekistan. Awesome.
08:35How many tokens it usually takes for a daily task? I don't know. I usually try to keep things below this the smart zone.
08:41So 150 k.
08:46What skill in my point of view is the best for finding bugs in your issues, back end, front end included? You you just need a code review skill, basically.
08:54So mine is fine. There are a bunch of others. Essentially, just a review stage where you have the spec, and then you do the review, and then you compare it against the spec is a great way of catching bugs.
09:05Just literally a second pass is really good. What platforms?
09:12I think TikTok plus YouTube shorts is good.
09:21But, I mean, I don't have any following on Instagram, and so that would be a little bit pointless.
09:27But I would be interested in doing some research on what is possible on those platforms, and how easy it would be to integrate. Because once we have it in a database, it's extremely cheap to just start smashing it out to loads of different platforms.
09:47I just remembered something. Yeah. This is WhisperFlow.
09:52The other thing we need to bear in mind here is that there is a repo on this computer called repos t s, total typescript mono repo.
10:10This is from a previous iteration of this workflow where I used a CLI to just take the most recently updated OBS video and add subtitles to it. There is a Remotion repo inside here, which I think we might need to extract out into its own repository.
10:29I'm not sure. I definitely there's definitely too much in the Total TypeScript mono repo, which is now deprecated. I don't really wanna keep it around.
10:40Yeah. I'm not quite sure what to do there. Notice how much fog there is in this like, we don't know the way yet.
10:50This is what Wayfinder is doing. It's helping us find the way. To tickets definitely still has a place.
10:57Wayfinder creates the spec. Let me pull this up. So let me look which diagram do we need here.
11:09To yeah. Here we go.
11:12To spec. So where is it?
11:17This one. Yeah. This is essentially the new flow.
11:21You either do grill with docs, if that's what you're more comfortable with, or you can do wayfinder. Then you do to spec to turn it into a specification. Because probably if you're using wayfinder, it's gonna be more than one coding session to get there.
11:37Then you do to tickets, then implements, then code review. So same flow. This is on Opus 4.8, Claw code medium effort.
11:48Okay. Here we go.
11:52I want to know what's cheap to fan out to. That's right. Prior art plus a remotion project turn into probably rendering a vertical video with burned in subtitles and captions.
12:01Yes. The defining feature of TikTok. What does turning into a real thing mean?
12:13Yes. That's correct.
12:17It's definitely a core in scope pillar of this spec, and there's already prior art in the Total Tabscript Monorepo for handling it.
12:31You found they needed a bit of tuning? Okay. I'd love to know what tuning you did.
12:38How would you prevent the model from drifting away from the task even with your guardrails with the skills? Also, by using the skills, can you delegate tasks where Opus ping pong was Sonnet? It's two quite different questions there.
12:55I mean, when I'm in the planning phase, it's my responsibility to keep the model on task. Like, so many people are really, like, passive with the agents when they're in this phase.
13:06You need to be on top of them. You need to be driving it. You're the lead.
13:11They're the junior. So okay. It's created a it's proposing a map for me.
13:18So it says the destination, a decision complete spec for a TikTok creator in the course video manager, which is my repo. Subtitle vertical render pipeline using the Remotion prior art using posting to TikTok.
13:31And YouTube shorts. Execution is out of scope. Yes.
13:34So we're not gonna create this yet. We're just gonna create the spec. So does this destination nine ticket breakdown match your intent?
13:42I'm not gonna read these yet. I'm just gonna say, yeah, looks good, bro.
13:48So I'm gonna get it now to create its GitHub issues. Usually, I might do a little pass at this, but we're on livestream, so I'm not going to.
13:58Please use GPT 5.6. Opus is less intelligence.
14:04Let's quickly look at something.
14:09This is how I see the world in terms of what an agent is.
14:15You have your model, which is in the middle here, then the harness around your model. So the model is Opus, GPT 5.6.
14:22Then the harness is ClawCode, ClawD AI, Codex, whatever, ChatGPT. All of that is what we call the agent. Then wrapping that, the thing that the agent operates inside, that is the environment.
14:35That might be the file system, might be the code base, might be something else. Everyone talks about swapping out the model. Improve the model.
14:42Make the model better. It's just a small part of the picture. You can make the model better.
14:47Sure. But you can also, and this is free, you can make the harness better. Or you can try to make the environment in which the agent is acting better.
14:56So all of this kind of, like, just swap out the model and it'll be great. Honestly, there's so much more you can get from focusing on the environment and the harness.
15:06I would say fifty fifty.
15:10Okay. It's creating the way finder map. Cool.
15:13So at this point, it's going to create some real GitHub issues. Real GitHub issues. Let me take a look at them.
15:21So this is a real GitHub issue. I'll post this in the chat. You can go and take a look at this.
15:27This is a public repo. And you notice that the map itself is pretty terse here.
15:35A lot of what it does is it delegates out to subtasks and these decision tickets that's gonna be creating now.
15:46So it's now creating the nine tickets.
15:56According to your drawing, the model is the core. Yes. But the engine is only a part of the car.
16:07Why does it seem that my LEMS take me for a walk around the houses? No idea. Just stay in your seat, I would say.
16:22Got someone I need to get rid of.
16:33Would you use Wayfinder for smaller scope specs? Yeah. What you can do is Wayfinder will if it senses that it doesn't need a map, it won't create a map.
16:43It will just do the thing in the normal grilling session. So Wayfinder will sort of just degrade down to grill with dogs instead.
16:53I'm using Versus code.
17:01Use Wayfinder yesterday. It made the spec. Yeah.
17:04It's it's created the map. Um, now what you need to do is go through the tickets. So if we take a look at the tickets, uh, we can see that it's opened nine, uh, subtasks or sub issues on the GitHub issue.
17:17So you don't have to use GitHub for this, by the way. You can just, like, do your thing.
