matheus
fidelis

I build AI products for small and mid-sized businesses.

Websites, apps, WhatsApp customer service and management systems, at One.Zi, the company I founded with Carollyna. AI writes the code. I decide what's worth building and I answer for what ships.

Matheus Fidelis
01About me

Hi, I'm Matheus.

I spent 9 years behind the counter of my family's video game store. That's where I learned to listen to every customer's story. I bring that to every product.

Matheus at home, with his laptop
São Paulo, --:--

44projects started since March

All built with AI, across my own products, One.Zi's and clients'. 7 of them are on this page.

7 are under Shipped, further down

  1. beforeThe counterNine years serving customers at my family's video game store, in Suzano.
  2. the firstpsiOneA system for psychologists with video calls and real-time transcription. My first full product.
  3. todayOne.ZiWebsites, apps and WhatsApp customer service for small and mid-sized businesses.
  4. off the clockRecords and dogsA stack of vinyl and two dogs, with Carollyna.
off the clock

My desk.

The records that play while I work and the tools that stay open all day. Click things, drag the needle.

Cassiano, Cuban Soul: 18 Kilates
Arthur Verocai, Arthur Verocai
Toca-discos Technics visto de cima
wilco — zsh
matheus@mac ~ %
G-Shock GA-B2100 watch with the strap open
02How I build

How I build.

I read six months of my conversations with AI: 4,736 voice messages and 230 rules. The way I work fits in six steps.

Each square is four of the 4,736 voice messages I've sent to AI since April.Hover over it.
phase 1

Ideation

Before opening the code.

01I argue before I decideWith people and with AI, until I find the hole in the idea.
02I point to a referenceI show something that already exists instead of writing a doc.

Do you think it's worth it?

~100×times I asked AI that, in October alone
phase 2

The conversation

How I talk to AI.

03I talk and reactI say it out loud, look at the screen and say what's wrong.
04What I repeat becomes a ruleAI keeps the why and doesn't make the same mistake twice.

Love it. Just one thing: the logo is purple on purple.

17%of my September voice messages include a “take a look”
phase 3

Execution

From agreed to done.

05I split it into tracksLook, data and an owner for each part agreed upfront, then several agents work at once and one merges and reviews.
06I prove it where it's realIt's only done when it works with people using it.

Did you test it? Where's the data?

R$ 1on a real purchase, refunded, before saying payments work
the filter

Before any project starts, it goes through four questions.

  1. 1Who, by name, is going to use this by Friday?
  2. 2What could block it from outside: app store, payments, platform?
  3. 3What's the taste reference?
  4. 4Which project gets paused so this one can start?
03Clients and products

Shipped.

Three One.Zi clients, two One.Zi products and two of my own.

Each one opens the full case, with what it does and what it's built with.More at One.Zi
04What's next

Got an idea on hold?

I'm the one who answers.

  1. 1You send a messageOn WhatsApp or by email, whatever's easier.
  2. 2We talk for about 20 minutesNo strings attached.
  3. 3I tell you what I'd doWhether it's worth building and where I'd start.
Tell me the problemSão Paulo, --:--
Send on WhatsApp
or by emailmatheus@onezi.com.br
notes

How each one was built.

It goes out on LinkedIn first and stays here.

  1. Oct 26I built a Mac app with AI. My job was spotting what was wrong.
  2. coming upFrom problem to app store: Boa Feira
  3. coming up44 projects in seven months, and why half of them stopped
  4. coming upHow I talk to AI when it gets things wrong