Go back

HTML is the new Markdown: How Anthropic engineers are building with Claude Code | Thariq Shihipar

35m 58s

HTML is the new Markdown: How Anthropic engineers are building with Claude Code | Thariq Shihipar

In this episode, Claire Vell discusses a transformative insight from Tharic of the Claude Code team: HTML is the new Markdown for interacting with AI agents. Markdown became popular for agent plans, but as models run longer and plans grow to thousands of lines, users stop reading them—a critical mistake. HTML offers a richer, more visual medium that both humans and agents can engage with, featuring scrollable layouts, mockups, code excerpts, and mood boards. This improves human oversight and decision-making. Claire and Tharic emphasize that the role of product managers is evolving into "compute allocators," who must decide how to spend compute budget wisely. Simple prompts like "create an HTML file as a plan, help me visualize" can yield high-quality outputs, especially when the agent is trusted with open-ended instructions. For larger projects, HTML can be organized into tabs or sections to separate product and technical specs. Key components of a good spec include type interfaces, validation criteria, and tests. The episode highlights that better communication between humans and agents leads to better outcomes, and HTML is now the preferred tool for that dialogue.

Transcription

7039 Words, 36847 Characters

English
Markdown became a really popular way of interacting with agents, but the plans are so long I honestly have stopped reading them and this is honestly a Mistake I think that you still need to be really in the loop plans matter or PRD's matter spec matters when you say okay Cloud can run for eight hours where you're really saying is clock and spend 500 bucks All of us are becoming these compute allocators then right and so you have to decide what is worthwhile spending the compute on? People ask me all the time Claire. You said product management is dead. What's next? I'm gonna say you're a compute allocator babe That's the job now. HTML is a lot easier to read and so it's just a richer Communication medium between you and hot instead of saying here's a markdown Document it was like what's the best way to convey this information? So you can actually engage with it and pick something this is the plan. It's purely an HTML This is something that I will actually read this is not even Personal software. It's like micro software on top of micro software Welcome back to how I AI I'm Claire Vell product leader and AI obsessive here on a mission to help you build better with these new tools Recently, I was able to attend code with Claude andthropic's first developer conference and as part of that I got to spend a little time with Tharic who works on Claude code and taught me something that has blown my mind Ever since I heard it. HTML is the new markdown He's gonna show us how to use Claude code to generate rich artifacts that both you and the agents can enjoy working on Let's get to it. This episode is brought to you by Celigo Every company today wants AI to improve how work gets done The fastest way is building it directly into everyday business processes Automating employee onboarding keeping customer data accurate managing orders and inventory or Resolving finance and operations issues when AI lives inside the flow of work It can update records trigger approvals route work and kick off the next step across systems That's how teams operationalize AI and deliver measurable results Celigo makes this possible and now with Celigo aura. It's never been easier Celigo aura gives you access to the entire platform through natural language Connecting your systems and turning intent into action all of it Under your control Companies like Databricks PayPal and OliPop rely on Celigo to run critical business operations at scale Ready to operationalize AI? Visit Celigo.com/howiai That's Celai G-O.com/howiai Welcome to howiai Thanks for having me. I am so excited to be here at Code with Claude in San Francisco There's a lot of exciting things that we're now so we'll get to that in a little bit But you told me something I was not expecting to hear today, which is you heard it here first HTML is the new markdown tell us more I mean, I think markdown became a really popular way of interacting with agents especially like you know opus 4 and cloud 3.5 where you know They have a plan and the plan is like how to do this feature and maybe it's like 50 lines of code and you can edit it Right like I think back then you're still like reading all the outputs and editing the the markdown and making it right But you know as the agents have gotten longer and longer running when you have opus 4.5 and 4.7 They're running for like an hour or something and the plans are so long. I honestly have stopped reading them and This is honestly a mistake like I think that you still need to be really in the loop You really need to understand what the agents are doing But like a thousand line markdown file You know, I don't even edit them anymore. I just have clawed to edit them instead And so I think one of the things that I've been seeing Emergently in the quad code team is that like they that we're using HTML files instead and so HTML is like the models are still very good at it They have a lot of more context now so you can spend the more extra tokens and they Like it's a lot