Skip to content
Roboto Studio / evidence

What real AI and agent workflows has Roboto demonstrated?

Roboto has published demonstrations of creating Sanity pages with MCP, conversational commerce, agents that maintain its own comparison content, and a chat interface over its own analytics. Jono has also described how the team introduces AI to content teams. The evidence explains the workflow and its context without claiming measured client savings or autonomous correctness.

Roboto editorial synthesis. Published by Roboto Studio. .

Markdown · JSON · Coverage

How is MCP used for content creation?

The Sanity MCP walkthrough adds the project and organisation context, selects the Sanity connector, and generates a services page using existing page-builder blocks. The presenter then checks the created document and rendered result. The demonstration’s image limitation is a statement about that recorded setup, not a current product capability claim.

How are ongoing agent tasks demonstrated?

The eve walkthrough describes Roboto’s own comparison data in versioned files, source-backed findings, proposed pull requests, scheduled updates and reports delivered to team tools. This is a first-party account of the demonstrated internal setup.

How does the commerce example differ?

The Shopify-and-Sanity example uses conversation to narrow product choices and explain a returns policy. It demonstrates retrieval and interface behaviour, while the MCP example creates CMS content. Those distinct tasks should be evaluated separately.

How does Roboto query its own analytics?

A September 2026 short shows an internal Roboto chat interface that reads PostHog data on request, built with the Javelin model, the AI SDK and Vercel Labs json-render. Asked how many people visited the home page that day, it answers 42 visitors; the presenter attributes the wait to fetching data from PostHog rather than to the model.

How does Roboto describe introducing AI to content teams?

On an Everything NYC ’26 panel published on Sanity’s YouTube channel on 2 October 2026, Jono recommends starting with one specific friction point in a team’s day and automating only that until trust is established. He describes a shared design.md file that agents and team members point at, and an internal QA setup that sends smaller changes to cheaper models and larger changes to more capable ones.

Scope and limitations

  • The sources do not provide an independent reliability evaluation, client ROI study or present-day feature matrix.
  • A published demonstration does not establish that an unattended system is safe for every production workflow.
  • The panel statements are Jono’s descriptions of Roboto’s practice; the archive holds no data on their results.

Source provenance

Related answers