Starling MX

Starling MX had an AI thesis and no product surface to put it behind.

The founding team had spent eight months proving that a language model could answer questions about a company's internal material. What they did not have was a place for that answer to live. Documents sat in Google Drive, brand files in Dropbox, decisions in Notion, and every AI response arrived in a chat window disconnected from the file it was talking about. Pilot customers liked the demo and could not adopt it, because adopting it meant leaving whatever tool the work already lived in.

The second problem was editing. Users would ask for a rewrite, get good output, then copy it into Word, lose the formatting, and re-upload. Three of the first five pilot accounts named that round trip as the reason they stalled.

Starling MX is in production with paying customers and has moved from proving the idea to widening it.

The platform now runs live for a set of mid-market customers in professional services, with document workflows and brand asset management as the two features cited most in renewals. Model routing sits behind an internal abstraction, so the team switched a summarisation path from one provider to another in a single afternoon when pricing changed.

Work in 2026 has shifted from core surfaces to depth: permissioned sharing across client organisations, an approvals path for generated documents, and usage analytics that show which AI actions actually get accepted. We remain engaged on the AI layer and the editor while their in-house team has taken ownership of billing, admin and onboarding.

Have something that needs to ship?

Scope agreed in writing, architecture first, weekly increments you can review. Start with a 30-minute call or send the brief.