7h • 💬 General discussion
The Unreasonable Effectiveness of HTML (worth a read)
Came across Thariq Shihipar's (Claude Code lead @ Anthropic) take and it reframed how I think about AI output.
The argument: we default to Markdown for AI-generated docs out of habit, but we stopped editing those docs by hand. Once the output is just read, Markdown caps its value: headers, bullets, tables that break past 30 rows. HTML unlocks SVG diagrams, inline annotations, interactive widgets, real layout.
He backs it with 20 self-contained HTML files generated by Claude Code, code reviews, implementation plans, incident reports, design explorations.
The catch he's honest about: HTML runs ~4–8x the tokens and 2–4x the generation time. So it's a lever, not a religion HTML for finished deliverables a human will read, Markdown for drafts, agent-to-agent handoffs, and anything in a repo.
6
4 comments
The Unreasonable Effectiveness of HTML (worth a read)
skool.com/cliefnotes
What we give away free beats most paid courses. Build durable AI systems with a Marine vet and Edinburgh researcher. 40+ lessons, growing.
Leaderboard (30-day)
1
🔥
+2014
2
+1652
3
+1263
4
+593
5
+527
Powered by