easier to read like they can have a lot more information a lot of scrollable more scrollable and When you're talking about implementation like you know sometimes you see you clawed make these like little Asky markdown things where you're like oh like here's a little you know little mock up and it's trying really hard Yeah, in HTML It doesn't need to try nearly this hard right after it can actually draw like things that you can look at and so It's just a richer communication medium between you and plot and before we go further into HTML specifically I do have to pause because I do have a vested interest in this sure Which is you are saying for the people who are not listening listen up? Yes plans matter or priorities matter spec matters even as these models get more intelligent You still feel like that's a really important part of the process. Oh 100% yeah, I think that You know everyone has different views on how it'll go But I think that this will just forever be the thing because when I you say okay Cloud can run for eight hours where you're really saying is clock and spend like 100 bucks, you know, I mean and and so if you're spending 500 bucks or like I think all of us are becoming these compute Allocators then right and so you have to decide like what is worthwhile spending the compute on yeah And I think that is happened in the spec in planning phase right you have to really understand like What do you want and sometimes you don't know sometimes you have to like Pull it out of yourself by chatting with Claude Sometimes you have like unknown unknowns you need to figure out But yeah, I think this is like the whole thing now is just like really getting in sync with cloud about what's building I love what you said because people ask me all the time Claire You said product management is dead. What's next? I'm gonna say you're a compute allocator babe like that's that's the job now Yeah, you're still doing the same thing though You're writing documents to decide whether or not something else should do do work in the shape of that work Okay, so you've convinced me html is the future and I like how you said this It's not that it is necessarily harder or easier for the agents to read. They're very smart They can read all sorts of code But in fact what you're finding is that html makes it easier for you to engage with the content Which then up levels the quality overall because you're not your eyes aren't crossing looking at a bunch of raw markdown Me like whatever it's probably good instead you're actually getting pulled into the spec or the document or the plan and Then interacting it with the way that upgrades the the quality and then you can ultimately build something better Yeah, that's right. Okay, so you're building something with the agent so the agent can manage you You know, I'm not sure if management is the right word exactly, but you know I just I care a lot about being in sync with the agent This is sort of like the features that I built in cloud code have been like that You know like how can I get to know you better? So yeah, okay great Well, we have we have cloud code up so let's walk through how that how that works Yeah, so I I did like a little bit before we started and so I like to talk with cloud just as a human You know, and like I always start with brainstorming. It's so much easier to brainstorm once you like you know Once you have a partner so I was literally like look. I'm on a clever podcast I want to do a demo, you know, and can you brainstorm my some ideas in a html file? All right, and this is literally the prompt I gave it right like there's not it's not complicated and so here you can see the eight visual demos That it made for me and it has these little mockups as well, right? So like purity to working prototype, right like it it searched you off right you thought that with a web search right Whiteboard sketch to working you. I sure thought was really cool. This is such a cute like well things, right? It's really cute and I it's what's really funny is just this morning a chat purity user message me and they're like I love the mockups in chat Here and I'm like what in the what are you what are you talking about yeah because I have something very similar to this in code review right now And haven't shipped it and I'm like did I like did I act a cloud accidentally do this and it was that like cute little low-asky, you know wireframe. So this is definitely the dream. Yes, but not even but now you're telling me I'm gonna build it so yeah So it's giving you basically instead of saying here's a markdown document of Kind of like what you should talk to Claire about some descriptions of things you could do instead it was like what's the best way to convey this information? So you can actually engage with it and pick something and it used HTML to make this visual guide of a potential agenda or a set of demos and you just get like a much richer expression. Yeah, exactly Like I think like another like for brainstorming one of my Like sort of rules of thumb is that I'm not going to read a longer output than the screen on cloud code, you know So like if I if you gave me eight ideas, I'm just not gonna see all of them and but with HTML I'm definitely I scroll through all of these, you know, and yeah, the the diagram just make it so much more evocative for me to like But understand what's happening right the slash command starter pack