AI First thinking: how an AI First consultancy works on the inside

An AI First consultancy is a services company that takes artificial intelligence as the starting point of every process, not as an add-on: faced with any task, project or internal tool, the first question is always “how can this be solved with AI?”. AI First thinking isn’t about using more artificial intelligence tools, it’s about changing the order of decisions: first the problem is described, then the solution is built with AI, and only the traditional method is used when AI falls short.

In this article we explain what that shift in mindset means and show it through a real, verifiable case: El Mundial de Cronuts, a complete application for our team’s World Cup 2026 sweepstake, built entirely with generative AI.

# What AI First thinking is (and isn’t)

AI First thinking is a decision-making criterion, not a technology. The term became popular when Sundar Pichai, chief executive of Google, wrote in his first shareholder letter as CEO, in April 2016, that the company would move from thinking “mobile first” to thinking “AI first”; Google later confirmed it as corporate philosophy on its official blog. Since then, it has defined organisations that design their processes assuming AI will be involved from the very first moment.

Being AI First isn’t having ChatGPT, Claude or Gemini licences for the team. Nor is it publicising that “we use artificial intelligence”. The difference shows in the order of the questions:

The practical consequence of AI First thinking is that small ideas stop being dismissed for lack of resources. When building a prototype costs little, many more things get tried, and that culture of experimentation is what later gets passed on to clients. Stanford’s own AI Index 2025 confirms the underlying trend: the use of generative AI in at least one business function went from 33% in 2023 to 71% in 2024, a rate of adoption no previous business technology had reached in a single year.

# How an AI First consultancy works on the inside

An AI First consultancy is recognised by a trait that’s awkward to fake: it uses AI in its own processes before recommending it elsewhere. Internal consistency is the proof of credibility — in English it’s called eating your own dog food: consuming what you sell.

In practice, this translates into three observable behaviours:

• Internal processes are automated first. Reports, data analysis, content and in-house tools are handled with AI agents before hiring someone or commissioning custom development is even considered.
• Natural language replaces technical specification. Briefs are written the way you’d explain the problem to a colleague; AI acts as the bridge to code. This approach is known as vibe coding, a term coined by Andrej Karpathy — OpenAI co-founder and former director of AI at Tesla — in a post on X on 2 February 2025, also named 2025 word of the year by the Collins dictionary. Karpathy described it as going with the flow of generated code without reviewing it line by line; in a professional context, that idea translates into describing the problem well and letting AI write a first version that a person then oversees and corrects.
• The whole team builds, not just the technical staff. In an AI First company, knowing how to frame a problem well is worth as much as knowing how to code it.

# A real case: El Mundial de Cronuts

El Mundial de Cronuts is the internal demonstration of our AI First thinking: a complete web application for the team’s World Cup 2026 sweepstake, built with generative AI from start to finish. We could have used a shared spreadsheet. We decided to do it the way we do everything: by describing what we wanted and building it with AI.

The process: from a PDF and a set of rules to an app in production

The process started from just two inputs, neither of them technical:

1. FIFA’s official schedule: a PDF with the tournament’s 104 matches.
2. Our points rules, written in natural language, the way you’d explain it to a friend.

With Claude Code, Anthropic’s coding agent, those two documents became a real application: a web interface in React, a real-time database in Firebase and cloud deployment with serverless functions. We then connected it to the official results API (football-data.org): every match that finishes is imported automatically, recalculates each participant’s points and updates the ranking. The World Cup’s 104 matches, in real time, with no one entering anything by hand.

La Gazzetta della Porra: AI-generated content every matchday

The app’s most celebrated feature isn’t technical, it’s cultural. La Gazzetta della Porra is an internal sports newspaper that an AI writes every matchday from the real standings data: a dramatic headline in the style of the sports press, player of the day, surprise of the round… and zero mercy for whoever’s at the bottom of the table.

The result is that AI doesn’t just automate the sweepstake: it turns it into the team’s daily conversation. That’s the nuance that separates “using AI” from thinking AI First — technology in the service of something as human as ribbing each other over a sweepstake.

# What a team learns when it builds with AI

Building internal products with AI is training in disguise. Every skill practised in El Mundial de Cronuts has a direct application in client projects:

That’s why we call it the most cost-effective team building we’ve ever done: while the team competes over the sweepstake, they’re training exactly the capabilities we later apply in our AI products for businesses and in our clients’ SEO and GEO positioning.

# How to apply AI First thinking in your company

Adopting an AI First approach doesn’t require a two-year transformation plan. These five steps replicate the process we follow internally:

1. Choose a small, real internal problem. Something the team puts up with or enjoys every day: a repetitive report, a missing tool, a World Cup sweepstake. Starting small lowers the risk and speeds up the first visible result.
2. Describe it in natural language. The brief is the new code: what you want, what rules it has, what data it uses. If you can explain it to a colleague, you can explain it to AI. The more specific the description, the more reliable the result.
3. Build it with an AI agent. Tools like Claude Code turn that description into working software that a human oversees and validates. The goal isn’t to eliminate human review, but to shift it from the code to validating the outcome.
4. Connect it to real data. An API, a database or a spreadsheet: the difference between a toy and a tool is that it feeds itself. Without real data, a prototype stays a demo.
5. Iterate on the team’s feedback. Every improvement is a short cycle: it’s requested, tested and adjusted. That’s how judgement gets trained, which is the one thing AI can’t supply.

# Common mistakes when adopting an AI First approach

The most frequent mistakes when trying to be an AI First company are about approach, not technology:

• Starting with the tool rather than the problem. Buying licences without deciding what they’ll be used to solve produces anecdotal use, not a change in culture.
• Getting stuck in the eternal pilot. A prototype that never makes it into daily use generates neither learning nor trust. The risk is real and has been measured: according to the The GenAI Divide: State of AI in Business 2025 report, from MIT Media Lab (the NANDA initiative), 95% of generative AI pilot projects in businesses fail to deliver a measurable impact on financial results, and only 5% manage to become embedded in processes and generate real value.
• Delegating AI to the technical department alone. The value of AI First thinking shows up when marketing, operations and management build too.
• Measuring only savings. The interesting metric isn’t how much less it costs, but what was previously impossible that’s now possible.
• Hiding the process. Showing how something is built — inputs, tools, limits — builds more trust than boasting about the result.