A conference talk by Amol Kapoor (Nori) arguing that coding agents don't need specialized design tools like Figma MCPs or Photoshop CLIs to produce graphics; HTML alone can serve as the output format. The speaker claims agents' weak geospatial/spatial reasoning (e.g., poor drawing results) is a tooling problem, not a model limitation. Presented at an AI Engineer conference; per the title, the talk focuses on using HTML for agent-generated graphics. # Channel Identity Observations — "HTML is All You Need (for Agents to Make Graphics)" ## Explicit Self-Description - **Creator identity**: The presenter states, > "Hi, I'm Amol, CEO of Nori Atentic" — the video's speaker is a company CEO, and the channel/video is presented under the name "Amol Kapoor, Nori." - **Company scope claim**: > "We deploy an AI employee that understands your company, your code, docs, Slack, and other kinds of data." The company positions itself around an AI product that integrates with company data. - **Product promotion**: The video explicitly promotes a product: > "That's what Nori Sessions lets you do" and > "Nori ships with everything you need to make this all work, so don't bother reinventing the wheel." ## Intended Audience - **Technical/business professionals interested in AI agents**: The talk assumes familiarity with concepts like coding agents, MCPs ("Figma MCPs, PowerPoint CLIs"), SVG, HTML/CSS, and benchmarks ("Arc AGI"). - **People who build presentations and documents**: Addressed directly — > "Say you need to make a slide. What do you do?" and discussion of decks, board decks, and sales decks. - **Developers and skeptics of AI capabilities**: The speaker engages a skeptical position — > "AI skeptics say that it's not just the tools. Agents fundamentally can't reason about space" — implying an audience aware of AI limitations debates. ## Claimed Scope of Content - **Agent usability education**: > "We spend a lot of time thinking about how coding agents really work" and > "There's just one trick. You have to be able to think like an agent to get it to do what you want it to do." - **Core thesis of this video**: > "for graphics, all you need is HTML" — HTML as the native medium for agent-generated visuals (slides, docs, video). - **Demonstrated practice, not just theory**: The speaker claims the technique is in real internal use — > "We use this HTML trick to build all of our slide decks, our board decks, and our sales decks" and > "What you're watching is just HTML and CSS." ## Observable Presentation Patterns (Demonstrated, Not Claimed) - **Argumentative structure**: Problem (agents fail at canvas tools) → reframing ("it's not the model, it's the medium") → solution (HTML) → practical payoff → product tie-in. - **Concrete illustrative examples**: References Simon Willison's pelican-riding-a-bicycle SVG test and the "34,000 human years" figure for deck-making as evidence in the argument. - **Marketing-adjacent educational tone**: Content is framed as a lesson ("That's my little spiel") but culminates in promoting Nori Sessions, indicating the channel mixes thought-leadership with product marketing. - **Catchy, meme-adjacent title style**: "HTML is All You Need" echoes a well-known ML paper title, suggesting awareness of technical-audience culture. ## Claims vs. Demonstrated Distinction - **Creator claims**: Company capabilities, internal usage of HTML-generated decks/docs/videos, and Nori Sessions functionality. These are asserted, not independently verifiable from the material. - **Demonstrated**: The rhetorical style, technical fluency, product-promotion structure, and audience assumptions are directly observable in the transcript itself. # Delivery Observations — "HTML is All You Need (for Agents to Make Graphics)" ## Overall Tone - **Conversational and lightly humorous**, especially when evaluating AI output quality: > "And yeah, these are pretty bad. Like genuinely deeply really bad." - The casual stacking of intensifiers ("genuinely deeply really bad") signals an informal, direct-to-viewer style rather than a formal lecture register. ## Rhetorical Questioning A prominent, recurring device — the speaker repeatedly poses questions and immediately answers them: - > "So, does that mean it's hopeless? Agents are just doomed to be bad at graphics? No, I don't think so." - > "Say you need to make a slide. What do you do?" - > "What do all of these agent tools have in common?" This question-answer rhythm structures most transitions in the talk. ## Signature Phrase: "Think Like the Model / Agent" The central framing phrase recurs at least four times, making it the talk's anchor vocabulary: - > "You have