I'm sharing this to help others understand how I work at this specific moment in time. I expect in 3 months for these practices to evolve 25-50%.
I've fundamentally changed how I work. AI is no longer something I use occasionally. It's part of my default operating system for work, writing, thinking, and building.
As Chief Design Officer at Nubank, my job is direction: setting vision, making decisions, raising quality, giving feedback, and helping teams execute at scale. Almost all of that runs through an AI-augmented workflow now.
When I catch myself doing a meaningful task without AI, I pause and ask: "How can AI help with this?" That question has become a habit, and that habit compounds.
1. I changed how I create
My old model: sit down, open a blank doc, write.
My current model: capture thinking anywhere → structure with AI → refine with judgment.
This article is an example. I started it by speaking into my phone while commuting. I could've spent that time passively consuming content. Instead, I generated raw material, pushed the transcript into my OpenClaw workflow, and had Arno (my agent) synthesize and structure the draft. Then Arno and I iterated back and forth on clarity, depth, and voice. I finished this article by hand.
Creation doesn't need to start at a keyboard anymore. It happens anywhere, everywhere, all at once.
2. I run an aggressive learning pipeline
AI is moving too fast for occasional catch-up. I stay current through X/Twitter, podcasts, direct tool experimentation, and conversations with builders. I spend around 30 minutes to an hour every day consuming AI related content. I have excised non AI related content from my personal media consumption pipeline (other than occasion sports or a show with my family). No time for anything else. Nothing else is as fascinating.
But the important part isn't consumption. It's curation.
I bookmark aggressively. I tag what matters. I return to those items and turn them into action. Over time, bookmarks become a living repository of applied insight, not just saved links. I feed Arno (my bot) references to critical content as a means to ingest and interpret it. I also constantly incorporate what I've learned by experimenting with new tools and models (see 5 below). These bookmarks fuel my explorations.
A good feed informs you. A good repository changes your decisions.
3. Context quality is now strategic
At Nubank, we have rich documentation and communication surfaces: Confluence, Jira, Google Workspace, Slack, meeting notes, design artifacts. In an AI-first world, these aren't just operational tools. They're context infrastructure.
I treat everything we write and say as part of an AI-readable corpus: strategy docs, execution plans, meeting notes, Slack updates, decision logs.
That means context freshness and quality are leadership concerns. If your context is stale, fragmented, or ambiguous, your AI output will be too.
I record most meetings (excluding sensitive 1:1s), keep strategy docs updated, and push for clarity in written communication because these assets directly affect AI usefulness across the org.
I want to underscore the importance of capturing meeting notes. Over 60% of the content I create, and that I care about, occurs in meetings. By capturing meeting notes this content becomes part of the context, instead of disappearing.
4. I start from retrieval, not a blank page
For internal writing and analysis, I start with Glean by default. Not because it gives me final answers, but because it gives me a richer starting state than memory alone.
Even when I know a topic deeply, retrieval surfaces adjacent context and relevant prior work I wouldn't have pulled manually. Then I shape, compress, and rewrite the output to match the specific need and voice.
AI doesn't replace judgment. It improves your first move.
5. I stay hands-on in the terminal
I keep a terminal open all day and actively build with multiple models and tools — Claude Code, Codex, Gemini — to understand their strengths in practice, not just in demos.
At Nubank specifically, we're built on Flutter with a backend-driven content architecture. As we evolve the actual design → production pipeline to be AI-first, we're pivoting the design org to build prototypes in a React Native library that better simulates what production will actually feel like. These prototypes are also higher fidelity for usability testing, so they serve double duty. I've built within that system myself and maintain my own prototypes in it.
I also learn through personal projects. I'm curious, somewhat obsessively so. Here is a short list of projects I'm actively developing using a range of models to get a feel for their capabilities, especially as they evolve over time:
A non-custodial Ethereum wallet with a built-in prediction markets. I built this over four hours last weekend to explore blockchain libraries and development opportunities. I'm excited to integrate this with Arno, my bot.
My personal website. First built in conversation with Arno. Subsequently developed with a range of models.
A data viz experimentation library to push on the capabilities of javascript frameworks, webGL, and UI frameworks. Inspired by Braz de Pina's exceptional work at kodo7.com.
A desktop visual translation app built in Swift. I use this when I need to translate designs authored in portuguese or spanish.
A tool that exports Apple Health data to JSON, built in a Swift app.
A tool for designing ceramic objects and slip-casting molds. This is a personal passion project and a way to apply special knowledge I have of ceramics to AI.
Staying hands-on matters for two specific reasons. First, if I want teams to evolve how they work, I need firsthand understanding of what these systems can and can't do today. That's the only way to lead the change credibly. Second, what a time to be alive. We have such an abundance of tools at our disposal to express our ideas!
6. Design is moving up abstraction layers
I think it's important for us all to think in abstraction layers. What work do we do today, how can that be abstracted and augmented with AI and agents, and what is the clear path to get there. I spend a lot of my time understanding this as I help Nubank become an AI-first company. I constantly ask my teams: "if we were to start a company today, what would be our ideal tools stack and our development process."
The fact is that most teams still spend most of their effort producing screens and specs. But modern, AI-first teams are moving up the stack. The first shift is already beginning. Instead of designing in Figma and handing off specs, designers declare intent and constraints, generate options in a terminal or cursor environment, and check code directly into deployment. The designer is still doing the craft work, just at a different layer.
We are on this path at Nubank. The time I spend learning and experimenting helps me help the company most efficiently and effectively navigate the path.
7. Arno, my agent
Arno is my personal OpenClaw agent. He runs on a local machine at home and has little to do with work. However, through my interactions with him, I can see how work agents (that automate more than code) are inevitable.
I've been building Arno up over time across three main areas: taste, health, and thought partnership.
On the taste side, I talk to him regularly about what I find beautiful, what feels overworked, what's genuinely interesting versus just current. That accumulation becomes useful — he surfaces creative references that are actually calibrated to my sensibility, not generic inspiration results. My goal is to push the limits on how embody and preserve my sense of taste in an agent.
On the health side, I built a Swift app that exports my Apple Health data and pipes it into Arno as structured context: workouts, steps, heart rate, meditation, sauna, cold plunge. I can ask questions about patterns and trends in a way that's more useful than watching rings close on a watch.
And on the thought partnership side I have Arno harvest information about topics I care about from sources I value. This is combined with my own philosophies that I journal to Arno in a blend of knowledge that "we" use to think and write about the things I share here.
8. Leading by example
Part of my role is transforming Nubank into an AI-first company. I share these practices with my team deliberately, and one reason I'm writing this publicly is so the people I work with can see the full picture of how I operate. I want AI-first thinking at every level of the org, and the only honest way to ask for that is to model it visibly myself.
This article is an AI-native artifact. I spoke the raw ideas into my phone while commuting, sent the transcript into OpenClaw, and had Arno synthesize it into structure. Then Arno and I iterated on depth, framing, and clarity.
That loop — human intent, AI synthesis, human refinement — is increasingly how I work.