AI First: what it is + how Fhinck became Brazil's 1st

TL;DR: An AI First company does not use AI — it operates on AI. The 10 principles that guided Fhinck's refounding (Brazil's first AI First company, founded in 2014) mark the gap between AI Enabled and AI First: 3x to 10x in operational results. A manifesto for C-level executives deciding architecture, not add-ons.

The 10 principles that guided Fhinck's transition to Brazil's first AI First company. No theory. No motivational narrative. Practical decisions we made — and would make again.

AI First vs AI Enabled vs AI Adopter — the stage that decides ROI

Before reading the 10 principles, it is critical to distinguish three stages of AI adoption that are commonly conflated. The difference is not semantic — it is 3x to 10x in real operational indicators.

AI Adopter is the company that uses AI in point cases. It has a corporate ChatGPT, a service chatbot, perhaps a copilot in the code editor. AI shows up as a personal productivity tool. Result: marginal gain (5–15% productivity on specific tasks), no structural change to operations.

AI Enabled is the company that integrated AI into some areas as a support layer. HR uses AI for resume screening, marketing uses AI to generate copy, support has AI on the front line but humans always behind it. AI is part of the stack, but the organizational architecture remains human — AI assists humans, it does not replace them.

AI First is the company where AI is the default of the operational architecture. When a position opens, the question is "is this work for a human or an agent?" When the answer is "agent," the agent is built — with no human backfill. Processes are designed end-to-end assuming machines will execute; humans enter only as qualified exceptions (strategic decisions, ethical context, targeted intervention). See the difference between AI First and AI Adopter in detail.

Fhinck crossed this transition between 2022 and 2025. It was not linear or comfortable. It was a refounding. The result is in the numbers at the end of this manifesto, and in the 50-to-6 case which describes the process in detail.

The 10 principles of Fhinck AI First

1. Rebuild > Add

We did not add AI to an existing company. We rebuilt the company for the era of autonomous agents.

The difference between "using AI" and "being AI First" is the difference between installing solar panels on an old house and building a house that breathes sunlight from the foundation. Both generate energy. Only one is efficient.

AI Enabled companies try to add AI to existing workflows — and stumble because the workflow was designed for humans. AI First companies redesign the workflow assuming AI by default. That is the concept of redesigning work for AI to take over, and it is where 95% of AI initiatives fail.

2. Speed > Perfection

We prefer to iterate fast rather than plan indefinitely. We learn by doing, not merely by thinking.

In a field that changes every week, a 12-month plan is obsolete by week 8. The competitive advantage belongs to those who execute fast, measure, learn, and adjust.

AI First is not a state — it is a cadence. Whoever waits to have "everything clear" before starting loses 12 months for every 3 months of those already executing.

3. Autonomy > Hierarchy

Every Fhincker has the freedom to make decisions. We trust in competence, not in cascading approvals.

Cascading approval kills speed. And in a lean company with agents, speed is survival. Trust the team. If they err, adjust. If they succeed, scale.

4. Impact > Activity

We measure results, not hours worked. What matters is transforming clients, not appearing busy.

Traditional companies value those who are visibly occupied. AI First companies value those who solve. The difference is radical in recruiting, promotion, and culture.

5. Innovation > Comfort Zone

We embrace the discomfort of exploring emerging technologies. If we are not a little uncomfortable, we are stagnating.

Comfort is the symptom of falling behind. In AI, staying comfortable for 3 months means being 12 months behind.

6. Replace > Assist

When a position opens, we ask: "is this work for a human or an agent?" When the answer is "agent," we build the agent — without backfilling.

Most companies build AI "assistants" for humans. Result: marginal gain, invisible ROI. AI First replaces within defined scopes. Brutal ROI.

This is the most radical principle of the manifesto and what most separates AI First from the other categories. AI Enabled adds assistants. AI First redesigns the scope of work so that the agent takes over — humans handle only high-complexity exceptions.

7. Learn Continuously, Unlearn Old Things

Every Friday, the entire company stops working and enters a classroom. We learn one new AI technique. Sharpening the Axe.

The pace of the AI market leaves no room for "we'll study when we have time." Anyone without a weekly learning routine is a quarter behind in 6 months. No exceptions.

