--- title: How Fyxer built an AI executive assistant people trust description: Fyxer, an AI executive assistant startup, built its email-drafting product on OpenAI models combined with over 500,000 hours of annotated human... canonical: https://daily.dev/posts/how-fyxer-built-an-ai-executive-assistant-people-trust-e6hqdkvlz twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: How Fyxer built an AI executive assistant people trust | daily.dev og:description: Fyxer, an AI executive assistant startup, built its email-drafting product on OpenAI models combined with over 500,000 hours of annotated human... og:url: https://daily.dev/posts/how-fyxer-built-an-ai-executive-assistant-people-trust-e6hqdkvlz og:image: https://api.daily.dev/og/posts/e6HQDkvLz.png og:image:alt: How Fyxer built an AI executive assistant people trust og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # How Fyxer built an AI executive assistant people trust **[OpenAI](https://daily.dev/sources/openai)** · 6 min read · 1 upvotes · 0 comments ## Summary Fyxer, an AI executive assistant startup, built its email-drafting product on OpenAI models combined with over 500,000 hours of annotated human executive-assistant workflow data. The system splits tasks among 30-50 specialized models handling classification, intent prediction, memory retrieval, and draft generation rather than one large text-generation model. Fyxer uses supervised fine-tuning, LoRA, and Direct Preference Optimization derived from user edits to continuously improve drafts, validating changes through A/B tests. The company reports 53% of AI drafts are accepted as-written, 90% user retention at 90 days, and revenue growth from $1M to $32M ARR in 2025. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does Fyxer's AI email assistant use multiple models instead of one large model to draft replies? Fyxer splits the email workflow across 30-50 specialized models rather than using one model for text generation. A reply-decision model first classifies whether a message needs a response, scheduling action, or is just informational; other models then analyze intent, predict likely outcomes, and retrieve relevant memories before a draft is generated in the user's voice. _Developers designing multi-model AI pipelines can follow architecture breakdowns like this on daily.dev._ ### What is Direct Preference Optimization and how is it used to improve AI email drafts? Direct Preference Optimization (DPO) trains a model from pairs of outputs rather than manually labeled examples. Fyxer captures the difference between an AI-generated draft and the version a user actually edited and sent, treating that edited pair as a preference signal, then validates each resulting model change through a statistically significant A/B test before shipping it. _Teams tuning production models with real user feedback can track techniques like DPO on daily.dev._ ### What percentage of AI-generated email drafts does Fyxer's assistant get accepted without edits? 53% of Fyxer's AI-generated email drafts are accepted as written, meaning the system correctly predicts intent and tone for a majority of real conversations without requiring user edits. The company also reports over 90% user retention at the 90-day mark and revenue growth from $1 million to $32 million in annual recurring revenue during 2025. _Founders benchmarking AI product retention and adoption metrics can follow similar case studies on daily.dev._ ## Similar posts on daily.dev - [Superhuman’s new auto-draft feature almost makes me like AI replies](https://daily.dev/posts/superhuman-s-new-auto-draft-feature-almost-makes-me-like-ai-replies-gbvb9szuu) · TechCrunch · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#deep-learning](https://daily.dev/tags/deep-learning), [#openai](https://daily.dev/tags/openai), [#lora](https://daily.dev/tags/lora) [View this post on daily.dev](https://daily.dev/posts/how-fyxer-built-an-ai-executive-assistant-people-trust-e6hqdkvlz) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"How Fyxer built an AI executive assistant people trust","url":"https://daily.dev/posts/how-fyxer-built-an-ai-executive-assistant-people-trust-e6hqdkvlz","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/how-fyxer-built-an-ai-executive-assistant-people-trust-e6hqdkvlz"},"datePublished":"2026-09-14T16:45:16.540Z","dateModified":"2026-09-14T16:45:39.632Z","description":"Fyxer, an AI executive assistant startup, built its email-drafting product on OpenAI models combined with over 500,000 hours of annotated human...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/393235ca359bb48976064e3882f40ddf?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/393235ca359bb48976064e3882f40ddf?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"OpenAI","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"OpenAI","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/984b3632316f4afaa3d5213d5c64fe65","url":"https://daily.dev/sources/openai"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/how-fyxer-built-an-ai-executive-assistant-people-trust-e6hqdkvlz","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,deep-learning,openai,lora","timeRequired":"PT6M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"OpenAI","item":"https://daily.dev/sources/openai"},{"@type":"ListItem","position":3,"name":"How Fyxer built an AI executive assistant people trust"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/how-fyxer-built-an-ai-executive-assistant-people-trust-e6hqdkvlz#faq","mainEntity":[{"@type":"Question","name":"How does Fyxer's AI email assistant use multiple models instead of one large model to draft replies?","acceptedAnswer":{"@type":"Answer","text":"Fyxer splits the email workflow across 30-50 specialized models rather than using one model for text generation. A reply-decision model first classifies whether a message needs a response, scheduling action, or is just informational; other models then analyze intent, predict likely outcomes, and retrieve relevant memories before a draft is generated in the user's voice. Developers designing multi-model AI pipelines can follow architecture breakdowns like this on daily.dev."}},{"@type":"Question","name":"What is Direct Preference Optimization and how is it used to improve AI email drafts?","acceptedAnswer":{"@type":"Answer","text":"Direct Preference Optimization (DPO) trains a model from pairs of outputs rather than manually labeled examples. Fyxer captures the difference between an AI-generated draft and the version a user actually edited and sent, treating that edited pair as a preference signal, then validates each resulting model change through a statistically significant A/B test before shipping it. Teams tuning production models with real user feedback can track techniques like DPO on daily.dev."}},{"@type":"Question","name":"What percentage of AI-generated email drafts does Fyxer's assistant get accepted without edits?","acceptedAnswer":{"@type":"Answer","text":"53% of Fyxer's AI-generated email drafts are accepted as written, meaning the system correctly predicts intent and tone for a majority of real conversations without requiring user edits. The company also reports over 90% user retention at the 90-day mark and revenue growth from $1 million to $32 million in annual recurring revenue during 2025. Founders benchmarking AI product retention and adoption metrics can follow similar case studies on daily.dev."}}]} ```