I stopped using /grill-me for coding. Here’s what I use instead:

A few months ago, I wrote a few sentences, about four sentences that have turned out to be the most influential four sentences I've ever written. I packaged these four sentences up into the grill me skill, which is a skill that you can use to get the LLM to interview you relentlessly. It interviews you until you reach a shared understanding, walking down each branch of the design tree, resolving dependencies between decisions one by one. I know this skill is influential because every single day I receive about five messages of people saying they've tried it and they love it. This skill is an absolute gamecher. What are your favorite skills? Grill me is great. I'm working on a project. Blah blah blah. The grill me skill asks me about ambiguities. Fantastic. Grill me skill is goated. At first, I felt like it slowed me down with all the questions, but after using it bit, I honestly think it might save time. You just one-shot everything after you've kind of gathered all the context and you tested the skill called grill me and blah blah blah blah blah. It's wonderful, wonderful, wonderful, wonderful. And after all that praise, you might think, well, you should probably stick with that skill, shouldn't you? That skill sounds pretty good. And it turns out I've actually built a better one. I'm never very happy when I'm resting on my laurels. I always feel like there's improvements to be made at every single part of my process. And now grill me has been replaced with a new skill. Let's open up a session so that I can explain a little bit more about where grill me goes wrong. I'm going to paste in a prompt that I've already added here. And this prompt is an idea for a new feature. I've just dictated this out. So you just sort of be spared the details of it. But essentially I want to create a new entity in my database and a new entity that my app is going to deal with. Currently this application deals in courses, deals in lessons, deals in videos and you know sections and a few other things. And I want to add a concept of pitches. There's this kind of Mr. Beast style axiom where you should be thinking about the packaging for your video before you actually work out what's going in the video. And that's what a pitch is in this kind of setup. A pitch is really just the packaging for the video, the title, the description, how I'm going to frame it to people. And I create a bunch of these pitches and then pick the best ones and then turn those into videos. Now, what you notice here is as I'm communicating with the agent, we're really focusing on language, right? We're really focusing on what is a pitch. I just had to communicate that to you so that you could follow along and the agent will need to extract that information from me too. But there's also some extra jargon in here that the agent doesn't know about yet. For instance, I've talking about standalone videos. What does a standalone video mean? Oh, of course it means a video that's not connected to a lesson or a course. Now, of course, I know that. That's kind of like a term of art for using this whole setup, but the agent doesn't know that yet. It doesn't have any concept of what that is. So during the grilling session, it's going to have to ask me what a standalone video is or try to figure it out from the code. So as I used grill me more and more and more, I would start to notice these times where the agent was being really really verbose and I would have to remind it, no, there's already a term for that. And often though there wasn't a term or I was kind of thinking about things in a very verbose way myself and that wasn't being challenged by the agent or we would actually land on some really good shared language and then that wasn't documented anywhere. So I started to feel dissatisfied with grill me because there was this piece missing from the puzzle which is we were able to communicate about the code pretty effectively but I would have to reexlain all of the nonobvious things about the codebase and about the domain the problem that we were solving before we could do anything productive. So I started to think to myself what is the thinnest layer of documentation I could use to just give the AI a bit more of a leg up. So I came up with this skill, the ubiquitous language skill. Ubiquitous language is an idea that comes from domain driven design. This is the big blue book by Eric Evans that everyone goes on about. And what it does is it's essentially you're trying to create a document which is the language that's used by the codebase, that's used by developers, and that's used by domain experts. In other words, people that know about what you're building, but not how you're building it. All of those three groups should be using a shared language because that means that the domain expert can go okay there's something wrong with this particular section of the app. The developer knows what they're talking about and the code also reflects that. So what I would do is in the middle of a grilling session when I noticed that we were needed to sharpen some language I would use the ubiquitous language skill and call it with you know ubiquitous language and try to create a ubiquitous language.mmd as we were going. So, I had grill me and I had ubiquitous language and I was using them both at the same time. And I realized, wouldn't it be great if I just combined the two into a new skill? And here is that new skill. It is grill with docs. It has exactly the same text as grill me at the top here, but it has a couple of extra pieces. The first thing it has is the ability to look for a context.md file. And this context.md