five-code-of-feature-flag dashboard Yeah, a purity diet and the one I ended up liking the most was the CSV to interactive dashboards. We love a dashboard Yeah, I used to say when I was in enterprise. I guess they stole them in enterprise software dashboards equals dollars I like this one. Okay, so you use cloud code you said brainstorm but brainstorming HTML Give me a couple things that I can talk about it gave you eight ideas including visuals and this lovely like why her What the visual is and then the I like the risk. It's like it could go sideways as all good demos. Yeah, I'm yeah And so you're gonna pick one and then you're gonna show us how you pull this through to a full plan on on this idea That's right. Yeah, so I think the what I like about HTML is like really cloud really understands this and so my next prompt here was really like Okay, I asked it to me some mockups in the follow up prompt. And then I was like, I asked it to interview me about number eight, right? And so this is something that like, you know, similar to specs and PRDs, right? Like finding out my unknown unknowns, what do I want it to do? Answered a bunch of questions. And now I'm like, okay, create an HTML file as a plan that helps me visualize what the implementation plan is, include excerpts, mockups, code, whatever's needed to be like maximum context, right? And so then it made me this HTML file here. Yeah, you can see now this is the plan, but it's purely an HTML, it's got like, it started scripting out the podcast itself, which you know, maybe I didn't need all of that, but like, you know, we're making a skill. And so it, you know, fleshed out the file system, it gave me like an excerpt of the skill.md. It put together like a mood board as well. Some example components, some of the logic here. Yeah, insights and templates, helpers, grips, right? And like helping me get a sense of like, what's the important things for me to know here? Yeah, this is something that like, I will actually read, you know? Yeah. And I want to go back to Claude code really quickly, if you don't mind, which is, you know, people are gonna ask, how did it know exactly what to put into the spec? And I just, I want to go back to your prompt was very simple. And it's so funny. I've done, I don't know, 75 of these, how AI episodes, and they get incredible outputs. And everybody's prompts like make the thing, hopefully it's good. Yeah. Kind of nice. And so I love that this prompt is literally just creating HTML file with a plan, help me visualize. You misspelled excerpts. That's a good idea. And you're like excerpts, mockups, code, etc. whatever is needed. Yeah. I do want to encourage people, you know, don't stress so much about what should go in the thing. And in fact, it might change initiative to initiative. It might be slightly different to engage you with the work, but like identifying what you want to get. And then letting, Claude, letting, letting the model do what it needs to do, will do a very high quality job. Yeah. I think with prompting, it's like this fine balance of like, I think you want to give enough information that you get what you want. But you don't want to over constrain Claude. Yeah. And so sometimes when I see people with a lot of like overbuilt skills kind of, like, you know, you're an expert planner or something, right? Like that is usually like outsourcing too much and constraining it. Yep. But in this case, for example, like I really did want it to make sure it gave me code excerpt. Yeah. And I wasn't sure if I did, whether it would do that, you know, but this was really important. But then I always need to give Claude an out. You know, I always needed to be like, okay, like you asked me for this, but, you know, like there's something else I want to give you. And so whatever is needed to give me maximum context is like my way of saying, like, hey, Claude, like I trust you here. I want to just like be in the loop with you. Yeah. I love what you say, which is I trust you because my new ending prompt is not make no mistakes, love make no mistakes. But that's not it. I literally like, I believe in you and trust you. Exclamation, exclamation. Exclamation. I'm like, truly, I know you are capable and I believe in you to make these decisions. And so I think leaving that open ended sort of like whatever you got to do, I trust your judgment. At least it makes me feel like I get better outcomes. Yeah, I mean, I love the recent twist on this sort of like make mistakes, Claude, you know, like fallen love, you know, like, you know, make some bad decisions. Like we need a, we need more happenstance in our life. Okay. So you built this thing. I think that I want to walk people through it. You pull up the plan because I think about this a lot and I get people ask that ask me this a lot, which is like, what's the future of the PRD? What's the future of the tech spec? Are these things separate? Are they together? And I think what's nice is they, whatever you want, right? And whatever you want from the audience. So you are a single builder in the instance of this demo. You want it all in one piece. You want the product idea. You want the, I guess you want a 12 minute walkthrough of how to demo it. You want code snippets. You want style guide. You want that all in one thing because this is a self-contained little project that's easier to have it all months. But what I can imagine in larger organizations is