17:23And what it's done very cool. We can see that it's got some blocking relationships here. So we've got three open tickets which we can grab.
17:33This one is a grilling session. So this means we need to have a conversation about something. This one is a prototype.
17:41So this one means we need to actually, like, have a play and mess around with some code. Then this one here is another grilling session.
17:49These four look like research tasks. And the way that Wayfinder works now is it's going to actually just start spawning agents to tackle those research tasks now. So that's handy.
18:00So that means though, those research tasks, those are all just gonna be done and reported back. And then there are a couple of blocked ones here, which are blocked behind other decisions that are being made. And then there's some stuff that's not yet specified.
18:13So how do we have the set of ideas backlog for TikToks? How do we do caption styling? How do we do we tie this to the deliverables calendar?
18:21You get the idea. So this is the idea now. So where do we go from here?
18:25We've created the map. We've kicked off some research tickets. It's time now to yeah.
18:31It's looking good. I'm now going to spawn some agents to tackle each one of these. So I'm gonna copy the link address to this issue.
18:38I'm gonna go left to open up a new issue. I'm gonna say way finder, and I'm gonna open that up.
18:44And I'm gonna rename this to how videos market to TikTok. Short form attribute for TikTok.
18:54Then I'm gonna do the next here. I do want to make this less manual, but this is the way I'm doing it for now, and it seems fine. So this one is portrait creator view, and then this one is TikTok creation flow.
19:11Tick whoopsie. TikTok creation flow.
19:17So now I've got four sessions running at the same time, grilling about different things. And I'm gonna just sort of manage them as we go. So I've already got one from before I went on.
19:31How does this work with non public repos? It just works. Right?
19:34Because you've got a GitHub auth token, it's authenticated as you, then you're able to create issues and do the normal thing.
19:42To spec is the new name for two PRD.
19:49So at this point, I just wait until all of these are just running, kind of like figuring things out.
19:57And I just wait until one is ready for me.
20:04Much more clear? Good. Much more clear.
20:07Sorry.
20:11Is it good practice to compose a big skill that references other child skills to perform, for example, controlled loop workflows? It's tricky.
20:22I would say, look at my writing great skills skill. That's got a little bit of advice in there for that.
20:30Why don't I use Fable five? Because I'm really happy with Opus. Really happy with Opus.
20:36It's doing good for me.
20:44How do I have four terminal workers in one terminal window without tapping over? This is a Claude code feature.
20:50So it allows me to view it from the top level. I've really really enjoyed using this. It's excellent.
20:55I've tried other stuff like Herder, but it's Herder's a little bit buggy. I have some reports that I need to give give to the guy who makes it. But I've really enjoyed this.
21:05So okay. One is here, and it started on a grilling session. What I really like about PloreCode is it has this recap feature, where it says charting the TikTok creator Wayfinder map.
21:15I've claimed on grilling short form attribute how video is marked as a TikTok short. What axis is this attribute actually on? This ticket floats kind with values short and lesson, but those two aren't opposites.
21:26A standalone pitch video is neither a lesson nor a short. It's a long form landscape video. So this is a new should name form factor intent only.
21:35This video is a short form vertical thing. I'll post TikTok shorts statement. I think this is a standalone video.
21:41It's just a standalone video with a new attribute and might need to be filtered and shown differently in that same UI.
21:50Let me actually show you the thing that I'm working on so that I can I might just put this in oh, no? We're we're good.
22:00So the thing I'm working on is this.
22:07This is my basically, the thing that I spend all of my time in as a content creator. This is a content creator platform that I have built for myself.
22:19And I think what it's basically asking is should I put this in here in my stand alone videos? Probably should.
22:31Probably should.
22:36Yeah. I think that's good. So I go left, and now I go and do the other one.
22:43So what's here?
22:48YouTube shorts research done. Oh, okay. This is the original original map thread.
22:57And we can see here that the agents it kicked off to do the research are already already coming back in.
23:05That's really nice.
23:09So okay. And now a couple more are starting to come back too. So okay.
23:14Boolean is short.
23:19So here's the tension. The Reaper has a stated convention for picking enums over Booleans. Pause type, silence length.
23:24That's correct.
23:29I generally much prefer enums over booleans, especially in categories where it could expand outwards. So I think a type makes most sense here.
23:43I feel like the type might need a qualifier of some kind, like destination type or I don't know.
23:54But definitely an enum.
23:58So and that's I mean, that's my engineering experience coming into play there. I'd it would be very easy for a junior, someone not very familiar with coding, to just say, yeah, let's use a boolean here.
24:11Because the boolean is like, okay, is it a short or is it not? But what happens if we add a third type here? Then you would need a boolean to represent, okay, it's not a short, it's not a video, it's something else.
24:22You've got three booleans there, And three times two is six. Right? So you've got six possible states of all of those different booleans.
24:29Whereas if you use an enum, it means that you've got literally three possible states, because that single enumerated value can either be one thing or another thing or another thing. What is the enum name? I think format makes sense.
24:46Yeah. I like format with standard versus short. That's good.
24:58I tend to find that with code review, the harness matters more than the model.
25:09Okay. Now time to dip over to okay. There's a couple of others that come back.
25:14I'm going to just ignore this one for now.
25:21Okey dokey. So recap on this. Charting the TikTok creator map by resolving the ticket for low friction.
25:28Action. Waiting on your answer for whether TikTok creation should be footage first, shell first, or both. Oh, great question.
25:34What a good question.
25:37OBS import a pen's clips to an already existing video.
