Fyxer’s AI Executive Assistant Gains Trust by Merging OpenAI Models with Half‑Million Hours of Workflow Data

The Challenge of Contextual Email

Professionals today juggle inboxes, meetings, messages and countless apps, making it easy for commitments to slip through the cracks.

Fyxer identified this gap and built an AI executive assistant that tracks conversations across tools to preserve context.

The assistant relies on the latest OpenAI models to understand each email, retrieve relevant history, and draft replies in the user’s own voice.

“There’s something called Moravec’s paradox,” explains co‑founder Archie Hollingsworth, noting that tasks humans find trivial are often hardest for computers.

For email, the paradox appears when two recipients receive the same message but need entirely different responses based on prior interactions and objectives.

Fyxer’s solution is to learn each user’s workflow so the AI can act like a seasoned assistant who already knows what matters.

How Fyxer Structures Its AI Workflow

Instead of treating an email as a single generation problem, Fyxer decomposes it into 30–50 specialized models, each handling a narrow sub‑task.

When a new message arrives, a “reply decision” model first classifies whether a response, a scheduling action, or simple visibility is required.

If a reply is needed, intent‑analysis models predict the likely outcome, such as moving toward a meeting, resolving a request, or continuing a longer thread.

Memory management models then decide which details should persist across conversations and which should be discarded after a single exchange.

Retrieval models compare the incoming email with stored interactions and surface the most relevant memories for the current thread.

OpenAI’s models power every step, from digesting the email’s content to re‑ranking contextual snippets and finally generating the draft.

Shantsila, a senior engineer at Fyxer, notes, “We use OpenAI models for everything from digesting the email, so we can understand what it is actually about, to pulling in and re‑ranking the context we want to include, to the actual email generation.”

Fyxer chose OpenAI because internal benchmarks showed superior performance, fine‑tuning options suited subjective tasks like tone, and the partnership offered hands‑on engineering support.

Archie Hollingsworth adds, “We chose OpenAI because they have the best models, and they’ve given us real access and a close working relationship. I can drop a question in Slack and get an answer quickly, and when we face a problem, the team comes to our office and works through it with us.

They show up.

The company has amassed over 500,000 hours of executive‑assistant workflow data, feeding real user feedback into the models to continuously improve accuracy.

Metrics show a 90% user retention rate after 90 days and a 53% acceptance rate for AI‑generated drafts, indicating strong trust in the system.

Lessons for AI Founders

First, breaking a complex task like email into many smaller, purpose‑built models yields better results than relying on a single monolithic generator.

Second, training on authentic assistant workflows—rather than synthetic data—helps the AI capture nuanced human judgment.

Third, close collaboration with model providers can accelerate development through fine‑tuning capabilities and direct technical assistance.

These principles illustrate a pathway for building highly contextual AI products that align with real‑world professional needs.

Fyxer’s approach demonstrates that combining robust language models with deep domain expertise and extensive feedback loops can produce an assistant that professionals trust.

Why This Matters: Fyxer’s AI assistant shows that tightly integrated, context‑rich models can achieve high user retention and draft acceptance, proving a viable route for enterprise AI tools.