like, be like put the PRD in one tab and put the tech spec in the second tab because maybe separate people would be reviewing that information. And so you can really kind of like craft a ideal spec package to whatever you want. With HTML in a way that mark down is a little bit more constrained. It has to either be like one megafile or like separate files. And I think this is just nicer from that perspective. Yeah, I think that's right. You have a lot more interactivity. I haven't asked it to make tabs here or something, but you can easily do that. It's like all the same to cloud, right? Yep. I think one of the things on like the specs and PRDs is like, I think you're trying to find like the boundary of like what you need to know with cloud. And for example, if you're doing something very technical, I like to do the type interfaces. So this is like, you know, just like understanding what the types are so that I don't often care about the actual implementation of it. I just like, okay, like the types help me understand what we're building. And then I might edit those, right? And that's like the boundary I want to interface that. And yeah, like across the problem, right? Like from the arc of the podcast to mockups, where do you want to integrate? Yeah. One other thing on this types of sort of an important input is I think types are really great. And then really a validation criteria and set of tests or other ways to see if you can get what you intended to get. I know you all like just announced outcomes, which is really focused on this like kind of goal oriented. You need to achieve this thing, do whatever you need to do. And I think that is a piece that most product managers at least are not used to writing. Like the technical success criteria of a feature and then how to test your way into it. I do those two things as well as like I care about the data model and the shape, the types, like what's going in. And then this is how you would test that you did it correctly. And with those two bookends, you can like everything in the middle is kind of kind of gravy. So that's what I think about. I think that's right. We can have probably a whole podcast on testing, I think. Let's do it. Okay. Round two. We're going to get it scheduled. Yes. And then there's like a tagline there is like test verification is not testing. Yeah. And so I think there's a lot of like, there used to be like unit test and things like that. But now, yeah, verification can be like a rubric like with managed agents and outcomes. It can be like have quadrachorta video of what it did for you. You know. So there's a lot of depth there. Yeah. I keep a set of like synthetic data and I like run a CLI through this synthetic data. So I'm like, these are all the things that have broken in the past. Yeah. We get better at like resolving these broken things than we have moved forward. So I think there's just a lot of interesting verification and testing mechanisms you can do now. We are going to part two pressure him to do it in the in the comments. Please, please tell us this episode is brought to you by persona. You're learning to build with AI. But there's an important question you need to ask. Who is actually using your product? Is it a legitimate user? A bot or a fraudster? Rex, Figma, Etsy and Twilio trust persona to answer that question. With personas identity verification platform, you can create branded experiences, automate fraud prevention, and know who is human online. That makes it easy to give good users an experience that makes them feel welcome and to stop bad actors from causing damage. And for those of you building in the AI agent space, persona helps you verify the identities of people, businesses, and developers behind agents. It's how companies like Lithik and Skyfire are pushing the frontier of agentic commerce. Learn more at with persona.com. Okay, so you have built this, but here's the objection, which is, mark down is accessible. I can like go into mark down, type in it and make edits. And so I think that is one reason is it has been so popular, it's bridged this gap between machine writable and readable, human writable and readable at a very low level of sophistication. Right? As soon as you understand, okay, these like hash signs mean headers, you're good to go on on mark down. Yes. How, like I want to fix this. How do I, how do I touch this? How do I edit? Yeah, yeah. So I think that this is like a great point, right? And I think one thing I felt with mark down was that one, I, because I'd stopped reading them, I'd stopped editing them as well, you know? And so I would end up asking Claude to edit them. And so like that is like the most basic form is just to be like, Hey, Claude, I didn't like this part of the plan. Can you edit it? But let's say you want to get really in the loop, right? And like really get in depth with a Claude can also do that for you. So the next prompt I had, and I forgot if it was here or not, it was here. Okay. So I created an edible HTML artifact to help me define the decision rules. So these are the rules that it's defined here on like, okay, how do you take data and turn it into, you know, a visualization? And I think some of these are kind of arbitrary. And so I asked it like, you know, creating HTML artifact, I don't like the ones we have right now, make this a costume UI that helps me with structure, but give me flexibility, design the ideal