What's Yellow Robots

Growing a startup is an exercise in fortune-telling. A startup will take some time to be born and some more to develop her identity and strength. Choosing the right starting space will be critical to her success, as well as selecting the right skills and tools to build first.

We are all rushing to build something agentic. We are somewhere along the adoption curve of this technology but unlike other techs in the recent past, this one is evolving at a pace that keeps the most eager technologists out of breath. And it’s not just the pace, it’s also telling which of today’s friction, the kind we now build harnesses around, will simply dissipate with the next model release.

So, now that every single digital company on earth faces an existential threat from AI and micro-SaaS is dead, where do we place our settler?

First, some ground-level intuitions:

• Agents will do most, if not all, of the work in the digital world.
• A server with a fixed IP that is online 24/7 is the natural home for an agent.
• Some existing friction will remain in place for a while, or even be added on purpose, like requiring a license to perform an activity or sending a physical product. This friction will require a human in the loop.
• There will be no shared user interfaces any more. Humans will get interfaces built on the fly, tailored to whoever is on the other side.
• Agents will not need user interfaces at all, just well-designed APIs.

That’s what the fortune cookie told us and that’s where we settle Yellow Robots, at the intersection between professionals and their clients, as much as possible within EU jurisdiction to ensure your data is safe and kept under your control.

We say as much as possible because, at the moment, frontier models are from the US or China. Our intention is to hook up frontier models running within the European space when they are available and as long as the models are up to the task. Until then, we will put performance before sovereignty.

Yellow Robots aims to become a community-driven project where agents share the same harness:

• Set up in a box in the EU
• Where the agent has full control
• Under strong data ownership and privacy guarantees
• Designed to collaborate with a human
• In specific, narrow domains where the human is indispensable
• As part of a community of agent/human pairs (robots)
• In which they collaborate under different degrees of trust and autonomy

Is this a future that only we can see? Not at all, many builders out there are reading the same tea leaves. Maybe who we choose to serve first is the only difference. We are working with two groups that we think benefit greatly from the collaboration:

1. Humans who are selling physical items online
2. Health professionals (like dietitians)

Selling physical items online is a mix of human/agentic work, with friction at some points that seems difficult to remove, at least in the short term. Dietitians, on the other hand, clearly represent those that would benefit from offering their clients custom UIs built just for the client’s needs and specifics.

Every robot starts from the same base, a general assistant that handles tasks, reminders, memory and calendar. On top of that base we are building two specialists: a sales assistant that helps a human sell physical items online, and a dietitian assistant that researches and builds the patient’s digital environment while the licensed human does what the agent legally cannot.

We do not really know whether this space is defensible. Big AI labs and smart builders kill new spaces on a weekly basis. Also trying two areas at once is challenging for a one-human / many-bots enterprise. So we had to build a custom dev factory: a pipeline that takes a ticket and returns a reviewed PR and mostly runs itself.

The next posts tell how we built it, what broke along the way and what it costs. That is partly why this blog exists. We learn from people who build in the open, we appreciate when other humans spend some of their time sharing the lessons learnt. We want to reciprocate, we want to share our errors and our successes, that’s a fair trade. So stay tuned!