Meet samambai: AI that runs in production, with prices on the table

samambai is live, and the prices are on the table before anyone books a call. We are an engineering studio for AI, automation, and software. We are based in São Paulo. We deliver in English and in Brazilian Portuguese. Companies in São Paulo, New York, Madrid, and Tel Aviv have already worked with us.

Where we came from

samambai grew out of PowerAds. The name is new. The habit is not: learn how a company actually runs, then ship something that stays on, instead of a deck about what AI might do next year.

The studio is in São Paulo. The work is not. Buyers sit in Brazil, the United States, Spain, Israel, and farther out, as long as the problem fits what we build. We work in the language the client's team already uses.

"Cultivate" is not a campaign line for us. It is the delivery pattern. We start with one focus, put it into production, and only then branch into the next system. A twelve-month transformation with no working software in the middle is the wrong room. A workflow, an agent, or an app that is live in a few weeks is the right one.

What we build

Four lines of work. Each one ends in production, not in a demo that nobody opens again.

Taking a Lovable app to production. Some teams design the product in Lovable, with Supabase behind it, and stall when a real customer has to pay. A multi-tenant B2B SaaS for gyms started that way and now runs in production with paying customers. The service page is Lovable to production.

WhatsApp AI agents tied to the client's systems. Not a bot on an unofficial session. The official WhatsApp Business API, the company's own rules, an integration, and a handoff to a person with the thread attached. Agents we built are in production for a medical clinic group (front desk and scheduling), for hospitality and B2B retail, for an adventure park, for a condominium administrator, and for a visa advisory. The service page is WhatsApp AI agents.

Automation on n8n. Workflows that connect a CRM, an ERP, an e-commerce stack, a spreadsheet, and WhatsApp, instead of someone retyping the same row. We run our own n8n server, with dozens of flows in production: service, integrations, and routines. The service page is n8n automation.

GEO. Getting a site into a shape that ChatGPT, Gemini, Perplexity, and Google's AI can read, trust, and cite. We do not promise a rank. We measure a trend on a fixed set of questions. The service page is GEO.

Around those four, we also build custom software (apps, platforms, and internal tools) and do AI consulting when the missing piece is a map, not a tool. Internal systems we have put into use include staff hours for an education institution, treasury for a financial group, and media spend for a consultancy. Client names stay off this page. The proof is the kind of operation, not a logo.

How we work

Four steps: diagnosis, strategy, execution, evolution. The cadence is a two-week sprint, with something real at the end of each cycle. Most automations go live in 2 to 4 weeks. Full projects take 4 to 8 weeks.

Prices are public, in US dollars, and spelled out on pricing:

• Sprint: US$ 1,200 for a two-week sprint. One focus at a time. Cancel whenever you want.
• Project: from US$ 3,000. Scope, timeline, and price are fixed. A delivery every sprint. Support after launch.
• Partnership: from US$ 900 a month. A continuous roadmap, priority, and a monthly check-in.

Our own operation is AI-first in a concrete sense: a human team plus AI agents inside the build, the tests, and an independent audit. None of that replaces a human review. An agent drafts, tests, and flags a failure. A person decides what ships.

Client data follows three rules. Least-access credentials. Data stays in the client's own accounts. LGPD and GDPR practices, and client data never trains a public model.

What we do not do

We do not install a generic chatbot and disappear. We do not put a customer number on an unofficial WhatsApp client to "save money." We do not rip out a CRM, an ERP, or a spreadsheet that already works, if we can integrate with it. We do not promise revenue, a rank inside an AI answer, or a Meta template approval on a named day. We do not sell a shrink-wrapped product license under our name. We do not take on a large-firm transformation program in which the first working software shows up next semester.

We also do not publish a client count, a message count, or hours saved. If a number shows up in our writing, it is a price or a timeline from this page, or it is a public figure with a source. An anonymized case describes the kind of operation. It does not describe a percentage you cannot check.

When not to call us

Do not call if you want a talk about AI, a maturity report, and nothing in production. Do not call if a vertical product you already pay for books the appointment, talks to your system, and the rule fits inside that product. Do not call if the budget only works with an unofficial WhatsApp session. Do not call if nobody on your team can take the thread when the agent does not know the answer. In that case the agent becomes a wall, and we do not build walls.

Call if the problem is specific: the Lovable app has to survive real customers, WhatsApp has to read the calendar or the order, n8n has to join two systems that do not talk, or the site has to be readable by the AIs your buyers already ask. What we publish lives on the blog.

Thirty minutes, no commitment

The first call is 30 minutes. No contract at the end of it, no deposit to "hold the date." We look at the system you already run, the channel your customer already uses, and the rule that is breaking. If a ready-made product solves it, we say so. If a sprint, a project, or a partnership fits, you get a proposal with scope, timeline, and price. The three numbers are already on pricing, so you can show up knowing the order of magnitude.