I run my business by talking to it.
No dashboards. No tabs. No clicking through software to find the one button that matters. I describe what needs to happen, agents execute, and a governed record keeps every action mine. This website is one of them — briefed by voice, built while I did other things.
Who I am, in plain terms.
Antonio Russo. I build systems you operate by talking to them, and I run my own company that way — most of it from a phone, describing what needs to happen while agents do it.
Before any of this, I founded a company, scaled it faster than it was ready for, and raised more than $25 million across the ventures I have been part of. Some of that went well and some of it I learned the hard way — which is most of why I have strong opinions about what should and should not be handed to a machine.
The interesting part was never the AI. It was discovering how much of a working day is spent operating software rather than doing work: finding the tab, remembering the process, clicking the path, copying between tools that were never introduced. Take that away and what is left is deciding — which is the only part that needed you.
The catch is that handing execution to a machine only works if you can prove what it did. So everything I build carries the same three constraints: scoped access, a record of every action, and ownership that stays with you. Without those you have not automated anything, you have just lost track of it.
I am a co-founder of Blvckbox, where we build this for companies — more than 100 automations running in production so far. This site is where I work in the open.
Speak. It happens. And you can prove what happened.
Three things have to be true at once. Most tools give you one of them and call it a revolution.
Talk, don't type
Describe the outcome in plain language. No dashboard to learn, no process to remember, no hunting for the one screen that does the thing. The interface is the conversation — the work happens after you stop talking.
Agents execute, you decide
Machines are good at doing and bad at judging. So they do, and you decide. Every workflow has a point where it stops and asks — because the moment a system stops asking is the moment you stop being responsible for it.
Governed, so it stays yours
Scoped access instead of the keys to everything. A written record of every action. Your data walled off, pooled with nobody, training nothing. This is the part almost no one talks about, and it is the part that decides whether you can actually use any of it.
It works the same on a business and on a life. The calendar, the follow-ups, the things you keep meaning to get to — same system, same rules, clear walls between what is the company's and what is yours.
You don't own the model.
You own the memory.
Models get replaced. The one you are using now will be beaten within a year by something cheaper and better. None of that is your asset — you are renting it, and so is everyone else.
What is yours is the record. Every decision, every document, every correction you made along the way, kept in your own store and structured so any model can read it. Change providers and it comes with you.
That is the thing that compounds. Work this way for a year and the system knows how you actually operate — what you approve, what you refuse, how you word things, who matters. Not because a model was trained on you. Because the context is yours.
It is the same asset whether it is holding a company or a life. Most people spend years generating it and leave it inside somebody else's product.
Nobody is training a private language model in their spare time, and anyone selling you that is selling something else. The real play is duller and far more durable: own the memory, rent the intelligence, and keep the walls where they belong.
Built this way
Businesses don't need more software. They need their existing tools working together.
and running in production
Outcomes over outputs — every workflow exists to create a measurable business result.
This site shows the way of working. Blvckbox builds it.
Designing this for a real company — the access rules, the integrations, the audit trail, the handoff — is what my firm does. Different job, different place. Start there and you get the version with the governance already in it.
Go to Blvckbox →
The AI operating layer for modern businesses.
- Systems architecture
- Agent & automation build
- Governed data
Runbook
What I learn building these systems, written down while it is still fresh. Procedures, not opinions.
Do Not
Automate
Conversations with operators about where the line is.
I build automation for a living. This show is about the other half of the job — what these owners refused to hand over, and what it cost them to find out where that line sits. No vendor pitches.
Questions worth
asking first.
If yours isn't here, ask it on the call — the answer is free either way.
Ask me directly →I describe the outcome I want in plain language — draft this, file that, chase this thread, publish that post — and agents carry it out against the real tools. There is no dashboard to learn. The interface is the conversation, and the work happens after I stop talking.
No. Dictation turns speech into text and stops there. This turns intent into action: the system reads context, does the work across the tools it has access to, and comes back with a result or a question. Typing was never the bottleneck — clicking through software was.
No, and that is rather the point. Describing what you want is the least technical skill there is. What it takes is a willingness to hand off execution and stay responsible for the judgment.
Both, and separating them is usually the mistake. The same governed system that runs your operation can hold your calendar, your family logistics, and the things you keep meaning to get to. One place, one way of working, with clear walls between what is company and what is yours.
Not a model, no — training one costs millions and it would be obsolete in months. What you are building is the memory: your decisions, documents and corrections kept in your own store, structured so any model can read them. The intelligence is rented and swappable. The record is the asset, and it moves with you when the models change.
Scope and a record. Every agent gets explicit, narrow access rather than the keys to everything, and every action it takes is written down and reviewable afterwards. Automation handles execution. I keep context, judgment, and the final call.
You do, and no. It lives in your accounts under your credentials, walled off rather than pooled with anyone else's, and it trains nothing. If you cannot answer who saw what and when, you do not have a system — you have a liability.
That is what Blvckbox does. This site is where I show the way of working; Blvckbox is where it gets designed and built for a business, with the governance and the handoff that go with it.
Tell me what you want to stop doing.
I am genuinely collecting the list.
Not a sales call. The tasks people would hand over first are the most useful research I have, and I read every one. If you want this built for your company, that is Blvckbox.