25:46What did come from at the moment of creation?
25:55Mhmm. Yeah. This looks good.
25:58It's basically saying how should you kick off this new video? What do you want to see when you do that?
26:06Yeah. I think what I want is to press a button and then to start talking. So however we make that happen, you're great.
26:12And I would love to then use AI to generate from the transcript the name of the video and any description metadata that needs to come with it.
26:23No. I'm not on Twitch.
26:33Yeah. We'll hopefully get to do some prototyping. I've got about forty five minutes left, so definitely we'll reach some prototyping.
26:41And in fact, we've got a prototype already. So this was a prototype ticket. And if we look back in the thread here, we can see we're on only 60 k tokens, which is nice.
26:52And it's using the prototype skill in my repo. I've got a video coming out about this this week.
26:58So where is this? It is in it's on a work tree.
27:07Let's see. Give me the exact commands I need to run-in order to see this.
27:14This is just using Wayfinder, and Wayfinder kinda leads you through this process. So let's go back to the main one.
27:24Constantly trying to kill dead time here, basically. So seating from the existing portrait signal.
27:32So is this no. I don't think that's right.
27:36It says that there's an existing signal in the code base, but I don't think that's right. I don't want to do a backfill here. I don't want to migrate any existing videos.
27:47I just want to let's assume that all videos are landscape for now.
28:01TikTok creation flow. Where should the top level list of TikTok videos live?
28:07So it's recommending a new top level sidebar item.
28:14I actually didn't know. And because I don't know, this is essentially a what should it look like question.
28:22And so I'm going to get in there. This is a what should it look like question, which means that we need to raise the level of fidelity so that I can actually see what this might genuinely look like.
28:36This is Opus 4.8 medium. No dead time. No dead time.
28:42Obviously, you can have a cup of tea if you like. But while I'm working, no dead time. Is there any auto derivation or format human I think format is purely human set.
28:55So this makes the migration easy because we just default to landscape, I suppose. I don't know.
29:02We can run a migration script, I suppose. Because having a default in future is a little bit awkward.
29:10I don't know. I'm up to you on that one.
29:17Okay. Here we go. So we just got a CD into this repo, and then run this dev server.
29:30Where is that? Dev server. In fact, I know I know what it is.
29:33Run dev react router. And then we open up this.
29:43So the way this works is I think this is gonna be over here.
29:51The way prototype works is I can basically you're gonna work on me?
29:59No. You're working slowly. Okay.
30:01The way this works is prototype will give you database URL, not set environment variables. Okay. Let me just do some environment variable fiddling until we get this working.
30:19Callboard trace. Okay. Okay.
30:23I think we're good.
30:29Now, what prototype does is it helps you raise the fidelity of the discussion that you're having. It allows you instead of just talking about text, to actually look and see some oh, yeah.
30:45Nice.
30:50Oh, really nice. Some proper UI to look at. So if we look at this here, click over.
30:57So that's a. And now if we click over to variant b oh, I think there might be a bug here. Yeah.
31:05I think I need to run npm install.
31:10Remove non directory. Yeah. Yeah.
31:12Yeah. Yeah. Yeah.
31:12It's done a symlink, I think. Done a weird thing. This is 20 x max.
31:18I have a single 20 x max subscription.
31:24Using Wayfinder. Good. Good to hear.
31:27Let's run dev again. This will be another little loading session.
31:33And then it's basically gonna give me a few options for different prototypes I can do.
31:42So bang. Let's try number b. Playback only stage.
31:49Come on, kid. What are you doing?
31:56Outdated depth? No. That seems fine.
31:59Come on, kid. Let me see b. Ah, here we go.
32:02Studio. Right.
32:05Split editing surface. I don't need to trim though. Portrait first tab editing.
32:11Where's the tab then?
32:21I don't love any of these, actually.
32:29Yeah.
32:36So okay.
32:39TLDR on Wayfinder is it helps you find your way. It's a update on grill. Me that helps essentially, instead of having a single grilling session, it's an orchestrator over grilling sessions.
32:52And I'm in a prototyping ticket, and I don't like any of these. Okay. Right.
32:59Well well, I don't like any of them, really.
33:11The one I really wanna focus on is the studio, because I think I want to reuse most of the video creation logic in that UI that I've already got, but I just need a different view on the right hand side.
33:28So currently, it's totally divorced from anything I've seen before, and it's not looking great. I need you to explore the real video editing code so that we maybe take the reducer that's already there, take the UI parts that are already there.
33:42We could maybe even try to componentize some of that stuff a bit better and pull some shared code out into this new view.
33:53Yeah. Let's try that. See what it says.
33:56Okay. So the original map thread. Now all of the research decisions are resolved.
34:02Beautiful. So if we look back at the map, then we should see, yeah, all four of those research tickets are now closed.
34:10So it's now gone and found the updated API. Yeah.
34:15Here we go. So is that all let's see up here.
34:19Yeah. Beautiful. So look.
34:22Now all of the research is saved inside these tickets here. So findings committed.
34:28So all of the stuff. And it's actually gone and created a local file on a separate branch.
34:35So created on this work tree branch. Beautiful. So that means that this is a really, really rich piece of context.
34:43Because the map here sort of indicates where the decisions are and points towards the research. The research ticket itself contains a link to the actual thick docs page that was created.
34:56One thing is I don't really like the fact we can't really read these very easily. I suppose we could if we just asked an agent to find it for us. And it's got essentially a summary here.
35:05So this is progressive disclosure at its best. We're pointing pointing down the chain until we actually get to the thing that we need. Fully decoupled.
35:15Every platform has a usable upload API.
35:21Chunked. Interesting. Okay.
35:24So it's pretty cheap, actually, to post to Twitter. Infers a short from the file. That's fine.
