18 June 2026
Anthropic's engineers argued agents should hand back rich HTML, not flat Markdown. We agree — and it's why Solutions Engine builds branded, interactive pages instead of walls of text.
Earlier this year, an engineer at Anthropic published a piece arguing that AI agents should hand back HTML, not Markdown. The argument is simple. Markdown was a transport format for blog posts. The output of a serious agent is not a blog post — it's a decision, a deliverable, a thing you put your name on. That deserves a real medium.
We agreed before we read the piece. It's why Solutions Engine has always produced branded, interactive web pages as its primary artefact, not chat bubbles, not Markdown, not even PDFs as a first-class citizen. A page can layer hierarchy, imagery, brand type, callouts, tabs, references and links. It can hold your voice. A wall of bullets can't.
When an agent hands you Markdown, three things happen. You read it and it feels generic — because every Markdown deliverable looks the same. You copy-paste it into your real tool. You re-do the visual thinking yourself. The agent did the easy bit. You did the hard bit.
Rich pages flip that. The visual thinking is part of the artefact. The hierarchy of the argument is something the engine reasons about explicitly — what's the headline, what's the proof, what's the call-to-action, what supports it on a second read.
It's small enough to think about whole. Big enough to carry an argument. Brandable. Linkable. Shareable. Editable. And — because it's HTML — it can be re-rendered into a deck, a microsite, a board memo, or a printable PDF without losing its shape.
Every output we make is a page. Sometimes a single one. Sometimes a small bound microsite of them. They share your brand fonts, your colours, your tone, and your library of approved imagery. They reference each other. They learn from your edits. That is the artefact. Everything else — the deck export, the share link, the embed — is a view of it.
Begin a session
