← back
HTML is All You Need (for Agents to Make Graphics) - Amol Kapoor, Nori
Takeaway
Use HTML as an editable source format for agent-generated visuals, letting browsers handle layout and rendering.
Summary
- Canvas-based design tools force agents to manipulate spatial interfaces, whereas HTML gives them a familiar structural language for layout.
- Browser rendering translates semantic structure and CSS into styled graphics, charts, typography, and motion without manual coordinate placement.
- Nori uses HTML for sales and board decks, documents, and video, with export to formats such as PDF when needed.
- Connecting agents to company documents, emails, and transcripts lets them gather content as well as produce the visual artifact.
htmlvisual-generationcoding-agents
Original description
Coding agents are great at writing code. But non-believers will still say things like "these agents aren't really powerful because they have terrible geospatial understanding." ARC-AGI is literally grounded on this premise! And it's true that if you ask Claude or ChatGPT to draw a pelican riding a bike, you get some goofy results. But if you ask me, the problem is the tooling. We build Figma MCPs and Photoshop CLIs and all sorts of things to just get the agent to make a single powerpoint deck. I'm here to tell you that all of that is just user error. Just use HTML. HTML is all you need. Speakers: Amol Kapoor (Nori): Amol is building Nori, the cheapest and most customizable AI employee on the market for any and all dev, ops, and sales automations. LinkedIn: / amolkapoor GitHub: http://github.com/theahura/