AMIGO is born: personal AI should cost pennies, not a subscription

Today I started AMIGO. The first commit sits in the repository with a timestamp, and it will stay there forever — not because I need to prove anything to anyone, but because keeping a record is a good habit. An idea with no birth date is an idea nobody can trace back later.

The thesis that brought me here is simple, and it has chased me for years: why does personal AI charge a flat subscription? I pay the same in January, when I barely open the app, as in the month I use it to solve ten problems back to back. That is not pricing — it is the vendor betting that average usage covers the extremes. Great for whoever sells it. Bad for whoever uses it, and, more importantly, a bad long-term business model for whoever actually builds it.

Twenty years understanding what a query costs

From 1999 to 2012 I was at Catho, building and owning the company’s job and résumé search engines on Apache Solr and Lucene — relevance tuning, synonyms, a 13-person multidisciplinary team reporting straight to the presidency. From 2010 to 2022 I was co-founder and Chief Data Officer at Semantix, which at its peak managed petabytes of data and billions of transactions a month, and which listed on NASDAQ in 2022 under the ticker STIX — one of the first Latin American deep techs to do so.

In between, I put search into production in places that do not forgive architectural mistakes: Magazine Luiza, Itaú, Vivo, Petrobras, DoiNet’s base of 70 million documents, the G1/Globo.com news recommendation engine running on Hadoop. Every one of those systems processes a query volume that only makes financial sense if the marginal cost of each search is pennies, not dollars.

Two decades of that life teach you something the generative AI industry seems to have forgotten: every query has a marginal cost. Search, indexing, inference — all of it costs pennies, not dollars, when the architecture is built right. A monthly AI subscription bill rarely reflects that real cost. It reflects what the market will accept without asking questions.

One assistant is not enough

The second half of the thesis: nobody wants “an” AI assistant. People want specialists. One to review a contract, another to think through strategy, another to just talk without judgment. Single model, generic prompt, neutral personality — that was already the minimum viable product of 2023. AMIGO is born around the idea of multiple specialized assistants inside the same app, each with its own voice — literally, with speech — and its own job to do.

Synapses, not a subscription

That is why AMIGO is born with a synapses model: you pay for what you use, not for the right to use it. That changes the relationship with the tool. You do not feel guilty for not opening the app this week. You do not feel like you are being billed for intelligence you never spent. It is pay-as-you-go applied to cognition, and it is how I believe personal AI actually becomes mainstream — when it costs pennies per interaction instead of a flat monthly fee nobody questions.

What Arvor is betting on

AMIGO is Arvor’s first consumer product, and it carries the thesis that guides everything we build here: specialized agents, cost aligned to real usage, architecture built to scale without turning into a generic subscription. If your company wants to think about agentic AI with the same discipline around cost and specialization, talk to us through Arvor’s agentic consulting.

---

Update: AMIGO has left the private repository behind and become a production app — iOS, Android and Web — with a marketplace of specialized assistants anyone can use. The synapses thesis still holds: you pay for what you consume, not for the right to have access.