Goodbye Markdown, Hello HTML: Why AI Engineers Are Switching Formats
Anthropic engineer Thariq argues that HTML outperforms Markdown for AI‑augmented workflows, citing higher information density, better readability, near‑zero sharing cost, interactive capabilities, and increased enjoyment, while also weighing token cost, generation time, and version‑control challenges.
Why HTML Beats Markdown
Anthropic engineer Thariq published a long‑form article declaring that Markdown is dead and HTML is the future. He frames the piece as a single purpose: to give Markdown a "death sentence" and explain why HTML should replace it in AI‑driven documentation and workflow scenarios.
Thariq’s Five Reasons
1. Information density crushes Markdown – Markdown can only express headings, bold, lists, and code blocks, whereas HTML can render tables, CSS styles, SVG graphics, JavaScript interactions, Canvas, absolute positioning, and virtually any visual element. The author notes that there is almost no type of information AI can understand that HTML cannot express.
2. Readability – Thariq claims that for files longer than about 100 lines, he stops reading Markdown entirely. In contrast, HTML remains readable, even for large documents.
3. Sharing cost is almost zero – Markdown files must be attached and opened with a compatible viewer, while HTML files can be uploaded to S3 and opened directly in a browser, making sharing with teammates, managers, or friends trivial.
4. Two‑way interactivity – HTML can host sliders, knobs, drag‑to‑adjust parameters, interactive kanban boards, and live preview editors. The article shows examples where Claude Code generates a prompt editor with a real‑time preview and a copy button.
5. Fun – Working with HTML feels more enjoyable, leading users to invest more effort and produce higher‑quality output.
Practical Use Cases
Planning and exploration – Instead of writing a plan.md, Thariq generates a set of HTML files to visualize multiple project directions side‑by‑side, then refines the chosen direction into mockups and code snippets.
Code review – HTML can render true diff views with inline comments, severity coloring, and flowcharts, which Markdown cannot provide.
Design and prototyping – HTML’s native interactivity makes it ideal for front‑end design work.
Reports and research – Claude Code can ingest Slack, code repositories, and the web to produce highly readable HTML reports, interactive explainers, or slide decks.
One‑off editors – For tasks that are hard to describe in plain text, HTML can be generated as a custom, single‑use editor (e.g., a drag‑and‑drop kanban for prioritizing Linear tickets, or a split‑pane prompt editor with live preview and token counter).
Cost Trade‑offs
A community member calculated that processing 425 Markdown files for a year costs about $6,600 in token usage, whereas the same content in HTML would cost roughly $11,000 – an extra $5,000 per year for tags that the model does not need.
HTML generation also takes 2–4× longer than Markdown generation, but Thariq believes the extra time is worth it because the resulting documents are actually read and provide far more value.
Version control is identified as HTML’s biggest drawback: diffs are noisy and reviews are harder, and no perfect solution is offered yet.
Q&A
Readers ask whether HTML consumes more tokens. The answer acknowledges the higher cost but points out that with Opus 4.7’s 1 M‑token context window the overhead is less noticeable.
Questions about version control receive a candid response: currently there is no ideal solution, and HTML is best suited for one‑off planning, reports, or code‑review documents where strict versioning is less critical.
Industry Perspective
Carpathi (卡帕西) echoed the sentiment, presenting an evolution from plain text → Markdown → HTML, and speculating that the next step may be interactive video generated by diffusion models.
The article concludes that both Markdown and HTML are human‑centric formats, but as AI becomes a co‑author, the “human‑first” design goal is shifting. HTML’s richer, semi‑structured nature aligns better with current AI‑driven workflows, even though it introduces new trade‑offs.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
Java Backend Technology
Focus on Java-related technologies: SSM, Spring ecosystem, microservices, MySQL, MyCat, clustering, distributed systems, middleware, Linux, networking, multithreading. Occasionally cover DevOps tools like Jenkins, Nexus, Docker, and ELK. Also share technical insights from time to time, committed to Java full-stack development!
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