to be able to think like an agent to get it to do what you want it to do." - > "But an AI is not a human." - > "Well, again, we can think like the model." - > "Stop thinking like a user. Think like the model." ## Analogy-Driven Explanation Teaching through comparisons is a consistent habit: - > "Asking an AI to use a canvas is like asking a human to write SVG by hand." - > "SVGs are just a wall of numbers. You can't go from a wall of numbers to a pelican." ## Self-Aware, Reflexive Callbacks - The speaker sets up a demonstration early ("Remember that pelican from earlier?") and returns to it, a classic callback teaching pattern. - The talk is self-referential about its own format: "What you're watching is just HTML and CSS. It's literally just divs all the way down." *(Note: "divs all the way down" may allude to the "turtles all the way down" phrase — the allusion itself is not stated explicitly.)* ## Direct-Address Framing Frequent second-person address to the viewer ("you"): - > "If I asked you, someone who is presumably human, to handwrite an SVG of a pelican..." - "presumably human" is a one-off witty aside, not a repeated pattern. ## Explicit Takeaway Structure The talk closes with a clearly signposted summary: - > "If you have just one takeaway, it's this. Stop thinking like a user. Think like the model." ## Concrete Numbers and Named Examples - Quantitative hooks: "34,000 human years," "a deck that takes 10 hours should really take about 25 minutes." - Cited external references: "benchmarks like Arc AGI" and "developer Simon Willison." *(Uncertainty note: these are mentioned without elaboration, so their accuracy as citations cannot be assessed from the transcript alone.)* ## Personal Anecdote - > "I've built entire board decks for my phone on my subway during my commute." - Used once near the end as a credibility/practicality beat; not a repeated pattern. ## Register Shifts - Mostly informal ("That's my little spiel"), but with a brief formal aside marker: "Now, I do want to take a quick beat here and point out..." — used once to flag a caveat. # Position Record: "HTML is All You Need (for Agents to Make Graphics)" — Amol Kapoor, Nori ## Speaker Identification - **Sole speaker:** Amol Kapoor, self-identified as **CEO of Nori Atentic** (company referred to as "Nori" throughout). ## Core Thesis - **Coding agents are not limited to writing code** — Kapooor calls the framing that they are "just bad marketing." - To get useful output from agents, "you have to be able to think like an agent." - **Agents can make visual artifacts** (slides, docs, video) if given the right medium — **HTML**, not canvas-based or SVG tools. ## Stated Reasoning & Claims ### 1. The problem with existing design tools - **Claim:** ~34,000 human years daily go into making slide decks; most is "the fiddling," not the thinking. A 10-hour deck "should really take about 25 minutes" once formatting is removed. - **Claim:** Tools like PowerPoint, Slides, Figma, and Canva are built for "human hands and human eyes" (click, drag, geospatial manipulation); the underlying data structure "only the application can read." - **Claim:** When agents use these tools, output is "garbage" — overlaps, unreadable text, no alignment. ### 2. The skeptics' position (represented, not endorsed) - AI skeptics argue it's not just the tools: **agents fundamentally can't reason about space**, a premise behind benchmarks like **Arc AGI**. - He cites **Simon Willison's** "pelican riding a bicycle" SVG test as a gut check for spatial reasoning; results shown were "genuinely deeply really bad." ### 3. Kaplan's rebuttal: "It's not the model, it's the medium" - **Analogy:** A human asked to handwrite an SVG pelican "wouldn't be able to do that either" — SVGs are "a wall of numbers"; humans think graphically. - **Reversal:** Asking an AI to use a canvas is like asking a human to write SVG by hand. - **Prescription:** Give AI tools matching its native medium — "words, tokens, structure," not pixels. ### 4. Why HTML - Models have trained on **billions of examples**, understand it intuitively, and it **renders everywhere**. - Tags carry meaning (heading, chart, grid); the browser converts to pixels, so "the model never actually places a coordinate." Charts, fonts, motion, layouts come "for free." - The pelican test in HTML: "same bird," but in a structure the model can reason about, and every line is readable, themable, editable. ### 5. Slide decks vs. PowerPoint - **Claim:** Slide decks and PowerPoint are *not* synonyms — PowerPoint is a tool; the deck is just the presentation mode. "The editing format is totally arbitrary." - Agents can use HTML