English
Italiano
We develop vertical AI agents, tailor-made
We start from a well-defined domain and from the people who work in it. We study its language, workflows and constraints and build an agent that works reliably in everyday use, rather than just in a demo.
Fabula Tabula is an Italian startup that designs and develops vertical artificial intelligence agents, with the operating model of a software factory.
What we do
Five activities linked by a single method, because an agent that is well designed but never maintained stops being useful within a few months.
Agent design
We draw the boundaries of the domain, collect the vocabulary of the people who work in it and put in writing what the agent must not do. This is the part that decides whether everything else will make sense.
Custom development
We build the agent around that vocabulary: sources, models and tools chosen for the specific context. The prototype is tested on real work, with the people who will have to use it.
Integration into existing workflows
The agent enters the tools already in use. One more application to open is abandoned within weeks; an agent that appears where the work happens is not.
Maintenance and evolution
After release we measure real use, correct and prune. Models change, sources change, the work around the agent changes: standing still already means getting worse.
Preliminary consulting
Before we build, we assess whether it is worthwhile. Some processes are not suited to an agent, and saying so straight away costs far less than finding out halfway through the work.
How we work
Every agent follows the same four-phase development process. That is what makes the work repeatable without making it generic.
Sowing
The domain is chosen and its boundaries drawn, listening to the people who work in it.
Cultivation
The agent is built around that vocabulary and tested on real work.
Harvest
It becomes a product with its own name and identity, and ships when it holds up to daily use.
Care
After release the long work begins: measuring, correcting, pruning.
The development process is described in full on the Agentic AI Farm page.
Frequently asked questions
The things we are asked most often, with the answer we would give in person.
What is the difference between a vertical AI agent and a generic chatbot?
A vertical agent is built on a well-defined domain whose language, workflows and constraints it knows, while a generic chatbot answers questions on any subject without knowing any of them in depth. The practical difference shows in the edge cases: the vertical agent also knows what it must not do, because its boundaries were decided during design.
What is an AI agent software factory?
It is a way of working in which every agent follows the same repeatable development process instead of being a one-off project: the domain is chosen, the agent is built on the vocabulary of that trade, released as a standalone product and cared for afterwards. Fabula Tabula calls these four phases Sowing, Cultivation, Harvest and Care.
How do you tell whether a process suits an AI agent?
A process is suitable when it has a recognisable vocabulary, rules that someone can state and an outcome a competent person can check. If instead the result cannot be assessed or the rules change case by case, the agent will produce plausible but useless answers: that is why the first phase of the work is about defining the domain, not writing code.
In which domains do you build a vertical agent?
In domains where there are people who actually do that work and are willing to talk about it, because the vocabulary of the trade is gathered from them. The first published agent, Fabulè, is dedicated to readers; subsequent domains are chosen using the same criterion, not by market size.
What happens after an agent is released?
After release the longest phase begins: real use is measured, wrong behaviours are corrected and whatever is not needed is removed. An agent stays in the Fabula Tabula catalogue as long as someone is taking care of it, because an agent that is not maintained gets worse even while standing still: the models, the sources and the work around it all change.
Which models and technologies do you rely on?
Models and tools are chosen for each domain during the Cultivation phase, not fixed once and for all. The choice is treated as a reversible decision: the agent is built so that it can switch models without being rewritten, because in this field the best technology changes faster than the problem to be solved.
What is the Fabula Tabula Manifesto?
It is the declared position from which the work of Fabula Tabula starts: artificial intelligence as an enhancement of human judgement, not as its replacement. In practice it means designing agents that leave the decision to whoever has to make it, and distrusting solutions that remove cognitive work instead of making it better. You can read it in full, or explore it as a visual work.
Who is Fabula Tabula?
Fabula Tabula is an Italian startup that designs and develops vertical artificial intelligence agents, with the operating model of a software factory. It operates in Italy and has published its first agent, Fabulè, dedicated to readers.