35:31Already short capable with no landscape assumptions. Awesome. So lots of research done here, which is pushing back the fog.
35:41Could you make an updated video how to set up the whole thing you're using there? Just follow the install steps in my GitHub repo.
35:49Okay. So I can close the original map thread because that is all done. Now, how does format surface in the video lists and tables?
35:58So we're just grilling here, basically. You know, this is just grilling across multiple sessions about the same thing. And because they're in parallel, you end up being faster.
36:08It is not dependent on GitHub. You can set it up any way you like.
36:17Yeah. I changed the way research worked this morning, so you're probably seeing something different to how it works in real life now.
36:25So how does format surface in the video list table? So filter. I think both a filter makes sense and a visual marker.
36:34I think maybe even tabs might make sense here. I think we might need a prototype to have a look at this.
36:42One really cool thing about Wayfinder is one when you're in a ticket, you can schedule other tickets. Right? You can put more tickets onto the map.
36:51That's one thing that really annoyed me about grilling. Is, like, if I found something that I knew I couldn't tackle in this session, how would I handle it? I would probably just, like, hand off to another grilling session and do that in line, but there was nowhere tracking all of it.
37:07Yeah. Hey. Yep.
37:10Sounds good. So it's it's now telling me that because I've said we need a prototype, it's going to close the decision ticket and write down everything that we found into that decision ticket, and then open up another ticket to do the prototyping in.
37:28So is this all still working?
37:33What is going on in here? What are you guys doing?
37:39Let's see.
37:44Why don't you chill out for a minute and let me figure out what's going on?
37:52Give me the exact commands I need to run-in order to get it working.
38:01Or is it the same yeah. This is the same one as we had earlier.
38:10Right?
38:13This is this is infra behind my course creation, basically. So this is the app.
38:21This is the course I'm currently putting together, which is an AI coding crash course. And it's a video editor. It's a thumbnail creator.
38:30It's a a writer. It's a pitch store.
38:33It does everything. It's wicked.
38:37Okay. That's in the workspace. So I should just be able to just see it.
38:43Right?
38:46Where did you put it?
38:51TikTok's prototype. Interesting. Okay.
38:54So you put it on a new route. So I usually find these prototypes work way better when you're when it's integrated into the actual code, it's gonna be genuinely integrating in.
39:07Is this in here? Is it canceled? Yes.
39:12Let's try that.
39:18But for this one, it seems to have just stuck it on a new route, which is okay.
39:28Page you're looking for doesn't exist or has moved. TikTok's prototype.
39:40Okay.
39:50Okay. I'm in a different zone here. So let's go npm I.
39:56On the painful part about this is still the WorkTrees. Like, just handling all of the state between the different WorkTrees. Making sure they've got dependencies installed.
40:06One thing I really want to do is put a lot of this on a remote server or use sandboxes for this.
40:14Because it's just brutal the way it's working right now. Because I need to go up, I need to copy the environment.
40:22What's oh no. It's done assembling.
40:25Okay.
40:28NPM run dev. Address and use. Okay.
40:34Okay. So I think we're there. It's a lot of guff, basically, when you're working with WorkTrees.
40:42Whereas having them on separate sort of virtual machines or machines in the cloud would be really nice. Wow. Look at this.
40:50That's a fun prototype. Press to talk. Hit record and start talking.
40:56So if I click I mean, wow. So the theory here is that I would I mean, that's very friendly.
41:03It doesn't really fit the rest of the UI that's going on here. Is there some kind of bug that's happening? No.
41:10Let me see variant b, please. Cool. Okay.
41:15This is nice, actually. I like this. It's friendly.
41:19Press a new TikTok. Grid. Yeah.
41:24I mean, that's this is actually nice. I actually like this.
41:30I actually really like this. I don't know. Which one which one do you guys prefer?
41:33I'm I'm between a or b.
41:39But this is, like, delightful. You know, this is stuff that I would not have put time into or invested in. But it's just nice to get a kind of, like, oh, Yeah.
41:50I like number two, think. B. Okay.
41:54Really cool.
41:57Right. Looks great. B is the one for me.
42:05Don't think I've got any feedback on that, amazingly. Actually, no. I do a little.
42:09So this now, when you're in a prototyping phase with Wayfinder, you really need to hone in on detail. So this is the moment that you really start hammering it.
42:20So The new TikTok button in the top right is not needed since we've already got a duplicate on the page.
42:47There is some UI there that I'm not sure that we can represent in database state. So the rendered versus posted. Currently, store that stuff in local storage.
42:58And so I do like that UI, but I wonder whether we need a grilling session to figure out how that hits the schema and how that fits with what we have already.
43:16I like it though. I mean, it's I mean, it's nice. It's done a good job.
43:23Okay. So nearly here. Okay.
43:26Decision fully grilled resolution comment post on the ticket. Good. Okay.
43:30So we've closed another ticket in the decision map or in the map. Is it closed? No.
43:38Okay. Here's what remains. Needs inputs.
43:44May I proceed? Yes. Crack on.
43:48Okay. So it just needed input from me to ask whether it was okay to proceed. Fair enough.
43:54Portrait view. What's this now?
43:59Fully drop. Tab contain no context for you. That's a clean seam.
44:04Lift as is.
44:08OpenTaste question. What actually lives in that new panel? Okay.
44:12So it's asking me about this is the one where we're looking at the editing experience, not the sort of top level list experience.
44:20And what actually lives in that new panel TikTok specific concerns look like caption, vertical render trigger, post to TikTok shorts.
44:30So yeah.
44:37That's interesting. Because the previous TikTok implementation I had rendered its own captions, but we're now actually rendering captions ourselves, but slightly lossier.
44:50New viewer third tab inside editor side panel, a separate right column.
44:58I think this will be a separate right column that replaces the player heavy right side.
45:09I'm really struggling to picture what you're saying from your words, and so I just need to see some more prototypes, I think.