8. Modern APIs = A Matter of Life or Death

Every critical system is evaluated by the quality of its API. Without a modern API, it is strategic debt that must be resolved.

An agent without an API is a blind agent. A legacy system without a modern API is the invisible bottleneck that paralyzes every AI First transformation.

9. Builder > Delegator

The AI First CEO gets their hands dirty. They build their own agents. They speak as an equal with the technical team. They publicly demonstrate what they know.

C-level executives who say "AI is for the technical team" have lost the game before it started. In 2026, that is equivalent to saying in 2010, "the internet is IT's job."

Being AI First starts with the CEO. You cannot outsource understanding of AI to a CTO and expect the entire company to transform — transformation flows from leadership that knows how to build agents, not from leadership that knows how to hire builders.

10. For Adventurers, Not the Complacent

Fhinck is not for everyone. If you seek predictable stability, rigid processes, and strict hierarchies, there are better companies for you.

But if you want to build the future of work, if you believe AI can democratize capabilities once reserved for large corporations, if you want to be part of something pioneering — you are in the right place.

AI First in practice — real numbers from Fhinck

A manifesto without evidence is motivational theory. These are Fhinck's operational numbers after adopting the 10 principles between 2022 and 2025:

From 50 to 6 people — lean team operating what previously required 8x the headcount. Details in the 50-to-6 case.
Revenue doubled — 2x revenue while keeping the team lean. Marginal cost near zero per new client served.
96% of customer service operated without humans — autonomous agents resolve 96% of tickets end-to-end. Humans intervene only in 4% (high-complexity exceptions or critical decisions).
800,000 active users across 15 countries — international scale without proportionality between headcount and users.
10+ years of Task Mining in production — Fhinck was founded in 2014 with the thesis of capturing corporate digital activity, well before the generative AI wave. That means real operational data to train agents, not a promise.

These numbers are not projections or fictional cases. They are the measurable result of applying AI First structurally, not as a veneer. The difference between AI Enabled and AI First shows up exactly here — in measurable ROI, not in narrative.

The question this manifesto leaves for you

If you are a CEO or board member, read the 10 principles again. For each one, answer honestly: does your company operate this way today?

If most answers are "yes," you are probably building the first AI First company in your sector.
If most are "no, but we are moving in that direction" — you have a window.
If most are "no, and we shouldn't" — you are operating a model that will face significant headwinds over the next 24 months.

Fhinck lived this manifesto and is reaping the results: 50 → 6 people, revenue doubled, 96% service coverage without a human operator, 800,000 active users across 15 countries.

Learn more about the AI First company concept or see how Task Mining captures the operational data foundation that makes AI First possible at /en/plataforma/task-mining.

Want to follow this path with method? Schedule a conversation.

Frequently asked questions about AI First

What does it mean to be an AI First company? An AI First company is one where AI is not an add-on — it is the operational infrastructure. Decisions, processes, and autonomous agents run on top of artificial intelligence by default.

What is the difference between AI First, AI Enabled, and AI Adopter? An AI Adopter uses AI in point cases (chatbot, copilot). AI Enabled has AI integrated in some areas as a support tool. AI First has AI at the core of the architecture — humans are qualified exceptions, not the rule. The outcome difference is 3x to 10x in operational indicators.

How does a traditional company become AI First? It is not a migration — it is a refounding. It starts by mapping what humans do today (Task Mining), identifying what AI can operate autonomously, and reorganizing processes so that AI is the default option and humans are the qualified exception.

Why is being AI First different from "using AI in the company"? Using AI in some areas is point automation. Being AI First is architecture — AI is at the core, not the periphery. The difference shows up in execution speed, marginal cost near zero, and the ability to operate 24/7 without proportionality between output and headcount.

What results has Fhinck achieved by being AI First? Fhinck went from 50 to 6 people while maintaining output, doubled revenue, runs 96% of customer service without a human operator, and serves 800,000 active users in 15 countries. It is the n=1 case of what AI First delivers when applied structurally, not as a veneer.

Are the 10 manifesto principles mandatory or suggestive? These are decisions Fhinck made and would make again. They are not mandatory — but companies that ignore most of them typically remain at the AI Enabled stage, not AI First. The outcome difference between the two is 3x to 10x in operational indicators.