I started programming at nine in Valladolid because I wanted a website about video games. It was called manucorporation.com, which felt like a serious name at the time. I filled it with whatever I was into, then slowly realized that building the site was more interesting than updating it.
At fifteen, my father bought his first iPhone, the 3G. I borrowed it as soon as I could, installed Xcode, and tried to make something that felt like the games I loved. That became my own take on Geometry Wars, built on cocos2d, the engine everyone in that corner of iOS development seemed to use.
The engine pulled me in as much as the game did, so I started contributing to it. The game started as Abstract War 2.0, grew into Infinity Field, reached the App Store Top 10, and ended up published by EA. I was still a teenager, sitting in Spain, watching something I made get played by hundreds of thousands of people. That breaks your idea of what a computer can do for your life in a useful way.
Infinity Field is how I got to San Francisco. At seventeen I flew out for a summer to work at JoyPad. I came home, started a computer-science degree, and felt restless almost immediately. The next year I went back for longer, joined an incubator, and built developer SDKs at OpenKit and TinySpark.
Then I came home again and wrote the first commit of Gin.
For a while I thought the pattern was games, then startups, then open source. It took me longer to see the simpler version. I like software that sits close to the machine and still has to feel good in someone’s hands.
That has taken me through compilers, language runtimes, web frameworks, a game engine, GPU rendering, and distributed infrastructure. The domain changes. The part I like stays the same. The obvious version falls over, and you have to understand the system well enough to make the simple version hold.
Lately that has meant AI agents. One of my favorite systems from the last few years was a batch pipeline that consolidates an agent’s project memory while it sleeps. Large offline passes run when the agent is idle, compressing and integrating what it learned during the day, so it wakes up with a smaller context and a better memory. Borrowing from how brains handle memory turned into a surprisingly practical systems problem.
What I’m working on now
I’m a principal engineer at Builder.io, where I set the technical direction for our AI agent platform across product, design, and code. My work is split between the developer tools, the applied-AI systems behind the agents, and the infrastructure that runs them.
That infrastructure spins up ephemeral Kubernetes clusters per workflow across AWS and GCP, connects to enterprise networks through multi-tenant VPCs, and tears everything down when the work is done. It serves billions of requests a day. The fun is getting a large, messy system to behave predictably while real customers are leaning on it.
Open source
Open source taught me to make work earn its keep in public. If an API is confusing, someone tells you. If a benchmark lies, someone finds the edge case. If the thing works, people quietly build on it for years.
- I created Gin, a Go web framework with around 88k stars and 290k+ dependent projects. I wrote more about its router in the Gin write-up.
- I helped take Qwik from proof-of-concept to production, including its Rust compiler and a resumability model that skips client-side hydration.
- I designed the compiler architecture and high-performance virtual DOM for Stencil, and worked on Ionic across Angular, React, Vue, and plain JavaScript.
- Further back, I was a core contributor to cocos2d, the game engine I used for Infinity Field.
How I think about engineering
My CV from when I was a teenager had one line that still makes me smile. “I want to make the best software ever, and build it so the people who use it love it.” I’ve grown more modest about the first half and more serious about the second.
The instinct that survived is treating performance and developer experience as features. A game should hit its frame budget. A framework should make the fast path feel natural. An SDK should explain itself in the first five minutes. Infrastructure should stay boring under load, whether the bottleneck is Cassandra, Postgres, MongoDB, Elasticsearch, or the queue between them. A browser should render a multi-gigabyte CT volume because the data path was built with respect for the GPU.
I keep coming back to the same shape of problem. Hide the right complexity, expose the right control, and leave the person using the thing with fewer decisions than you had to make while building it.
Where to find me
I’m @manucorporat on GitHub and @manucorporat on Twitter, based between Lisbon and Budapest.