Why we say 'agents,' not 'automations'
The difference between an automation and an agent isn't marketing — it's whether the thing finishes the work without you.
There are a hundred companies in Spain right now selling "AI automation." Some of them sell good products. Most of them are selling Zapier flows with a chatbot bolted on the front. We don't think that's what AI is going to be remembered for.
Automations are brittle by nature
An automation is a pipeline. Step A fires, step B runs, step C sends an email. If the input shape changes — a new field appears in the form, the email arrives in Dutch instead of English, the customer mentions two issues instead of one — the pipeline breaks or, worse, silently does the wrong thing. Automations require the world to behave.
Agents don't.
Agents take goals, not instructions
When we ship Lucien to a holiday rental agency, we don't give it a flowchart. We give it a goal:
Answer guest messages well. Cite the property data. Stay in the guest's language. Escalate when you should.
Lucien decides each turn what to do — look up the property's wifi password, draft a reply in German, ask the human for context, hand off to the agency owner. It uses the same tools a human would: read messages, query a database, write a draft, send a reply, log the conversation.
The job description is the system prompt. The toolbox is integrations. The judgement is the model.
Why this matters for businesses in Spain
Spain's growth markets — short lets, real estate, hospitality, healthcare — all run on conversation. Most of those conversations are repetitive but not scriptable. You can't write a flowchart that covers every guest question, every tenant complaint, every patient query. You'd be writing flowcharts forever.
What you need is something that can handle the long tail without breaking. That's what an agent is.
What this looks like in practice
A guest messages Lucien on a Saturday at 11pm:
"hola, the AC isnt working in the bedroom and we have small kids, what should we do? we are at villa palmera"
A flowchart would need to: detect language → match property → check AC FAQ → respond. If the guest mentions kids or the property name is misspelled, half of those steps fail.
An agent reads the message, identifies the property, pulls the AC troubleshooting steps from the property profile, recognises the urgency from "small kids," drafts a calm bilingual reply with steps to try, and pings the on-call human if the steps don't work.
That's not a Zap. That's an employee.
The bet we're making
Coluum exists because we think the right unit of "AI for business" isn't a tool — it's a named, scoped, trained agent that you hire the way you'd hire a person. With a job. With boundaries. With someone who looks after it.
That's why every agent we build has a name. Lucien is first. There will be more.