How we are growing the team, and how we are not

The origin is unremarkable and it shapes everything. A group of computer engineers needed to use, deploy and manage AI agents in their daily work. The tools on the market could not do it: no way to control what an agent was allowed to do, no way to know what it cost, no way to prove what it had done. So they built what was missing, and that became the Yaju Agent Orchestration System.

Companies that start like that tend to lose the thing that made them good at a fairly predictable size. Here is what we are doing about it.

Hiring for the boring layer, deliberately

The work that makes the platform worth using is attribution, ownership, policy enforcement, evaluation and audit trails. It is not the part that gets attention.

So we hire people who find that interesting rather than tolerable, and we say so during interviews rather than after. Someone who wants to spend their time on model experimentation will be happier somewhere else, and establishing that in an interview is kinder than establishing it in month four.

Growing the surface slower than the demand

The consistent pressure on a platform company is to add capability faster than it can be governed. Every prospect asks for one more integration, one more mode, one more exception.

Our approach has been to add reach only when the governance extends with it. A connection an agent can use but nobody can audit is not a feature, it is a liability we shipped. This makes us slower than we could be, and we think it is the trade that keeps the product coherent.

Remote by default, in person on purpose

We are distributed, with a preference for working together in person where it is practical. Certain work, particularly early design and difficult disagreements, genuinely goes better face to face, and we would rather say that than pretend distribution is universally superior.

What makes the remote default function is writing. Proposals are written before they are discussed and decisions are recorded with their reasoning, so that being in a particular conversation is not a prerequisite for understanding why something is the way it is.

Named owners, for people as well as agents

We apply the same rule internally that we apply to agents: everything has a named owner. Not a committee. A person who decides after listening, and writes down why.

The failure mode of a growing distributed team is not conflict, it is drift, and drift is what happens when responsibility is diffuse enough that nobody is wrong.

Ways in

The Careers page carries open roles. The Scholars Program is a remote, part-time route with a mentor, open to people without prior research experience, with applications each September. The Open Development Community is free to join and always will be.

That last point is worth stating plainly: we do not charge for participation in the Open Agent System Development Initiative, and we are not planning to.

What we are keeping

The reason the company exists is that operating agents was harder than building them, and almost nobody was working on the harder half. If we ever find ourselves shipping capability faster than we can govern it, we will have become the thing we were built to replace.

Next

Careers, the Scholars Program and the Open Development Community pages cover each route in. Yaju AS is operated by CAP CONSULTOR LLC from Barcelona, and questions can go to contact@capconsultor.eu.