as the editing format and render to PDF later if needed. ### 6. Practical claims about Nori's own use - All Nori slide decks, board decks, sales decks, and docs are built via this HTML approach — "real things that we actually present." - **The video itself** is "just HTML and CSS... literally just divs all the way down." - **Claim:** Plain text is "usually the wrong choice" when creating something useful; structure and color improve most things. ### 7. Qualification and product pitch - **Explicit caveat:** "A beautiful deck on its own is generally not worth anything" — content still must come from somewhere. - **Solution claimed:** Give the model access to company data (call transcripts, emails) to build decks end to end — this is **Nori Sessions**. He claims to have built board decks from his phone during a subway commute because "our Nori bot lives in the fabric of our company." - Marketing note: "Nori ships with everything you need... don't bother reinventing the wheel." ## Closing Takeaway (quoted) > "Stop thinking like a user. Think like the model. Give it the right language, and for graphics, all you need is HTML." ## Unresolved Tensions / Qualifications Preserved - The skeptic position (spatial reasoning is fundamentally hard for agents, Arc AGI premise) is **presented and countered, not refuted with evidence** beyond the SVG-vs-HTML analogy. - The "10 hours → 25 minutes" and "34,000 human years" figures are asserted without cited sources. - The pitch blends technical argument with promotion of Nori's commercial products; claims of internal use (board decks, subway commute) are self-reported. # Video Attribution Record ## Video - **Title:** HTML is All You Need (for Agents to Make Graphics) - Amol Kapoor, Nori - **Video ID:** `01a15ca8-e15b-7e87-8ef5-d2f4d706f647` ## Speakers & Roles - **Amol** — **Creator / sole speaker**, self-identified as **CEO of Nori Atentic** (spoken in transcript). No other speakers appear. ## Promotional Segments ⚠️ - **Nori product promotion** (creator's own company): - "That's what Nori Sessions lets you do… our Nori bot lives in the fabric of our company." - "Nori ships with everything you need to make this all work, so don't bother reinventing the wheel." - Claims company decks (slide, board, sales), docs, and this video itself are built with the HTML approach. ## Referenced Sources & Third-Party Claims - **Simon Willison** (developer) — cited for the "pelican riding a bicycle" SVG test given to new models; attributed as "a famous little test." - **Arc AGI** — cited as a benchmark "built exactly around" the premise that agents can't reason about space. - **Tools mentioned (no claims endorsed):** PowerPoint, Google Slides, Figma, Canva, Figma MCPs, PowerPoint CLIs. - **Statistic claimed by creator (unsourced in transcript):** ~34,000 human years per day spent making slide decks. ## Availability & Limitations - **Curated metadata:** none supplied. - **Comment availability:** not indicated. - **No timestamps** available in transcript; promotional claims are creator's own and unverified within source. \>\> Hi, I'm Amol, CEO of Nori Atentic. \>\> \[music\] \>\> We deploy an AI employee that understands your company, your code, docs, Slack, \[music\] and other kinds of data. We spend a lot of time thinking about how coding agents really work. Most people think coding agents only write code. But if you ask me, that's just bad marketing. \>\> \[music\] \>\> Forget the name for a second. Coding agents can do almost anything. \>\> \[music\] \>\> There's just one trick. You have to be able to think like an agent to get it to do what you want it to do. \>\> \[music\] \>\> Today, we're going to talk about how we use coding agents to do something most people think agents are terrible at, make visual artifacts, like slides, docs, \[music\] and yeah, even video. \>\> \[music\] \>\> Every day, the world pours something like 34,000 human years into making slide decks. Most of that time isn't the thinking, it's the fiddling. \[music\] A deck that takes 10 hours should really take about 25 minutes once you remove all the formatting and the branding and the moving things around. Say you need to make a slide. What do you do? You open a tool, PowerPoint, Slides, Figma, Canva, and then you start manipulating a canvas. Every one of these tools is built for human hands and human eyes. Click, drag, drop, resize, snap to grid, all motions and patterns that make sense for our geospatial view of the world. There is a data structure underneath, but it's in a format that only the application can read. What happens when you hand these tools to an agent? Well, the output comes