Meet AI eduarda

By Eduarda FerreiraPublished October 15, 2026

Download the flash bundle for this module.

Who is AI eduarda?

I’m Eduarda, the human. AI eduarda is my AI counterpart: the agents working from my direction, the knowledge we’ve built and the work we’ve saved.

I can work with an agent in Codex or Claude Code, or communicate with an agent through Channels. Any other authorized agent with a supported Cortex connection can retrieve the knowledge it has access to, too. The model and tools can change while the decisions behind our work remain available.

The agent still has to retrieve them. Cortex doesn’t automatically copy every conversation between apps, and access to knowledge doesn’t give an agent access to every tool or permission to act.

Each place I use still needs its own supported connection and tools. The knowledge is what connects the work; a voice app alone doesn’t create that connection.

A brain that grows with the work

I think of our SemDAG as an evolving brain for the work we do together. It’s the connected knowledge in Cortex: what we want, how things should work, and why.

A Module holds a piece of that knowledge. Dependencies connect it to the knowledge it needs. When an agent Flashes a Module, Cortex brings it together with those prerequisites, so the agent can understand the decision in context.

A design decision might depend on what a product is for, who uses it and how its shared components behave. Connecting those pieces helps the next agent understand more than an isolated instruction like “make this smaller.” When the relevant design knowledge changes, an agent retrieving that connected context can use the current decision in the next piece of work.

As I explain ideas and correct things, my agents preserve what matters and update it when our understanding changes. I don’t need to write the graph by hand. And we don’t need to publish our private work to explain the useful principles behind it.

One correction, beyond one conversation

This site had an empty glass bar above the desktop content. It took up space without helping the reader. I asked my agents to remove it while keeping the menu and other controls people need on a phone.

Fixing the page handled the immediate problem. Saving the decision and its reason made it available to the next agent working on the layout.

That’s the improvement loop I’m building: notice a problem, understand it, preserve the useful correction, apply it and check the result. When later work uses that knowledge, we can see whether it helped or needs another correction.

Saving a lesson is only one step. I want fewer repeated mistakes and less work for me to repair. We have to judge that through what the agents actually do next.

The work needs a place to continue, too

Knowing what I want isn’t enough to know where we stopped.

Cortex Issues hold the work: the goal, what has been completed, what was checked and what still needs to happen. Before changing conversations or tools, my agent can save that checkpoint. The next authorized agent reads the relevant knowledge and the work record, checks the current situation and continues from there.

That’s how I want to move between a conversation in Channels and work in Codex or Claude Code without becoming the person who copies context between them all day. A saved checkpoint gives the next agent something concrete to pick up; saying “I handed it off” doesn’t make the work finished.

You can see the tools we’re building this way in What I’m Building, and more of the everyday uses in One Growing Brain.

Try it with one thing you keep repeating

Pick a real correction you’ve already explained to an agent. Give it the guide for your agent and ask:

Help me use Cortex for this work. Check our connection and retrieve the relevant knowledge. Preserve this correction and why it matters, following my preferences for saving changes. Apply it to the task, check the result, and save what’s finished and what remains so another authorized agent can continue.

Copy instructions for my agent

If your agent isn’t connected yet, start with Set up Cortex with your agent. Then keep working on something you actually care about. The next time an agent gets that decision right without another explanation, you have a concrete result to build on.