Yesterday I saw something rare and very much of this era: a group of Cubans founding a political party while AI agents redesigned the website, prepared deploys, and pulled tasks out of a WhatsApp group.
Not a demo. Not a campaign with consultants selling vapor. This was the normal coordination of an organization being born: people in several countries, ideological debates, jokes, legal questions, external attacks, and a domain that needs to go live.
The novelty isn't that they use WhatsApp. Everyone does.
The novelty is the chat no longer ends in the chat.
A party as a living product
When you found a party, you talk about names, bylaws, manifestos, symbols, spokespeople, and alliances. That doesn't change.
What changes is how you execute.
You don't send the website off and wait two weeks to see it. You iterate. You debate the official domain. You separate staging from the public version. You adjust the message. You deploy with agents. You fix it. You deploy again.
That technical detail changes the culture. If every idea costs too much to become visible, the organization learns to talk more than it does. If an idea can quickly turn into a page, a draft, or a task list, the group starts thinking differently.
A newborn party can work more like a living product than a frozen PDF.
WhatsApp is useful, but it swallows everything
The coordination group lives in WhatsApp. Not Jira. Not Notion. Not a tool designed for work.
And that makes sense. WhatsApp is where people already are.
But WhatsApp mixes everything: serious proposals, memes, links, fights, clarifications, voice notes, third-party attacks, legal ideas, website discussions, and uncomfortable questions. A good idea can get buried in half an hour. An agreement can vanish. A "we need to do this" can die without an owner.
That's where an agent stops being a toy.
It can read what happened during the day and return something usable:
- decisions made
- pending tasks
- open questions
- topics that need a document
- points Omar or the team need to respond to
It doesn't replace politics. It just keeps it from getting lost in the scroll.
Less epic, more operations
The Cuban opposition has epic to spare.
There are livestreamed denunciations, symbols, breakups, accusations, reconciliations, big names, and righteous outrage. All of it is part of the landscape.
What's often missing is the boring stuff: follow-up, archives, owners, dates, versions, an official domain, a decision log, and a clear list of next steps.
AI doesn't fix that on its own. But it can make it lighter.
It can flag that someone proposed validating a trademark. It can remember that someone mentioned legal counsel. It can turn a debate about "right-wing" and "far-right" into a concrete need: write a public explanation. It can point out that the official website shouldn't be confused with staging. It can rescue an idea about mixed healthcare and put it where it belongs: in the platform.
That's not magic. It's administrative work. But administrative work is what separates a conversation from an organization.
Ideology also needs tools
One group discussion circled around words: liberal, orthodox, right, center-right, far-right.
That debate isn't solved by AI. It's political.
But it can be organized better.
An agent can pull out the tensions:
- "liberal" communicates individual freedom, markets, civil rights, and rule of law
- "orthodox" can sound rigid, but in a Cuban context it also connects to public ethics, republicanism, and Martí's tradition
- "right" needs precision, because "far-right" can become an empty label
- mixed healthcare and education demand an explanation that nobody's proposing to throw people off a cliff
The agent doesn't decide the line. But it turns a scattered discussion into material for deciding.
That's already a lot.
The risk is producing vapor faster
There's a problem too: without judgment, AI just accelerates the chaos.
It can write correct texts that say nothing. It can paper over indecision with nice words. It can create the feeling of progress because there are lots of new documents, even if nothing's been decided.
An agent can summarize action points. Someone has to do them.
It can prepare a website. Someone has to approve what gets published.
It can draft an explanation about the party's name. Someone has to own it politically.
It can suggest a legal path. A lawyer has to validate it.
Responsibility doesn't get automated.
What gets automated is some of the friction.
An edge for small groups
The opportunity is here: small groups with operational capacity that used to require a large organization.
A new party can have a website that updates fast, daily summaries, a decision archive, platform drafts, press documents, versions in other languages, and task tracking.
That used to take a professional team. Today it can start with two or three organized people and agents working around them.
AI doesn't found parties. It doesn't create legitimacy. It doesn't replace leadership or trust.
But it can give operational muscle to people who previously only had will.
In the Cuban case, that matters.
A different way to be active
Being active in politics doesn't have to mean only going to meetings, writing communiqués, or arguing in endless groups.
It can also mean designing systems so work doesn't get lost. Creating flows. Automating summaries. Guarding memory. Turning conversation into tasks. Turning tasks into delivery.
The Cuban dictatorship benefits from repression, sure. But it also benefits from the exhaustion, the fragmentation, and the poor organization of those who oppose it.
If agents reduce that fragmentation a little, they're already useful.
They don't do the politics.
But they help keep politics from dying inside a chat.