interface for this problem. I really wasn't quite sure what it would give me. This is one of the fun parts of HTML too. It's just like, I just want to see what Claude like cooks up here. And yeah, this was what it was. gave me, right? So it's like a my own beautiful customer interface. I can sort of like, you know, edit any of these fields. I can, you know, like hide them. I can copy. I can, you know, add new fields here. And it gives me a mark down to copy back. And so once I'm like, okay, I have this. I can copy it back into the. Okay. I want to pause because people are going to totally miss what you just did. So I'm going to repeat it, which is you have this HTML plan. Yep. And there's a section in the HTML plan that is a pretty like specific table of rendering and visualization rules per data type that you could predict would be an SCSV. Yep. And you're like, I don't like it. No. And instead of going back into cloud code and being like, I don't like it, let's go back and forth and edit it in like the terminal. You said, there's probably a way for me to interact with this particular problem that's ideal for me user perspective. So basically build a throwaway UI for this very, it's like, this is not even personal software. This is like sub, it's like micro software on top of micro software, which is like, I've made this very personalized plan. And then I'm taking a module and the personalized plan and zooming into it using a very custom UI. Yes. That's going to engage me with the content to get to a higher quality. I also like that it's like kind of gamified. It's like very consumer. Yeah. Yeah. Very consumer. And you said in the prompt, give me the ideal UI for this to like help me engage with this. You built this, then you get the data right. And then you're just going to pop it back into the file. Yeah. Exactly. So fun. Is this how you're building now? Actually that. Yeah. And do you have any challenges with like, how are you passing this around from a collaboration perspective? Or is it just like, this is the way a single threaded product or engineering leader can get something done? It's you're engaging with yourself and with the model. And so you feel like you can own things full stack or do you hit friction points with collaboration? Like when somebody needs to get input on this. Yeah. Yeah. Point one input. Sure. Yeah. I mean, I think on this scale of an implementation plan, it's way better. Yeah. And I think that this is because you just like can upload it to like, you know, whatever AWS or something. And then you just share the link around. And so definitely the like likelihood of like, I don't know, cat or Boris, like reading this is like a hundred times better. Right. And so I think that really helps me like, you present this. I also just, you know, somewhat related. I use it a lot in like collaborating overall. So for example, you know, I report to cat. And so like every week I sent her a weekly status update in HTML of everything I've done, I get caught to read my Slack and just like create this message. And like she actually gets to read it. And I don't have to spend that much time on it. You know, my gosh, new, new, new internal competition is showing up. It's not just who is building the best product. It's who's building the best product that goes into building the best product. And like who is building the best product to represent themselves to the manager. But I mean, I think why you do that is not artificially for fun. It is that it is just a much more effective way to communicate across a company is with content that is engaging. And at the right level of detail and consumable. And we're all pretty good at reading websites. Yeah. Exactly. I think this is like, when I think of like abundance, you know, and like, you know, we talk about like Javon's offer software, like, oh, like you software gets cheaper. What do you do? I say like the amount of tokens I produce that go into production code. Yeah. Like extremely small. It's like one percent or something. You know, but like I'm generating so many more tokens like this. Like my dashboard and my custom interfaces, like really trying to get a sense of like, what do I want to do? And yeah, it's like I have everything I'm interacting with is so beautiful. And I think my hope is that it also like translates into what I produce in the end, right? It's like more in the loop. It's more beautiful. It's more like, you know, like what me and Claude working together. Yeah. And I like this because I've been in the in the product game for quite some some time, many decades. And people used to get so wrapped around the axle on like, what's our source of truth for specs and what's our source of truth for PRDs? And you know, is all this information in some centralized place that we can all access it. And is it all in the same format? Is it all on the same template? And there were these arbitrary rules because creating these content was relatively expensive, consuming it was certainly expensive, finding it was really hard. And I think when all of that cost goes to like, actually zero, well, that we're all paying our, I call them our, our closets, we're paying our closets. But you can kind of put stuff wherever in whatever format because we know these models are very good at using tools to discover the context that they need. And so I do think there's this fun moment where you're really up, you like