How Fyxer built an AI executive assistant people trust

OpenAI’s startup spotlight covers Fyxer, a European AI executive assistant that pairs frontier models with more than 500,000 hours of EA workflows. The product follows work across inbox, meetings, and apps, with email as the clearest demo.

Cited metrics: 90% user retention after 90 days, and 53% of AI-generated drafts accepted as written. Architecture uses dozens of narrow models for reply decisions, intent, memory retrieval, and tone rather than one catch-all generator.

We’re watching trust built from decomposition and feedback loops — not from pretending one model is your chief of staff.

We’re aligned with Hollingsworth’s Moravec point: drafting a “simple” email is hard because context and relationships are the job. Breaking the workflow into 30–50 specialized predictions is how you stop a general model from improvising through your professional life.

Choosing OpenAI for benchmarks, fine-tuning on tone/intent, and hands-on eng support is a vendor story — and a reminder that distribution of help still matters for startups. The 53% accept-as-written rate is more honest than 90% marketing magic; it admits drafts are a collaboration. Useful assistants that learn per-user memory and stay editable earn trust. The downside if this scales is invisible commitment tracking that acts without clear consent, or memory that retains the wrong details across threads.

I’m watching how Fyxer handles memory expiry, cross-tool agency beyond email, and whether retention stays high when novelty fades.

Fyxer is positioned as a Europe & UK technology startup building on the OpenAI API. Co-founder Archie Hollingsworth emphasizes Slack access and in-person problem-solving with OpenAI’s team.