file will document all of the shared language that's inside that context. Now context is like super overloaded. So I'm a sort of uncomfortable but maybe okay with it. It's essentially a bounded context in DDD is a part of the app in which you speak a shared language. So if you have a massive monor repo, you can have a context map here and have many different contexts inside. So that's how you would scale this to an enormous repo. But still if you just have one pretty big repo where all the application is speaking the same language and the domain experts speak the same language then you can just use a single context.md here. So it's instructed to look for this existing documentation to pull in this shared language and then during the session it's got some extra additions here to challenge uh language usage against the existing glossery to sharpen fuzzy language discuss concrete scenarios cross reference with code and update it as you go. So this essentially helps you really sharpen your language as you're using the grill with doc skill and this pays off as you go. I was asking some folks for feedback on this and I got some really nice quotes here. So this guy used it for the whole of today and at the start it asked him to define a lot of terms. Some terms were hard to agree on and ones he would most definitely forget but four or five sessions in he started noticing that Claude was picking up the context during the grill session and it magically aligned with the thoughts I had before the words came out of their brain. So that's what you get out of this. By documenting the non-obvious stuff, by agreeing on a shared language, you really can nail down and get a magical alignment between you and the AI where you just have to use far fewer words to communicate what you mean. For instance, here's the one that I have in my repo here. We essentially just have a little description about what the um you know what the repo is. Then we have a course and a course repo and we have all of the entities inside here, course versions as well because I have multiple versions. And if we look for the one that we were looking at before, which is standalone video, it is just down here. So we have an exact specification for what standalone video means. Now, now the grill with doc skill knows to look for this. But I also add a context pointer into not inside that claw. MD, but inside the local claw.md here. So we have just this domain docs, a single context layout, context.md at the repo root. and you see this extra little bit of uh documentation for more information about where this stuff is. One final thing that Grill with Docs does is that there are some things that sharpening the fuzzy language will help with, but there are some things that it won't. And so I wanted a layer that would explain all the nonobvious decisions that weren't able to be captured inside context.md. And so for that, I've gone with an architectural decision record. These ADRs here are really like simple markdown files that sit in your repo that essentially document all of the nonobvious decisions. You only want to create an ADR when the decision is hard to reverse because if it's just like, oh, we use this library instead of this library and they're kind of interchangeable, then you can always just swap later. It would be surprising without context, and plenty of decisions in a repo are surprising without context, especially more complex ones, and the result of a real trade-off. In other words, that this decision has consequences down the line. And I've got an ADR format inside here that um the LLM uses when it creates these ADRs. So now we understand all the pieces. Let's go back up to here. Let's replace grill me with grill with docs. And let's actually start this grilling session to see it in action. All right. So the first thing it has done is it said context.md is rich. Standalone video is already defined as a lessio a video with lesson ID equals null. and it says before going further I want to surface attention with the glossery. This is what you'll often find with grill with docs is that it really focuses on the language before you then actually go and talk about implementation details. It says there's cardality between pitch and standalone video. It's asking whether one pitch holds many standalone videos or one pitch corresponds to exactly one standalone video. I think it might make sense to follow its recommendation here and go with okay we have one to many relationship here. So, I'm just going to say correct. Very nice. Next up, it's noticing that there's a terminology collision with the standalone video. So, it's saying that you have a standalone video that are either totally standalone or they can be related to pitches. So, I think it's basically asking whether we keep standalone video as any pitched or unpitched lesson or we redefine it to mean specifically unpitched, unlistened video. Now, the way we answer this question is going to influence everything afterwards. is going to influence whether standalone videos get their own section in the UI or whether we have standalone videos and videos with pitches kind of mixed in because the language that we're defining here is also the language that the user is going to be seeing on the page when they navigate around this. And I think I'm agreeing with it. Actually, I was actually going into this thinking we would probably have a separate section for pitched videos, but I think yes, I agree it makes sense to keep pitching as just metadata on the standalone video itself. This, by the way, is whisper flow. What I'm using for dictation. It works. Okay. Uh I'm not sure I recommend it, but it's fine. It's now