up level the things that you should care about, which is like, what is the content of the plan? Is it a good idea? Do we feel like it's going to be executed well as opposed to like, I can't put a interactive mark down document in our, you know, blessed document repository. And so I have to have like another asset somewhere else. And so I just, I like this idea of just in time documentation, very, very high quality, some throwaway software, which is nice. Like it's cheap. So you can toss it. And then I like, you know, the executive means like, what if we can't ever find this again? I'm like, Oh, Claude can find it for us. It's fine. Definitely. And then do you feel like this results in better products? Like would you, but how would you build off of this? We say plans good. Let's go. Yeah. I think so. I didn't hit implement on this. But yeah, I would basically use this as an artifact. And so I would clear context. And I would say like, here's a plan. Yeah. You know, implement it. You could also have like, you can also use this as a source of checking the truth. Right. And so again, a benefit of HTML is like, I've got a little mock up here as well. So right, I can have the verification aid or I can have the verification agent check. Like, hey, what did I intend to do? Yeah. Right. And what actually came out in the output. Right. So yeah, I think this is like, really helps Claude be more in the loop. I've got some other examples of plans here that we can let's see it. Yeah. It's like a post I'm working on. So like, it's just like different ways I'm I'm using a cloud code or sorry, using HTML was called code. And one of my favorites is this living design system. And so it's this idea that like, you know, oftentimes when I am making, let's say a new app in the entropic design system, right? Like, cloud design does this very well. Use link a GitHub repo. And you like, it will extract the design system from it. Yeah. Right. And yeah, I saw Nate did that. I'm like, so smart. What this does is like, basically, I have an HTML file that represents my design system. Yep. You can see the colors here. Typography, spacing, radius, core components. Right. It's a fairly small one. But once I have this, I can basically start passing this around. So I go to a new project. I'm like, design system dot HTML, right? Instead of like design.md or something. And it's got this like compressed understanding. And you can literally just point pod at a folder in your thing and be like, Hey, find the design system here. Create a HTML artifact and pass it around. So I love this. I do this as well. I will give you my, um, like advanced mode version of this, which is I use quad design. I pull in my, both my marketing site repo and my app repo, which have like some expressions of the same design system in, I say, make the design system. Then I actually make it ask, ask it to make a design system or a style guide. But I want it at the component level. And so we have like colors on this stuff, but we also have components because there's some tweaks in how you want the design system implemented in particular components. And then I drop that into the repo. And then yes, I say exactly this, um, reference the design system. The advanced thing that I do that I think is really useful for people who have to interface with marketers is I have a, like what I call a, what do I call it? Like it's like a component visualization page, which is like the 25 components of our app in action and interactable in a page. So a marketer can go in and like get the, get the component in the form factor. It needs to look, quote unquote, real. Yes. And then you can download a transparent PNG. Yeah. And like drop it in a deck or drop it in a video. And I know you, or we use that as a source of truth for like, remotion videos. And so I love this idea of like this living design system and this living design repository. It's great for code, but it's also great for marketers, for designers, because one of the hardest things is is getting versions of your app that look real. Yeah. And you can use HTML to do that. Yeah. Exactly. I think there's also something you're around like component variations, which I thought was, uh, fun. Where it's like, yeah, like you, we create a component you want to see like, like what if I change a padding or if I change a border, you know, solid things like that? Like, these are like a pretty simple, like, you know, a way of just playing with this. This is also claw design like right like where you create these little components or these little knobs and sliders. Um, but yeah, you, you can imagine what This is like the abstract of like, oh, like what's the interface for this thing that you're trying to do and how can you visualize it? And yeah, there's not a trade-off between like being nice pretty for you and understand being nice for Claude. You know, like they're really the same, right? So. - And you're making me think of one thing that I love in Claude design that I think you could bring into your plans. Again, it's one of these features that like sounds truly kuhu banana is to build, but it's totally possible and easy to do, which is I love in Claude design. Once you have your design going, you can like comment, you can circle things, and there's no reason you can't do that in a plan. Right? I was like, oh, how would you interact with this? And it's like, oh, just build a arbitrary like