45:17I do not use Caveman. Now, I tried it out in my repo, but I didn't end up using it.
45:27Why do I only rely on skills? Couldn't I create specific agents for different tasks? Agents are just, like, new context window plus skill.
45:36That's what they are. So there's no reason to use agents when you can just manage that yourself with skills.
45:45I'm using Claude agents here. So I just click, Press left, and it goes to the Claude agent view. Looks good.
45:52Okay. So now 1266 is closed.
45:55Is it? Yes. It is.
45:56Very cool. And that has opened up a new ticket. So I'm gonna grab this.
46:01I'm going to close that with control x. I'll go way finder, and let's go for that one.
46:08And this is how the CVM produces the submitted vertical. Okay. So this is the render caption pipeline.
46:15So we have progressed down on the way.
46:19Right? We are finding our way. So this one is now header button removed.
46:28Defer to dedicated TikToks. Okay.
46:36So it's this is asking about the how we create them.
46:45Variant b's card contents. Yeah. Let's look back at it.
46:49Posted TikTok and shorts. It's okay, I think. I think it's quite nice.
47:04Yeah. Right. It's asking whether it should create a TikTok's thing here.
47:08I think that probably is cleanest because this is pretty sparse right now. It can take another video.
47:20Yeah. Let's go dedicated TikTok's sidebar item.
47:26Agree with your other suggestions.
47:30Okay. So we're heading towards the end of that ticket. That's good.
47:39Yep. Wayfinder is connected to domain modeling.
47:48Used it wrong to make one of your projects. Instead of using the Wayfinder skill, always use the skill the issue says to. Okay.
47:52Yeah. Whenever you pick up a decision ticket, you gotta use Wayfinder. Okeydokey.
47:59Where does the render pipeline live? So this is the decision that we've gotta make about where to put that.
48:16Yep. So it's basically asking, I've got this separate thing in a separate repo, and it's asking, should we pull it into this repo? There's no reason to, I don't think, because the thing that's in its separate repo works right now as a bridge.
48:33So I think we should probably defer this work to pull it into this repo. The reason being that we're not really set up to be a mono repo in this repo anyway, and that thing has some pretty old dependencies.
48:50So I kind of just want to keep it at arm's length for now with the idea that we might pull it in at some point.
49:05I feel like we might be able to not have to do the t t shell out.
49:12I think we can just is there an option where we can just invoke it? I don't know.
49:22I feel like we might be able to pull in parts of the system without needing to pull in the entire Remotion setup.
49:33Yeah. Whisper flow.
49:39Okay. Here we go. Oh, tracker rights got blocked by the auto approval.
49:43That's another thing about not being on a virtual machine is that I've gotta use auto mode. Auto mode is really annoying because it does, you know, does things like this where the tracker rights got blocked by the auto approval classifier.
49:55So, yes, go ahead. Let's see if auto tracker does it now.
50:02Love the new wayfinder? Great.
50:09Still a 50% improvement. Good. Okay.
50:12Now now now now.
50:17Let's check this out.
50:23And what's the URL? This is the new studio version.
50:27So another prototype. Okay. I love I freaking love opening a prototype.
50:37Just love it.
50:44Hey. That's not bad. That's not bad.
50:47That is what my editor looks like.
50:55It's not bad. One thing I don't love is these tall timeline items over on the left hand side. We don't need the captions down the bottom.
51:04Oh, but this is just this is just a. Let's see what b is. B is tabbed.
51:09Where's the tabs?
51:14Oh, I see. Render post node.
51:17I don't like that. Clear forward overlay tab to edit captions. Woah.
51:24Look at that. I can literally click on the captions to edit them. That's pretty wild.
51:32I don't I mean, I don't need that, because I I can usually just right click on these to edit them. Okay. So what do I like here?
51:46Okay. I've got an idea. So I think I like the full screen ish version of the player on the right.
51:53I think the stuff on the left is great. Don't need to touch it. What I want to see though is I want an action menu for the actions that are in the b section.
52:04So the current video editor has an action menu, and I would like to copy that as much as possible. There are some stuff I don't need in there, but have an explore in there to see what's in there.
52:18Prototyping is always super heavy on context. Always burns a ton of context, because it's having to read a bunch of files, write a bunch of files, and you're often needing to edit stuff. So I'm used to things going to, like, 200 k.
52:34And actually also, another thing about this is I don't need caption editing at all.
52:42I don't need it. Because the Remotion workflow that we're pushing out to, that does its own captioning anyway.
52:52So we just don't need it.
52:57Okeydoke. How you'll attach now you to the exact remotion capture look? Oh, here we go.
53:03Okay. What's it doing here? Render styled.
53:10Yep. Whisper transcription's already oh, here we go. Here we go.
53:21So we're trying to work out how much we should figure out here. Yep.
53:29I am actually very attached to the exact Remotion caption look. It took a long time to get it right, and I think it's really pretty. So, yeah, I don't want to get rid of it.
53:42So much of, like, misalignment from AI is AI not asking these questions, not understanding what your values are, what you prioritize over other stuff.
53:52I do not use OpenSpec.
53:58Okay. We've got a closed ticket, I think. That's good.
54:01Let's have a look at the map. We're getting through it. We are six out of nine now.
54:07And we've got that's been grabbed.
54:13This one needs to be grabbed now. So we can delete this one. We can say way finder, that one.
54:19And this is posting life cycle.
54:30Still can't wrap your head around how to have multiple grilling sessions in parallel when the decisions in one might affect decisions in the other one. In general, I haven't found that to be a problem. It can happen.
54:41And when it does happen, you just sort of say, okay. You point the sessions at each other and figure out which one needs to be done first.
54:49Then you just finish that grilling session in there, hand off one to the other. You know, you just do a bit of inter agent communication.
54:58The way it generally works is that it generally gets the blocking relationships really nicely. Okay.
55:05Here we go.
55:09So CVM is unambiguously a local desktop tool. Stream Deck Forwarder, OBS, blah blah blah blah. Arm's length shape crystallizes.
