How do you pick the first process to hand to an agent?

When I sit down with an owner for the first time, there is almost always a list already waiting for me: the four or five things they most want off their plate. I have learned not to trust the order that list arrives in. It is sorted by pain, and pain is not the same thing as readiness.

I have been sorting lists like this for a long time, well before agents existed. Eight years inside Tupy, Latin America's largest foundry, taught me to read a real operation before I ever gave anyone advice about one: I sat with logistics, with sales, with invoicing and export, and I built the systems the sale actually ran through, from the order to the tax book. Later, when I co-founded Sizebay, I wrote the first validation algorithm myself and made the first sale with my own hands before I ever taught a salesperson to make the next one. SEBRAE, Brazil's national agency for small business, does not hand out its accreditation for showing up. It validates a consultant's actual track record first, and it is demanding. I passed that review, and since 2019 I have used the same method on hundreds of small and medium companies: interview every department, find out what the data already answers, and build only what is actually missing.

An agent is just the newest thing I build with that method. The interview does not change. What changes is that the answer used to be a spreadsheet, or a CRM finally connected the way it should have been from day one. Now it can be an agent.

The instinct I correct most often

In engagement after engagement, the instinct is the same: reach for the process that hurts. The owner who dreads writing estimates. The shop that loses a week every month reconciling invoices. The service business where one person is the only one who can talk a nervous customer down. All of that pain is real. None of it, by itself, tells me the process is ready for an agent.

Painful work is usually painful because somebody in it is making a judgment call, and judgment is the last thing you hand to a beginner, whether that beginner is a new hire or a new agent. Automate the painful process first and you have not removed the judgment. You have just made it harder to see where it went missing.

The four filters I actually use

This is the working method I bring into the room, the same one that used to sort spreadsheet requests and now sorts agent candidates.

What week one looks like

Week one never starts with the agent. It starts with the process, written the way it actually runs today, exceptions included: the client who always asks for the discount, the invoice that never matches the purchase order, the job that comes in on a Friday afternoon. I ask whoever owns the process to define, in one sentence, what a correct output looks like: something the reviewing person can check against in under a minute.

Only then does the agent run, and it runs alongside the person who used to do the work by hand, for a few cycles, not one. Every correction goes back into the written process, never into a private prompt that only one person can see. That is the same discipline I have carried since Tupy and since Sizebay: the value is never in what I know, it is in what the team can keep using after I leave the room. By the end of the first week, the team owns two things: a document and a habit of checking it. The document is worth more than the agent.

Then you repeat it on the next process, and this time the team runs the loop without me.

What this has to do with Holybiz

This is not a new skill I picked up because agents arrived. It is the same craft I have practiced since Tupy and since Sizebay, aimed at a new kind of hire. At Holybiz, this is where every engagement starts, the way Janderson and I run our own company: map how the work actually happens, find the one piece that is ready to hand off, and leave the owner with a team, agents included, that they can run without us.