AI in the band? Yes, but it doesn't get the keys

I use AI in the systems around Virya, quite a lot actually. What I don't do is give an agent access to Gmail, Sheets, Drive and every social account and tell it to "grow the band". That sounds impressive right up to the first confidently invented email address.

What the model does#

Its useful role turned out to be much narrower. Models are good at language and ambiguity: classifying a reply, deciding whether a contact is a festival or a radio show, summarising a long thread, translating a message or preparing a draft around facts I've already verified. The deterministic workflow still owns identity, status, retries and every external action.

Constrained output#

For example, a reply classifier may return positive, negative, neutral or unclear, add a one-sentence summary and suggest the next step. It can't invent a fifth state because the email "felt promising". The workflow validates the output before it touches the CRM.

Drafting works the same way. I pass the real contact type, the reason we're writing, the material being offered, the goal and the links that must appear. AI turns that into decent English or Polish. It can improve the sentence, but it can't change the facts.

What stays with people#

Sending stays with a person. So do deletion, merging verified records, rights decisions and anything that could damage a relationship. A model can propose a collaboration email, but it can't quietly grant paid advertising rights or decide that today is the day to message five hundred people.

Facts and suggestions in separate fields#

I keep facts and suggestions in separate fields. A verified address is data. An address produced by a model is a candidate waiting for verification, however high its confidence score.

When the model fails#

When the model times out, returns broken JSON or contradicts the structured input, the workflow keeps the previous valid state and puts the record into review. The whole process still has to work on a day when the AI provider is down.

Learning from corrections#

Human corrections are useful data too. I can compare what the model suggested with what we finally decided, and see which signals led to good conversations. That's a better way to improve the system than handing the model more permissions.

So yes, AI is part of the band machinery. It handles the fuzzy bits and saves us a lot of repetitive work. People still handle taste, relationships and the decisions that matter. That seems healthier to me than pretending the robot is our new manager. :)