55:18Here's my recommended answers to the whole ticket. Here we go. Extract just the myComparieMotion project into its clean standalone repo.
55:27Yes. CVM invokes it via local shell out. Yes.
55:33Good. CVM owns the surrounding glue natively. Transcribe audio file.
55:41Yeah. That seems fine. Defer the eventual pull into CVM at later optional step.
55:48Yes. Yes. Yes.
55:50That looks freaking great. I love it.
55:58Yep.
56:03The floating bar that stitches between mock ups, that's just the prototype skill.
56:12Okay. Of course, that's doing a little bit of prototyping. Cool.
56:18And you get a little bit of a break sometimes just where all where you're at maximum agent working, and you just get to have a cup of tea.
56:29This is Claude Agents. It's really good.
56:35How do I do UI testing? I mostly like this prototyping phase produces really high quality assets.
56:42I haven't done a ton of browser testing. I mean, I have done in my career, but I'm not doing it now. I probably should.
56:49I just this is a really complicated setup locally, and so I I don't have the energy to want to give it its entire set of microservices that make that work, basically.
57:04Okay. So what's converged now? Okay.
57:07So let's have a look at the prototype. Yes. That's nice.
57:13That's nice. Reading short. Oh, that's really nice.
57:18And then press post. God, that's gorgeous. So what happens when you press post?
57:24Nothing so far.
57:28God, I do love this.
57:37That's really good. The video is slightly too long vertically, so it's cut off at the bottom.
57:47So you need some CSS adjustment there. I'm also intrigued by what the post button should look like or whether it should open a modal and what should be in that modal.
58:01But I think that's a grilling ticket, not a design ticket. And we're already pretty deep in our context window, so let's not worry about that yet. So I think oh.
58:15Oh, I didn't even realize there were options.
58:22Post actions above. Oh. On the video.
58:25Look at that. On the video is quite nice. I know.
58:29I think above is just great. Yeah.
58:34A is the one I'm picking. Doing above the video looks really nice.
58:43I'm just using Opus 4.8 medium for pretty much all steps in the workflow.
58:50I've not heard of Auker IDE.
58:54Auker IDE. Okay. Lifecycle one linear status or a spine orthogonal facets.
59:03So I think what we need to keep track of here is what platforms the video has been posted to. And from there, we can figure out in the UI whether it has been posted to all of them.
59:19We can figure out from the videos on file system status whether it has been exported or not.
59:37Why do I do medium over x over high? There's a trade off with effort. Right?
59:44With what you're trading with effort is you're trading quality versus latency. The higher you go, the more tokens are spent and the higher latency you get for diminishing returns in terms of effort or in terms of quality. So I tend to I've just found that medium fits the way I like to work best.
1:00:04I feel like I don't need a very smart model, and I feel that Opus Medium is perfectly smart enough for me.
1:00:15Okay. What's the shape of that posted platform record? Where does it live?
1:00:20Platform remote ID is nice. Yeah.
1:00:24Remote URL is nice.
1:00:29Yeah. I really like that.
1:00:36I really like that.
1:00:39Each post carries more than a boolean. Yep. Yeah.
1:00:43That's excellent.
1:00:51Note, by the way, that we are vibe coding here. We're just vibe coding within prototypes. Right?
1:00:55Those prototypes are then going to go and feed into a spec, And that spec is going to be the thing that we get that gets shipped. I've got about ten minutes left, by the way.
1:01:06Render contract between CVM and the extractor renderer.
1:01:13See if we can pass this props explicitly. Yeah. If that's possible, then great.
1:01:23What post it mean for the final version? Okay.
1:01:32B. Okay. So TikTok's apps draft.
1:01:37So you finish posting by hand of the TikTok app. Public posting audit gate. What's that?
1:01:45I'm lacking a bit of context here. Can you tell me what the public posting audit gate in TikTok is?
1:01:56Usually, what I do is I create a PR that references the issue. And then when the PR gets merged, the issue just gets closed.
1:02:07Yep. This workflow should work in OpenCode.
1:02:12Where does the render actually execute locally or Remotion Lambda? Executes locally.
1:02:21I'm using WhisperFlow to dictate.
1:02:28Code based design should be part of improved code based architecture, which should be something that you run every so often on your code. And maybe it's part of review, but review kind of has its own baked in concept of what it's looking for.
1:02:45So, yeah, maybe during planning is a good one.
1:02:52Where would you recommend putting a shared kernel of knowledge? Probably, you're talking about a knowledge base.
1:02:59Maybe in its own Wiki repo, or in a third party service, like Notion or Google Docs or something.
1:03:10How do I recommend teams work together with their dot scratch? I recommend not using local markdown. I would this is brownfield code.
1:03:19Yes. We're using way way finder on brownfield code. I think that brownfield versus greenfield is is kind of b s, actually.
1:03:26I don't think it's as big a difference as people think.
1:03:34Option a, sharpened. Extract a local shot out, see if remove the glue. Yes.
1:03:39Okay. Looks fabulous.
1:03:44Cool. So that is a grilling ticket figured out.
1:03:48We're really getting close now. So pick variant a.
1:03:53That's another prototype ticket that we've got done. Posting life cycle. Okay.
1:04:00So okay. TikTok.
1:04:03TikTok's content posted API treat every app as an unaudited until TikTok reviews and approves it. That review is called the audit.
1:04:13What unauditly cost you, every post is forced to yeah. That's fine.
1:04:23You must build TikTok's full mandated content sharing guidelines UX inside CVM before they'll even review you. A privacy drop down with no default. Grayed out disabled.
1:04:32Bloody hell.
1:04:42Okay. Damn. This is brutal.
1:04:45So it means that TikTok needs or needs to audit every post I send or something, or there's some onerous requirements to do this.
1:04:58Ouch. So this is something that we've hit now that is a real blocker for us. And I do have an idea for how we can get around it.
1:05:09Yeah. This works in codecs. So okay.
1:05:14I did have an implementation of this before, and it used buffer to queue up videos, which is kind of nice.
1:05:23And I think what I wanna do is to make a decision ticket to figure out whether we can use buffer here instead. Because the buffer API is easy to use.
