All data in this deck is fictional · leadership session · 2026
The artifact is the demo:
HTML in the enterprise
What LLMs can produce for business users and
engineers, one self-contained HTML file at a time.
Vishal Shah · This deck is itself a single .html file · press →
The problem
AI made writing cheap.
Reading is now the bottleneck.
- An LLM drafts a 9-page status update in seconds. Nobody reads page 4.
- Long documents flatten structure: timelines, diffs, dashboards and decision tables all become paragraphs.
- The result is familiar: we skim, we miss things, we ask for "a quick summary" of the summary.
The idea
Ask for one HTML file instead of a wall of text.
A single .html file can hold the timeline as a timeline, the diff as a diff,
the dashboard as a dashboard. Interactive, printable, and openable on any machine your company owns.
This is not a product feature. It works in any capable LLM: GPT, Claude, Gemini,
Grok, DeepSeek, or the assistant your enterprise already provides.
Why HTML: argument, not vendor claim
Six properties enterprises happen to love
- One file. Self-contained, no dependencies to break or to security-review.
- Opens everywhere. Every locked-down laptop already has a browser.
- No new licences. No extra tooling, viewers, or runtimes to procure.
- Plain text under the hood. Versionable and diffable in git like any code.
- Attachable. Drops into a change ticket, evidence pack, or mail as-is.
- Works offline. Zero external requests, by construction.
Lens 1 · LLMs for business users
The whole workflow is four steps
Write a brief, or steal one from the gallery: every artifact carries its own.
Paste it into your assistant and iterate a few rounds on content and layout.
Save the answer as a .html file.
Open it in a browser. Send the file instead of the wall of text.
Expect similar output to the gallery, not identical. The examples are the bar to aim for, reached by iterating.
Lens 1 · the artifacts
Six everyday documents, upgraded
Every artifact contains its own brief. Open one and scroll to "How this was made".
Lens 2 · LLMs for engineers
Let the agent write the evidence next to the code
An LLM coding agent works in your repository. The same run that changes the code can also produce
the review pack, the ADR, the post-mortem, the release evidence: as files in the repo, reviewed
and versioned like everything else.
Engineering teams live on evidence. HTML evidence diffs like code, attaches to the change
ticket, and still opens in two years with no tooling.
Lens 2 · the artifacts
Engineering documents built to be understood, not just read
Footnote · inside enterprise tools (every row has a primary source)
"Can our tools do this?" Documented.
| Fact | Source |
|---|
| Microsoft 365 Copilot Pages: create, edit and preview runnable code ("lightweight apps") |
Microsoft Support |
| Code previews: Cloud Policy, default enabled, admin can disable |
Microsoft Learn |
| GitHub Copilot agent mode edits files under enterprise policies |
GitHub Docs · policies |
| EU data residency: "code, prompts, and Copilot responses never leave your region during inference processing" |
GitHub Docs |
| Business/Enterprise data not used to train models; IP indemnification (duplicate-detection filter set to "Block") |
GitHub Copilot Trust Center |
The honesty slide
What I am not claiming
- The examples were built with an LLM coding agent over several iterations. They show the format's potential, not a guaranteed first answer. Each one says so, on itself.
- Every artifact ships its own brief. Reproduce it with your model and judge the result yourself.
- Different models, different output. Similar, not identical. The format's properties are what carry over.
- An LLM can be confidently wrong. Treat every artifact as a first draft to check — a person verifies and owns the contents before anyone relies on them.
- All data is fictional: ABC Corp does not exist.
How to start on Monday
One experiment per lens
Business users
- Take your next status update's raw notes.
- Ask your assistant for a single-file HTML one-pager (steal the brief from artifact B1) and iterate.
- Save as .html, open, send that instead.
Engineers
- On your next significant PR, have your agent draft a review pack or ADR beside the change.
- Commit the artifact to the repo. It diffs like code because it is code.
- Attach it to the change record, and review it like any other evidence.
Close
Ask for the artifact.
Gallery, briefs and sources. Open, MIT-licensed, one folder of .html files:
github.com/vishal8shah/HTML-with-Claude
Idea credit: Thariq Shihipar's "The unreasonable effectiveness of HTML". The enterprise
lens, the briefs and all examples here are new work.
Vishal Shah · built with Claude Code, an LLM coding agent, which is itself the point ·
back to the gallery