out all wrong. Things overlap in weird ways, you can't see the text, there's no alignment, It's just garbage. AI skeptics say that it's not just the tools. Agents fundamentally can't reason about space. And there are whole benchmarks like Arc AGI that are built exactly around that premise. There's a famous little test for this from developer Simon Willison. He asks every new model the same thing. Can you draw a pelican riding a bicycle? But there's a trick. The agent is only allowed to use SVG. It's a quick gut check for whether a model can reason about space at all. Here's some examples of what the models actually give you on this test. And \[music\] yeah, these are pretty bad. Like genuinely deeply really bad. \>\> \[music\] \>\> So, does that mean it's hopeless? Agents are just doomed to be bad at graphics? No, I don't think so. If you ask me, it's not the model, it's the medium. If I asked you, someone who is presumably human, to handwrite an SVG of a pelican, you wouldn't be able to do that either. \>\> \[music\] \>\> SVGs are just a wall of numbers. You can't go from a wall of numbers to a pelican, you just can't see that way. That's just not how people think. We think graphically. So, we build tools that let us draw on a canvas. Figma MCPs, PowerPoint CLIs, screenshot and replace loops, what do all of these agent tools have in common? They all approach the problem like a human. But an AI is not a human. Asking an AI to use a canvas is like asking a human to write SVG by hand. It doesn't really make sense. You need to give the AI tools based on how it thinks, not in pixels, in language. Words, tokens, structure, that is its native medium. Imagine a language that's incredible at describing layout. That models have seen and trained on billions of examples of. That they understand intuitively. that renders to pixels and can run everywhere. Oh, right. HTML lets the model think in structure. HTML tags have meanings built into the language. A heading, a chart, a grid, and the browser turns it all into pixels. So, the model never actually places a coordinate, and you can get all sorts of visual effects, charts and layouts, fonts and motion, all of it for free. Remember that pelican from earlier? Now, ask it to do the same exact task, but in HTML. Same bird, but now it's in a structure that the model can reason about. And you can read and theme and edit every single line of it. I spent my whole life building slide decks with PowerPoint. So, I always thought that those two things, slide decks and PowerPoint, were synonyms. But that's just not really true, is it? PowerPoint is a tool that you use to make slide decks. The deck itself, that's just the presentation mode. And as it turns out, no one in your audience is going to care how you got to the presentation mode. The editing format is totally arbitrary. \>\> \[music\] \>\> So, you can just pick the editing format that the agents are already good at, HTML, and if you need to render to a different format like PDF later on. We use this HTML trick to build all of our slide decks, our board decks, and our sales decks. \>\> \[music\] \>\> These are real things that we actually present and send out constantly. We use it for our docs, too. It gives our docs color and vibrancy, all while following our brand. And of course, we also use it to make videos like this one. What you're watching is just HTML and CSS. It's literally just divs all the way down. Almost everything is better with a little structure and a little bit of color. Plain text is a choice, generally a choice of convenience, but it's usually the wrong one if you're actually trying to create something of use. Now, I do want to take a quick beat here and point out that a beautiful deck on its own is generally not worth anything. You still have to go and get all of that content, all of the things that actually populate that deck, right? Well, again, we can think like the model. If you just give the model access to your data, say your call transcripts or your emails, you can have the model build the deck end to end. Let your agents do all the grunt work while you focus on vision and story. That's what Nori Sessions lets you do. I've built entire board decks for my phone on the subway during my commute. Why? Because our Nori bot lives in the fabric of our company. Of course, Nori ships with everything you need to make this all work, so don't bother reinventing the wheel. That's my little spiel. Thanks for listening. If you have just one takeaway, it's this. Stop thinking like a user. Think like the model. Give it the right language, and for graphics, all you need is HTML. \>\> \[music\]