How Fyxer built an AI executive assistant people trust

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: How Fyxer built an AI executive assistant people trust · Channel: OpenAI · 03:07 · Watch on YouTube

Topics: AI agents, Task and workflow automation, Markets and economy, AI models

Summary

Fyxer is an AI assistant designed to help people manage emails, clients, colleagues, and conversations without missing important details. The video highlights Mike, a commercial real estate manager whose email workload was taking time away from his family, and describes how Fyxer aims to prioritize what needs his attention. The company uses OpenAI models to understand emails, select relevant context, and improve generated drafts. Its founders say OpenAI made the product possible, helped them learn from executive-assistant workflows, and supported the company as it grew.

Key points

Fyxer helps people who manage clients, colleagues, and conversations every day.
Mike, a commercial real estate manager, used to spend one to two hours catching up on emails after work.
Fyxer aims to sort and prioritize email so that messages needing the user’s attention appear first.
The product supports team collaboration to help prevent important matters from slipping through the cracks.
The team uses OpenAI models to digest emails, understand their content, and retrieve and re-rank relevant context.
Fyxer says close collaboration with OpenAI provided ideas and technical input on improving models, costs, and efficiency.
The founders say GPT-3 was a turning point that made the assistant feel workable.
The company says OpenAI enabled it to learn from 500,000 hours of executive-assistant workflows in a matter of months.
The founders attribute their product development and business success to learning about customers and incorporating that learning into the models.

Chapters

Start	Chapter	Summary
00:00	An assistant for busy professionals	Fyxer introduces its email assistant through Mike, whose work was keeping him from spending proper time with his family.
01:02	Building with OpenAI models	The team describes its workplace and explains how OpenAI models support email understanding, context selection, and better drafts.
02:05	The moment the product became possible	The founders recall GPT-3’s impact and say OpenAI helped them learn from executive-assistant workflows as the business grew.
02:36	Learning from customers	The team credits customer understanding and OpenAI’s support with helping turn its learning into a better product.

Quotes

This year, I think you’ve saved me from getting a divorce. — 00:00

We want Mike to get to a place that when he opens his laptop, the prioritizing has already been done for him. — 00:31

We use OpenAI models for everything from digesting the email — 01:33

It’s actually working. — 02:05

doing hard work of learning our customer really well. — 02:36

Concepts and entities

Fyxer, OpenAI, ChatGPT, GPT-3, Mike, Fluxos de trabalho de assistentes executivas

Full transcript (original language)

00:00 We’ve had real customers say to us, “This year, I think you’ve saved me from getting a divorce.” >> [music] >> Fixer is an AI assistant for people managing clients, colleagues, and conversations every day to make sure they don’t miss anything important. >> The divorce guy, he’s real and just awesome. He’s called Mike. He’s a commercial real estate manager managing hundreds of different properties, hundreds of different people. And so, all day on the road, can’t text while driving. And so, comes home in the evening and has to like catch up on

00:31 emails. And that one to two hours a day he’s unable to like have proper family time. >> We want Mike to get to a place that when he opens his laptop, the prioritizing has already been done for him. It’s all been read, it’s all been sorted, and all the stuff that actually needs him is right there at the top. Fixer multiplies across the team now, meaning everyone can collaborate on anything important to avoid anything slipping through the cracks. >> [music] >> We’ve been in this office six months, I think. We’re all in the same floor

01:02 [music] now. Like, we all have lunch together, and it’s just like a nice vibe. >> We already know that ChatGPT is like the second biggest AI tool that our customers use besides Fixer. I mean, their models are obviously just state-of-the-art. So, when they release a new model, it’s really easy to create something really good. >> This is [music] engineering, where all the real work happens. This is Joey, Eleonora, David. They work closely with OpenAI. >> Literally my first day at the company, I met with Matt, the CTO, and he was like,

01:33 “Okay, your job is going to be to figure out a way to make the drafts that we generate for people better.” Various parts of that flow integrated using OpenAI. So, we use OpenAI models for everything from digesting the email, so we can break down and understand what the email is actually about, to pulling in and re-ranking the context that we want to include. We’ve had an amazing experience working with the team at OpenAI. Guessing their thoughts on how we can improve our models, how we can improve costs, how we can make things more efficient, and actually getting some of that technical input as well.

02:05 >> [music] >> So, with the release of GPT-3, it was the moment where I just like got exactly the answer back that I expected. >> So, my brother called me, and he was like, “It’s actually working.” We’d been building an executive [music] assistant agency together, but it was OpenAI that actually made Fixer possible. It allowed us to learn from the 500,000 [music] hours of EA workflows in just a matter of months. And the result of that was in our first year we went from zero to 30 million in ARR. >> [music] >> One of the real keys for Fixer

02:36 succeeding as a business and delivering a better product than anyone else has literally just been about doing hard work of learning our customer really well. And what OpenAI was then really pivotal for us in doing was being able to take that learning and successfully incorporate it into how the models work. >> I can see the white sea mentality is absolutely coded into the OpenAI team. At any moment, I can drop a question in Slack, [music] and I’ve got an answer coming out pretty quickly. I’ve got what feels like a very kind of

03:07 entrepreneurial [music] and dynamic team. Honestly, it feels like a five-star hotel.