The Current

Fyxer details how it built an AI email assistant on OpenAI models

The company says its executive-assistant tool combines OpenAI models with data from 500,000+ hours of human EA work, and reports 53% of drafts accepted as written.

useful tools · for smb · September 14, 2026

In a case study published by OpenAI dated September 14, 2026, the company Fyxer described how it built an AI executive assistant that drafts email replies in each user's voice. According to the account, Fyxer combines OpenAI models with more than 500,000 hours of annotated executive-assistant workflows collected during years of running a human-powered EA service. Rather than treating email as a single text-generation task, Fyxer says it uses 30–50 specialized models, each handling a narrow step: a reply-decision model classifies whether a message needs a response, a scheduling action, or is informational; further models analyze intent and predict outcomes; and retrieval models surface relevant stored memories. Co-founder Archie Hollingsworth is quoted saying, "Breaking the problem into many smaller models works much better than asking one model to write a good email." Fyxer says it uses supervised fine-tuning and LoRA to create task-specific variants, and Direct Preference Optimization to learn from the differences between drafted and user-edited emails. Every drafting change is A/B tested and shipped only on a statistically significant improvement. Fyxer states it chose OpenAI after internal benchmarks and cites 53% of AI-generated drafts accepted as written. The article is a vendor case study; figures come from Fyxer via OpenAI.

  • Fyxer reports 53% of AI-generated drafts are accepted as written
  • System uses 30–50 specialized models and data from 500,000+ hours of EA workflows
  • Published as an OpenAI case study dated September 14, 2026

What it means for you

This is a marketing case study about one email-assistant product, dressed up as an engineering lesson. The useful takeaway isn't 'buy Fyxer' — it's that the best AI email tools work by learning your writing patterns from your edits over time, not by generating perfect replies on day one. The reported 53% draft-acceptance figure comes from the company itself, so treat it as a claim, not an independent measurement.

Try this

If email eats your day, trial one AI email assistant (Fyxer or a competitor) for a week and consciously edit its drafts rather than rewriting from scratch — that's how these tools learn your voice. Judge it on whether it saves you time, not on the vendor's stats.

Who should care

Solo professionals and small-business owners who spend hours a day on email and could hand off routine replies and triage.

Skip this if

You don't have an email overload problem, or you're not comfortable giving a third-party tool access to your inbox.