An explanation becomes more interesting when we can put it to the test. The books ask how a model generates a response, where its limits lie and what changes when we give it access to tools. The Lab creates a space to follow these questions through two projects: Velia and Baby Velia.
From dialogue to work that continues
Asking for a summary of several documents helps reveal the difference between answering and carrying out a task. We need to access the files, understand which passages matter and check that the summary is supported by its sources. If a document is missing, the next step may be a question. The outcome of an action changes how the work continues.
An agent uses a model to choose steps towards a goal, calls on available tools and takes their results into account. Velia explores this architecture in an assistant for macOS. The prototype connects conversation, documents, research and persistent notes; permissions and available components determine what it can do.
Two agents, two paths
Baby Velia begins with a different question: what might an agent with interests and initiative of her own choose to pursue? The design imagines a small presence that could revisit a creation, try another approach or set an idea aside. Her identity, goals and activity cycle would be distinct from Velia’s.
Their relationship also allows for exchange: Velia could offer a lesson, and Baby Velia could interpret it and try it in her own work. The idea is to make the move from advice to experience observable. A lesson remains a proposal until its effect has been checked.
Change needs a comparison
When a procedure changes, one success tells us about that particular attempt. To understand whether the new method helps, we need to observe other cases and check what already worked. Keeping versions, criteria and results also makes it possible to go back when a change makes the work worse.
- An experience
- A variation
- A test
- A comparison
A Lab to follow over time
Velia is a working prototype in development. Baby Velia has a design under review, mockups and animation studies; her autonomous cycle has yet to be implemented. The Lab pages introduce both paths. The blog can follow their progress, describing what was tried, what we observed and which question comes next.
You can start with the overview or visit the individual project pages. Understanding LLMs provides foundations for reading about these experiments; the Lab accompanies the move from ideas to tests.