Skip to content
Nahuel SoriaES

AI agents and automation for companies

Nahuel Soria designs AI agents that do real operational work: triaging errors, watching production systems, writing and publishing content. Every day he runs his own fleet of about 20 agents that keeps his products maintained, with the controls written in code rather than in the prompt.

fromUSD 3,000quoted by milestones

fromUSD 3,000quoted by milestones

What a useful agent does

A repeated task with a result that can be checked. My fleet has one agent that hunts bugs every night and opens pull requests, another that triages production errors with no ability to change anything, and a whole news site, Onda Corta, that writes and publishes articles after passing plagiarism, fact and duplicate checks that run in code.

Why the controls live in code

The rules that matter are written as validators that reject the model's output, because an instruction in a prompt can be ignored. Every agent starts read-only and reports; write access is earned task by task, with caps and review. None of mine can move money.

How to start

With a single agent on a task someone does by hand every week today. It is measured for a month before deciding whether to widen its scope. It is quoted by milestones, like custom software.

See it working

Frequently asked questions

Can an agent break something in production?

Every agent starts read-only and reports. Write access is granted per task, with caps and review. In my fleet only one writes code, and it does so through pull requests.

Which models do they use?

Claude as the primary model, with other providers as fallback. The choice depends on the task and the cost.

Is it useful for a small business?

Yes, when there is a repeated, checkable task. Manicuría CABA, a home-visit business, publishes to Instagram automatically.

Order sheet

Tell me what you need. The first call is 30 minutes, it is free and you leave with a price.

Book a call
Calendly