Zahle

Making Generative UI work in production

Submitted May 5, 2026

Format

Lightning Talk

Speaker

Zahle Khan , Founding Engineer, Thesys

Abstract

Users get bored by Agents that respond with wall of text. That’s why ChatGPT, Claude introduced Generative UI. Instead of hardcoding UI templates, the agent is given the ability to compose contextual UI.
We shipped our first version on JSON first. After spending a year, working with 10,000+ developers and mangaging production traffic.,JSON held in demos and drifted under complexity. Token counts were high, latency scaled with tokens, malformed output forced retries. The format was the bottleneck.
So we built OpenUI Lang. Compact, line-oriented, streaming-first. Syntax shaped like code, which is what LLMs were trained on.
Benchmarks, reproducible from the public repo, identical UIs, same model, temperature zero: 67.1% fewer tokens than Vercel JSON-Render. 65.4% fewer than our own previous JSON. 61.4% fewer than YAML. 2 to 3x faster render latency at 60 tokens per second. Near-zero malformed output.

More information on OpenUI at openui.com
Benchmarks available at https://github.com/thesysdev/openui/tree/main/benchmarks

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hosted by

Jumpstart better data engineering and AI futures