saying, okay, we need some status semantics here. So each pitch can be idle or scheduled or shipped here. I think I want just to allow I think I want to allow just free form transitions. So I don't really care about this being automated. We can always layer on automation later, you know, like when you press uh send to YouTube or something, it goes into a new status. But for now, I'm okay just like toggling it myself. That's fine. It's now asking, can a pitch exist with zero videos? Absolutely, a pitch can exist with zero videos. The point here is that we're kind of generating um pitches, ideas for videos. You know, this is the Mr. Beast thing. We're trying to uh get the packaging ready first. And this relationship, this language also goes into things as concrete as deletion cascades. So, I think I'm going to say ondelete restrict here. This is mostly because I just uh like um restrictive deletions and mostly what I do is I tend to archive instead of delete when I actually do these. We're now getting into more implementation details here. So, I think um instead of just boring you with actually like implementing the grilling session here, I'm just going to say could you save what we have into context.md so far. If there's anything we haven't figured out, grill me about that before you make the adjustments. And let's see what it goes for here. And okay, it has made a bunch of updates into context. Specifically, it's added a bunch of pitch information here. So, we've got pitch, the actual entity itself. We've got pitch status, the status the pitch can be in. Pitched standalone video is a little bit awkward. I might want to grill it about that. And then unattached standalone video. That's also like it's basically saying standalone standalone video. Now bear in mind I'm like I might seem like pretty, you know, anal retentive about this language. This might just feel like bike shedding to you, but this is going to affect every part of the code that's generated. All variable names, all file names are going to be based on these uh context.md documents here. And so getting this right is absolutely crucial for feeling aligned with the AI. Now, of course, we don't want to just endlessly bike shed. So I'm going to call this now. I'm going to say that's good enough. Let's ship with this. We can always change and refactor to new language later. So, let's quickly talk about the benefits here. What you actually get from going through this ceremony. The first thing that you get is concise replies. The AI is able to use fewer tokens to speak to you because you have this shared language and it doesn't need to verboseely repeat everything or redescribe everything. It just says, "Okay, standalone videos are changing. We uh needed to make a change to the pitches and how the pitches display." This concision is also reflected in its own thinking traces as well because of course AI uses language to think to itself and so it's able to be much more aligned with your intention and actually use fewer tokens when it's thinking. This is something I've observed and it feels pretty nice. And finally because the planning documents because the way that you're speaking with the AI is also aligned with the way the code looks as well then you end up with easier to navigate code because it's able to just okay I need to find all the information about pitches. let me just search for it. And of course, this makes sense because these are all the same benefits described in domain driven design itself. So, the same techniques that work with humans also, it turns out, work with AI. You're probably thinking though, is Grill Me dead? Did I just kill Grill Me? Did its creator come along and stab it in the back? Absolutely not. I think Grill Me is an excellent, excellent skill, but Grill with Docs is better when you have a codebase. In my skills, I have moved GrillMe into the productivity area here. So this is for general use cases for use cases where you don't have a codebase. I had someone, this is the most amazing story, who said that they were writing a eulogy for their mom and they use GrillMe to get the AI to grill them about their mom and surface all these amazing stories. And so Grill Me has incredible use cases outside of engineering. And of course, if you are really early on in a project, actually really early on in a project, I'd still probably recommend using Grill with Docs because you just get so much more out of that shared language. And often at the start of a project is where you're trying to establish that shared language. So essentially, the rule is when you have a codebase, use Grill with Docs. When you don't have a codebase, use Grill Me. I update these skills super duper regularly and I'm often thinking new thoughts about the skills or even how best to use them without changing the skills themselves. So, I keep everyone up to date on this with my AI skills for real engineers newsletter. This is just an addition to the already good newsletter that I have that just gives you a few extra skills updates or maybe one a week when they happen. I really freaking hate email spam and so I'm not going to spam you, but this little page will help you basically keep up to date with all the skill change logs and have some nice extra additions here that you can just take a look at and learn how to use the skills better. Otherwise, thanks for watching and I'll see you in the next one. Thank you so much for following along. I really really appreciate it. And if you enjoy this skill, then do let me know in the comments how you got on with it, what you noticed, and do raise an issue on the skills repo itself if you think there's something that I could improve.