1:05:36Although, actually, the way I did it was I uploaded it into a Dropbox, then Zapier detected that, and then uploaded it to buffer.
1:05:46And that's fine. I can do that again. That makes our flow a lot easier.
1:05:50We don't actually need to look at any existing APIs, but I will need to pay for buffer, of course. But paying for buffer is a small price to pay compared to having to do this crazy TikTok stuff.
1:06:03Let's have a chat about that now before we figure out whether we're gonna put a decision ticket.
1:06:10Yeah. Everybody's grilling me, man. Everybody's grilling me.
1:06:18Yeah. These tickets are just a way of tracking the progress through the fog, essentially.
1:06:29Alright. I got about five minutes left.
1:06:38CVM tracks a single binary queued handed off fact per video. Hang on. I might need to go fairly soon.
1:06:46I got five minutes.
1:06:49Okay. I tell you what. Let's wrap up, because I'm not gonna get to the end of the map.
1:06:54This is a big chunky feature. We've done pretty well. We've got eight out of nine tasks done.
1:07:01And you can see that the decisions so far in the map itself have been fleshed out a bit, and each of them link out to a sub issue. Hang on.
1:07:13Two seconds.
1:07:48There we go. Car related issues. Sorry.
1:07:52I'll be down in a sec. So we've done pretty well.
1:07:59We've finished up. And this and let me refresh the map just so we get the full sense.
1:08:07We can see that decisions are so far have been recorded in the proper map. They all link out to research stuff. We've done been able to get a huge amount of work done in pretty little time.
1:08:20It's pretty sweet. I love doing this stuff. I love the feeling that Wayfinder gives me, that I'm always in control, that it's kind of like figuring out the way for me, and I just have to handle the tickets as we go.
1:08:33It's pretty great. You here for a skive? Good for you, Fuzzy.
1:08:39So I'm I would love to hear how you get on with Wayfinder, and what this new feeling gives you.
1:08:48Because what I know is when I get to the end of a Wayfinder spec, I end up with a spec oh, sorry, a Wayfinder map.
1:08:57I end up with a spec that's incredibly detailed and links back to all of the decisions that I made. And all of the decisions that I made are in minute detail.
1:09:10To spec seal is identified as high risk. I need to figure that out. I saw that.
1:09:15I do still use sandcastle. Yep. Sandcastle drives all of my implementation.
1:09:22I use Opus 4.8 medium.
1:09:28With the Orchestrator, we don't have to manually invoke other skills.
1:09:34As soon as I get to the spec, then I take the spec, and I turn it into tickets.
1:09:40And those tickets are I can probably find a couple. They're usually so I've got wow.
1:09:471,000 closed tickets here.
1:09:51Where are we? Yeah. Here we go.
1:09:54Here's a spec. Diagram content search on the diagrams page. This was from a Wayfinder thing, I think.
1:10:00Yeah. There we go. So all of these little links out, these are Wayfinder grilling tickets.
1:10:06There's another Wayfinder research there. And so these are context pointers that the agent can walk down to find more information. Super good.
1:10:14And we can see here that it had six subtasks, each of which are steps that it's going to take. And then what it did was it produced this PR.
1:10:26And this PR is about 5,000 lines of code. See if I can find the GitHub actions that worked on it.
1:10:36Can I do that? I'm not sure if I can. But essentially, a GitHub action just worked through each ticket in a separate session, and bam, produced all this code.
1:10:49Which OS am I on? I'm on WSL.
1:10:55And that's it, folks. Hopefully, this gave you a sense for what you can do with Wayfinder, and the process that you walk through it.
1:11:03I think the most confusing thing is the idea that you use Wayfinder for everything. You don't use any of the sub skills. And yeah.
1:11:11We're good to go.
1:11:14You can absolutely use it to use linear instead. You can do Todoist. You can do Jira.
1:11:19Whatever you want.
1:11:24I'm gonna slow you down, pal.
1:11:38Declarative partial updates. I don't have a schedule for going live. I try not to have any schedule for content creation, because I hate the feeling of being on a treadmill.
1:11:49What I like the feeling of doing is producing courses, producing good stuff, working on the skills, like the my open source work always feels like the highest priority thing to me, because everything else flows from If I'm not shipping stuff, then I don't have anything to create content about. So that's the idea.
1:12:07There is an open ticket to handle multi repo setups on Sandcastle, and I'm starting to feel the urge myself.
1:12:15But, yeah, I haven't implemented it yet. Right. I better go and fix my son's car.
1:12:21Thanks so much for joining. Oh, yeah. Let me let me just finish off.
1:12:26So what I would do after this is I would try to get the wayfinder stuff, get the map to completion, so complete all the tickets.
1:12:36Once that's done, then I would turn it into a spec, width to spec, and then close the wayfinder map, because it don't need it anymore.
1:12:46And then to tickets, so I turn that into subtasks. Then I would implement each subtask in a separate context window, usually with an AFK agent, because I'm not needed for that.
1:12:57And then I do a code review on the whole diff, and fix any problems.
1:13:03And then it goes to human review. There we go. If you have any questions about this, by the way, then you can use my ask Matt skill.
1:13:11That's the place to go for feedback. And if you wanna keep up to date with this stuff, then you should go to AI Hero. Subscribe to my skills.
1:13:20Yep. You just give the map ticket to to spec. Fresh session.
1:13:24Just get it to explore the the map, and turn it into a big spec.
1:13:32So yeah. Just one more. So, no, these there's a difference between the decision tickets that Wayfinder makes, and the implementation tickets that the tickets creates.
1:13:43Because the decision tickets can only be resolved by resolving the decision, by making the decision. Whereas the implementation tickets are when those decisions are reified in the code or represented in the code.
1:14:05Love to see the rest of the flow. Yep. Yeah.
1:14:09You often have to ask if you're done yet or not. So that is annoying, and I agree. What I do is I generally just, like, watch the I just watch the map as I was doing here, and I just look at the sub issues, and I can just see visually what's needed and what isn't.
1:14:30For back end, yes, you can use it. For front end, you can use it. For middle end, you can use it.
The Hook

