MarsBased

Martian Day: the crew back on Earth.🚀

We're 100% remote, so every few months, we land in Barcelona to meet face-to-face.

Our founders, Xavier Redó and Jordi Vendrell, opened the day with a look at the state of MarsBased: where we are, what we've achieved, and where the spaceship is heading next.

Remote works. But some things are better in person.

We just ran our first hands-on test of Claude Design, led by our own Anna Vidal: importing one of our client design systems straight from Figma to see what it could actually build with it.

A few things stood out. The importer flagged deprecated and half-finished components on its own, which was a nice surprise. Uploading and processing everything took closer to 30 minutes than the promised 5. And even with a full design system in place, Claude still needed real product and business context before the output made functional sense, the DS alone wasn't enough.

We also tried the "export to Claude Code" option, expecting a real bridge between design and pull requests. Right now it's closer to a fancy shortcut for the same HTML export button. Given it's still in Beta, we wouldn't be surprised if that changes soon.

Read the full story: https://marsbased.com/blog/2026/09/16/we-imported-a-real-client-design-system-into-claude-design-here-s-what-we-learned

We just ran our first hands-on test of Claude Design, led by our own Anna Vidal: importing one of our client design systems straight from Figma to see what it could actually build with it.

A few things stood out. The importer flagged deprecated and half-finished components on its own, which was a nice surprise. Uploading and processing everything took closer to 30 minutes than the promised 5. And even with a full design system in place, Claude still needed real product and business context before the output made functional sense, the DS alone wasn't enough.

We also tried the "export to Claude Code" option, expecting a real bridge between design and pull requests. Right now it's closer to a fancy shortcut for the same HTML export button. Given it's still in Beta, we wouldn't be surprised if that changes soon.

Read the full story:

"Rails no escala" es una de las frases que más se repiten en desarrollo de software, y casi nunca es cierta. El problema casi siempre está en una base de datos mal planteada, consultas sin índices o querer que la tecnología resuelva lo que debería resolver la arquitectura.

Para este nuevo episodio contamos con Cristian Planas junto a Àlex Rodríguez Bacardit. Cristian pasó 11 años en Zendesk optimizando sistemas a escala planetaria, y hartarse de leer ese mito en redes le llevó a escribir Rails Scales.

Soluciones que sí funcionan, como el sharding o el archivado de datos. Por qué la IA acelera escribir código pero jamás te salva de un mal diseño de arquitectura. Y el día que Cristian borró por error la base de datos de PlayfulBet en producción.

Todo esto y mucho más en nuestro nuevo episodio:

La importancia de la escalabilidad en Ruby on Rails | Cristian Planas ¿Rails realmente no escala o el problema es cómo gestionas tu base ...

A few months ago, we shared how we set up Claude to run our performance reviews. Today, we’re looking at what actually happened when we put it to the test.

As our COO, Jordi Vendrell, details in his latest article, using AI saves dozens of hours digging through Linear tickets, meeting notes, and Slack threads. But real-world testing exposed one core limitation: AI is incredible at gathering evidence, and completely blind to judgment.

It will instantly find the two times a Tech Lead flagged a risk over six months. What it can't tell you is whether two times is a win or a red flag for their level of responsibility. To the model, a simple typo and a major delivery failure get presented with the exact same neutral tone.

The takeaway? AI is staying for the heavy lifting of evidence retrieval, but human context has to drive the final call.

Read Jordi's full blog post on what worked, where the model struggled, and how we adjusted our review process: https://marsbased.com/blog/2026/09/09/the-judgment-gap-in-ai

When we started MarsBased in 2014, we made a decision that felt almost reckless for a new agency: zero outbound sales. No cold calling, no mass emails, no hunting down decision-makers at conferences.

Fast forward to today, and our 30-person team has worked with HP, Ford, FC Barcelona, Moody's, and Citadel Securities.

People often ask how a boutique agency lands enterprise accounts without a sales team. The truth is messier than "good networking." In the first few years, corporations rarely take a chance on you, you're too small, too new, too risky. So we relied on a few counterintuitive moves: we turned down huge clients early on when the fit wasn't right, which built us a reputation as an agency with real standards. We rode the startup acquisition wave, keeping the technical context on products that later got bought by giants like Ford or Citadel. And we learned to spot the "fast-track" procurement lane that opens up whenever a corporate leader has an urgent fire to put out.

Our CEO, Àlex Rodríguez Bacardit, goes deeper into all of this in our latest episode of Building MarsBased: https://youtu.be/uLCD0i73-Pk

Curious to hear from other founders: what's been your biggest headache with enterprise procurement?

How we landed HP, Ford & FC Barcelona as clients | Building MarsBased How does a small 30-person development agency land work with global...