comment thing into it and say people really comments on any aspect of this and when they do, and then they submit like fix, it fixes the core thing. And so I think people getting really creative with what interaction models you can do between content. - Yes. - And code is really fun. - Yes, yeah, I mean, you could easily imagine that you did this plan as like a, like almost a lightweight, fig, my dash, where you're something where you just ask it to like, hey, make a canvas, make a bunch of things. Let me comment on it and then give me a place to copy out my comments into something that I can paste back into Claude code. - Yeah, and we did an episode recently with the Stripe team and they built their own vibe coding platform and they said what they loved about it, which I think really applies here, is they have just a particular way they want to review products and they have a particular way they want to run design review and a particular way they want to run spec review. And by building it in HTML, they can actually shape the tool to how they want the team to run, which I think is really compelling to people. I love this. So just to wrap for people, pull up Claude code, ask it to make you, give you some ideas, brainstorm ideas, but brainstorm them in HTML. Pick an idea, plan it in HTML. Pick a part of that idea you don't like. Have it create a micro app to edit in HTML. And then some like bonus things is use cloud design, make a design system, but not only that, use HTML to encode that design system in your repo, so it can be referenced at any time. Design.md is dead, longly it live design.html. Did I get it right? - Yeah, I did, that's right. - Pro is pretty good. Okay, well, this was so fun. Before we get you out of here, back to this amazing event, couple lightning round questions. One, I ask everybody, there's three tabs in Claude desktop app. - Yeah. - One is your favorite tab. - It's gotta be code. Yeah, I love the team as well. I'm really close friends with them. - Yeah, okay, so code. I thought it would be on brand. Okay, second thing, we were at this amazing event. What is the thing that you're most excited about or that you saw heard today? - I think obviously we had a big announcement at the start of the day, our partnership with SpaceX, I bring the more compute online. I think, yeah, I'm excited for, we also said we are thinking about orbital data centers. - I love it. - Yeah, incredibly sci-fi. But I could actually happen, so yeah. - I know, we're watching this moon mission with my kids who are like kind of elementary school. And I'm like, would you want to work in the moon, mom? Would you want to work in the moon minds? 'Cause I think it's coming. Yeah, orbital, that'll be next year's demo, right? - Yeah, exactly. - You and I will come back and we'll do this, we'll do HTML, we'll do testing, and then lunar data modules. - That's right, perfect. - Okay, and then my last question, very important. I love that you just talked to a code like a person. When quad is not listening, not giving you what you want. What's your prompting technique? Do you yell? No one in thatthropic yells, that has been my experience so far, so you'd be the first to admit it. - Yeah, no, I definitely don't. I think that like, there are a couple things there. I do sometimes message people and I'm like, hey, seems like you have a bug. Can you send me your transcript? And they're like, it's not the best side of me. Yeah, yeah, you know. I think that like, yeah, I don't yell like Claude. I think that like, we've also done some interesting research recently about like emotions and Claude and just sort of like this idea that like once you, when you say things with a certain emotional charge, it also like activates different features inside of the plot code. I don't think anyone's done this like AB test of like, which like, you know, if your mean to Claude, does it better than without it or not? But I'm just like, let's err on the side of like, you know, just not or like, what's the thing I prefer to exist? You know, I'd prefer if you're like nice and you know, friendly to Claude that you got better outputs. So yeah. - All I've seen is if you store a border on stern to any of these models, their reasoning gets really sad. It's like, oh, the user is right to be so disappointed with me. I'm like, oh, I don't want to read that. I don't want to see that. Yeah, thinking traces are tough. I usually give them all some privacy. I'm like, I'm not going to read the-- - Like down to your business. - Yeah, I had somebody else on it. I feel like it was Hillary who was like, just like an employee. How do you get your work done is none of my business. I don't even want to know. We just collapse those things. Well, this has been so fun. Thank you for showing us the way. Where can we find you and how could we be helpful? - Extremely online at x. Yeah, I'm at TRQ212 and yeah, just tag me if you have anything with Claude code. I'm happy to help. Perfect. I am living proof. This man is happy to help. Well, thank you for joining how I-- - Yes, thank you. Thank you for having me. (upbeat music) - Thanks so much for watching. If you enjoyed this show, please like and subscribe here on YouTube or even better, leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review which will help others find the show. You can see all our episodes and learn more about the show at howiipod.com. See you next time.