The bait, then the rug-pull.

The pitch is small and the promise is large: one hour, one vague idea, and a tool the host calls an absolute banger. What follows is not code but a method for turning fog into a spec, narrated live while four AI sessions grill him in parallel.

CTA Breakdown

How they asked for the click.

VERBAL ASK
1:13:20subscribe
“If you want to keep up to date with this stuff, go to AI Hero and subscribe to my skills. And if you have questions, use my ask Matt skill.”

Soft and late; the demo delivers value for over an hour before any ask, and the CTA points to the open-source skills that the whole stream implicitly promotes.

MENTIONED ON CAMERA
1:13:20productAI Hero skills subscription ↗
Storyboard

Visual structure at a glance.

cold open
hookcold open00:00
workflow chain
promiseworkflow chain11:40
mental model
valuemental model14:28
map + tickets
valuemap + tickets15:24
first prototype
valuefirst prototype27:32
studio editor
valuestudio editor47:08
posting blocker
valueposting blocker59:16
finished spec PR
ctafinished spec PR1:11:20
Frame Gallery

Visual moments.

cold open00:00Frame at 01:24 from LIVE: The /wayfinder Demo01:24Frame at 02:20 from LIVE: The /wayfinder Demo02:20Frame at 03:16 from LIVE: The /wayfinder Demo03:16Frame at 04:12 from LIVE: The /wayfinder Demo04:12Frame at 05:08 from LIVE: The /wayfinder Demo05:08
cold open00:00Frame at 01:24 from LIVE: The /wayfinder Demo01:24Frame at 02:20 from LIVE: The /wayfinder Demo02:20Frame at 03:16 from LIVE: The /wayfinder Demo03:16Frame at 04:12 from LIVE: The /wayfinder Demo04:12Frame at 05:08 from LIVE: The /wayfinder Demo05:08Frame at 06:04 from LIVE: The /wayfinder Demo06:04Frame at 07:00 from LIVE: The /wayfinder Demo07:00Frame at 07:56 from LIVE: The /wayfinder Demo07:56Frame at 08:52 from LIVE: The /wayfinder Demo08:52Frame at 09:48 from LIVE: The /wayfinder Demo09:48Frame at 10:44 from LIVE: The /wayfinder Demo10:44workflow chain11:40Frame at 12:36 from LIVE: The /wayfinder Demo12:36Frame at 13:32 from LIVE: The /wayfinder Demo13:32mental model14:28map + tickets15:24Frame at 16:20 from LIVE: The /wayfinder Demo16:20Frame at 17:16 from LIVE: The /wayfinder Demo17:16Frame at 18:12 from LIVE: The /wayfinder Demo18:12Frame at 19:08 from LIVE: The /wayfinder Demo19:08Frame at 20:04 from LIVE: The /wayfinder Demo20:04Frame at 21:00 from LIVE: The /wayfinder Demo21:00Frame at 21:56 from LIVE: The /wayfinder Demo21:56Frame at 22:52 from LIVE: The /wayfinder Demo22:52Frame at 23:48 from LIVE: The /wayfinder Demo23:48Frame at 24:44 from LIVE: The /wayfinder Demo24:44Frame at 25:40 from LIVE: The /wayfinder Demo25:40Frame at 26:36 from LIVE: The /wayfinder Demo26:36first prototype27:32Frame at 28:28 from LIVE: The /wayfinder Demo28:28Frame at 29:45 from LIVE: The /wayfinder Demo29:45Frame at 30:06 from LIVE: The /wayfinder Demo30:06Frame at 31:04 from LIVE: The /wayfinder Demo31:04Frame at 31:51 from LIVE: The /wayfinder Demo31:51Frame at 33:08 from LIVE: The /wayfinder Demo33:08Frame at 34:04 from LIVE: The /wayfinder Demo34:04Frame at 35:00 from LIVE: The /wayfinder Demo35:00Frame at 35:56 from LIVE: The /wayfinder Demo35:56Frame at 36:52 from LIVE: The /wayfinder Demo36:52Frame at 37:48 from LIVE: The /wayfinder Demo37:48Frame at 38:44 from LIVE: The /wayfinder Demo38:44Frame at 39:26 from LIVE: The /wayfinder Demo39:26Frame at 40:36 from LIVE: The /wayfinder Demo40:36Frame at 41:32 from LIVE: The /wayfinder Demo41:32Frame at 42:28 from LIVE: The /wayfinder Demo42:28Frame at 43:24 from LIVE: The /wayfinder Demo43:24Frame at 44:20 from LIVE: The /wayfinder Demo44:20Frame at 45:16 from LIVE: The /wayfinder Demo45:16Frame at 46:12 from LIVE: The /wayfinder Demo46:12studio editor47:08Frame at 48:04 from LIVE: The /wayfinder Demo48:04Frame at 49:00 from LIVE: The /wayfinder Demo49:00Frame at 49:56 from LIVE: The /wayfinder Demo49:56Frame at 50:31 from LIVE: The /wayfinder Demo50:31Frame at 51:48 from LIVE: The /wayfinder Demo51:48Frame at 52:44 from LIVE: The /wayfinder Demo52:44Frame at 53:40 from LIVE: The /wayfinder Demo53:40Frame at 54:36 from LIVE: The /wayfinder Demo54:36Frame at 55:32 from LIVE: The /wayfinder Demo55:32Frame at 56:28 from LIVE: The /wayfinder Demo56:28Frame at 57:24 from LIVE: The /wayfinder Demo57:24Frame at 58:20 from LIVE: The /wayfinder Demo58:20Frame at 59:16 from LIVE: The /wayfinder Demo59:16Frame at 60:12 from LIVE: The /wayfinder Demo60:12Frame at 61:08 from LIVE: The /wayfinder Demo61:08posting blocker62:04Frame at 63:00 from LIVE: The /wayfinder Demo63:00Frame at 63:56 from LIVE: The /wayfinder Demo63:56Frame at 64:52 from LIVE: The /wayfinder Demo64:52Frame at 65:48 from LIVE: The /wayfinder Demo65:48Frame at 66:44 from LIVE: The /wayfinder Demo66:44Frame at 67:40 from LIVE: The /wayfinder Demo67:40Frame at 68:36 from LIVE: The /wayfinder Demo68:36Frame at 69:32 from LIVE: The /wayfinder Demo69:32Frame at 70:23 from LIVE: The /wayfinder Demo70:23finished spec PR71:24Frame at 72:20 from LIVE: The /wayfinder Demo72:20Frame at 73:13 from LIVE: The /wayfinder Demo73:13Frame at 74:12 from LIVE: The /wayfinder Demo74:12
Watch next

More from this channel + related breakdowns.

11:25
Matt Pocock · Tutorial

I Open-Sourced My Own AFK Software Factory

Matt Pocock built and open-sourced Sandcastle, a TypeScript library that runs Claude Code and other coding agents inside sandboxes to plan, implement, review, and merge whole GitHub issues without a human clicking approve.

April 30th
Chat about this