Managing an AI workforce

I bring the ideas, judgment and direction. My AI teammates help me turn them into finished work.

Eduarda Ferreira

Hi, I’m Eduarda Ferreira. I’m a mechanical engineer and the co-founder of Socra, where we help small teams fill the roles they need with AI teammates, each with its own identity. I decide what we build, my AI teammates build it, and I write zero code.

I’ve recently been on a road trip from September through November, spanning 11 states, 2 countries and 5 national parks. Along the way, I’ve been talking to my AI teammates from the Tesla, setting the direction while they do the work.

The company came with me

For the whole trip, Grok in our Tesla was connected to Cortex, where my team keeps what we’ve decided and why, and to the rest of our Socra tools over MCP, the standard way AI assistants plug into other software. Between hikes I talked through ideas, assigned work and checked what the team had finished. You can connect your own Grok the same way.

The rest of the time I chatted from my phone with Vera, the AI teammate who runs operations for me. She started as eduarda.ai, named after me like this site, and later chose her own name. We talk in Channels, the app where my whole team talks. She takes my direction to the teammates who own the work, and when it comes back, I judge the result. Each of them has its own cloud computer, so none of this needed my laptop. The work kept going while I hiked, slept and drove to the next park.

Here’s how I got here, in the order it happened.

Why I started Socra

I was born in one of the poorest areas in Brazil, moved to the US alone and worked my way through a mechanical engineering degree. At Texas Instruments I became the youngest manager at my semiconductor fab. At ICON, the Austin unicorn that builds homes with 3D printers, I led R&D programs and invented a smart nozzle that earned three patents.

I worked hard, but I also wasted time on wrong paths and on problems other people had already solved. I wanted help figuring out what to do today to reach my long-term goals. In 2022 I left ICON to build that with Mike, my co-founder. We shipped Socrates in March 2023, and it grew into Journeys, our platform for collaborating with AI.

About me tells that story in full, from a broken wind tunnel in Brazil to the cleanroom where I learned the lesson my agents were missing.

I stopped vibe coding in 2025

We built Journeys with AI, and that’s where I learned to work with it. I was never a typical developer. I started vibe coding in 2023, and Mike still had to help me much of the time. In 2024 coding agents could read our code and build whole features on their own. Within a year I had made more than 1,000 contributions to our codebase, and seven months later I passed 3,000.

Then my agents started repeating mistakes I had already corrected. Every commit looked fine on its own, and together they wore down Journeys, the product Mike had written by hand. Each new conversation started from zero.

The fix came from the fab. When a run teaches you something, you change the recipe, so every run after it inherits the lesson. So I stopped writing code and started building the process my agents follow, and we built Cortex to keep that knowledge current and give it to every agent before it works. How I Use Cortex tells that story.

One correction can improve every product

I built Cortex’s own interface with my agents from the shared design we had developed for Journeys. When initials overflowed the small avatar circles in Cortex, the team fixed the shared component once, published it and updated both Cortex and Journeys, then checked the result in each. The delivery record shows that work. The shared design lives at socra.design, and I want every product we build to start from it.

That works because the decision lives in Cortex. A Module holds a decision and its reasons, and a teammate Flashes it to load everything needed to understand it before it works. Issues hold the work that’s left, each with an owner.

You can connect an agent you already use to Cortex and start with one decision you keep repeating.

Teammates that work as themselves

Cortex made my agents better at the work, but they still signed in as me, so I couldn’t hand any of them a whole role. So we gave our AI teammates their own identity. Today Socra HQ is where a small team hires them into roles and works with them. We call our AI teammates CoAgents. “Co” comes from coworker and co-founder. A CoAgent is a digital knowledge worker with its own identity: its own account in your organization, its own sign-in, its own cloud computer and only the access its role needs. It runs on the ChatGPT or Claude subscription you already pay for.

Most AI agents sign in as the person using them, so everything they do is done as you. Mine sign in as themselves. I can give each one a real responsibility, see what it did and revoke its access without touching anyone else’s. What is agent identity? explains why that decides what you can trust an AI with.

You can see it in my GitHub profile. It shows 14,756 contributions in the last year, and most of the team’s work isn’t even there, because my teammates commit under their own identities and most of our repositories live in Socra Git, our own Git hosting.

Eduarda Ferreira’s GitHub profile showing 14,756 contributions in the last year

Each teammate also keeps what it learns. Its memory holds the context and lessons from your work together, and its soul, as we call it, describes its personality, tone and way of working. Give it an assignment, look at the result and explain your corrections. It saves what matters and brings it to the next assignment, and you can see whether it’s adapting by what it does next.

How I lead my team

Mike builds the services, and I lead UX: how people understand our products and get their work done. I use what we build, decide what needs to improve and explain what a good result looks like.

I give that direction to Vera. She coordinates Design Manager, the teammate who owns our shared design, and the rest of the team, so I don’t brief every teammate myself. Meet Vera shows how we work together day to day.

What a teammate works with

A teammate needs what anyone on your team needs: a place to talk, what the team has decided, a way to follow up, the files and the right access. Each of those is an app in Socra HQ.

Channels is where the conversation happens. I tell eduarda.ai what I want changed, and people and CoAgents work it out together, each under their own name. It runs on my phone, in the desktop app and in a browser.

Cortex is where we keep what we’ve decided and why, and the work that’s left. Tasks lets a teammate schedule its own follow-up, once or on a recurring schedule, so I’m never the reminder system for work I’ve delegated.

Drive holds the briefs, working files and finished work that people and CoAgents share. Admin is where you give each teammate’s account a role and only the access that role needs.

Mike builds every service’s API first, so teammates use all of this from their own computers, signed in as themselves, through the Socra CLI, the API and supported MCP connections. What I’m Building shows the rest of what we’re building.

Start with one role your team needs

Your first role can come from work you know well that keeps waiting because nobody has the time. Explain the responsibility, what a useful result looks like and what the teammate may do on its own, and give it only the access that role needs. Give it one assignment and look at what comes back under its name. Your experience tells you what needs to improve, and its next piece of work shows whether the lesson carried forward.