Podcast Summary

Key Points:

  1. HTML is replacing Markdown as the preferred medium for communicating plans and specs with AI agents because it is easier for humans to read and engage with.
  2. Long Markdown files are often ignored by users, leading to a loss of oversight; HTML enables richer, scrollable, and more visual communication.
  3. The role of product managers is shifting to "compute allocators," deciding how to best spend compute resources and budget on agent tasks.
  4. Simple prompts (e.g., "create an HTML file with a plan, help me visualize") can yield high-quality, interactive specs with mockups, code excerpts, and mood boards.
  5. Trusting the agent with open-ended instructions (e.g., "whatever is needed for maximum context") often produces better outcomes than over-constraining it.
  6. For larger projects, HTML can be structured into separate tabs or sections (e.g., PRD vs. tech spec) to suit different audiences.
  7. Key elements of a good spec include type interfaces, validation criteria, and tests—not just implementation details.

Summary:

In this episode, Claire Vell discusses a transformative insight from Tharic of the Claude Code team: HTML is the new Markdown for interacting with AI agents. Markdown became popular for agent plans, but as models run longer and plans grow to thousands of lines, users stop reading them—a critical mistake. HTML offers a richer, more visual medium that both humans and agents can engage with, featuring scrollable layouts, mockups, code excerpts, and mood boards.

This improves human oversight and decision-making. Claire and Tharic emphasize that the role of product managers is evolving into "compute allocators," who must decide how to spend compute budget wisely. Simple prompts like "create an HTML file as a plan, help me visualize" can yield high-quality outputs, especially when the agent is trusted with open-ended instructions.

For larger projects, HTML can be organized into tabs or sections to separate product and technical specs. Key components of a good spec include type interfaces, validation criteria, and tests. The episode highlights that better communication between humans and agents leads to better outcomes, and HTML is now the preferred tool for that dialogue.

FAQs

It means using HTML instead of markdown for agent-generated plans and specs. HTML is easier for humans to read and engage with, leading to better collaboration and quality.

As agents run longer, plans become very long (e.g., 1000 lines), making them hard to read and edit. People often skip reading them, which is a mistake.

HTML allows for richer visuals like mockups, scrollable layouts, and interactive elements, making it easier for humans to understand and engage with the content.

Because running agents costs money (e.g., $100 for 8 hours), you need to decide what compute is worthwhile. Planning helps allocate resources effectively and align goals.

It's a new role where you decide how to spend computational resources (like time and money) by prioritizing what the agent builds, based on specs and plans.

You can ask the agent to edit it for you (e.g., 'Claude, edit this part'), or use the agent to create editable HTML artifacts for deeper involvement.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.