On July 5, 2026, Vincent Luciani, Co-founder and Executive Chairman of Artefact, was the guest on the Generation Do It Yourself podcast, hosted by Matthieu Stefani, for a three-hour interview. This conversation pulls back the curtain on the behind-the-scenes reality of the artificial intelligence industry, from corporate data governance to the true state of digital sovereignty.

1. The Myth of the Race for the Latest Model (A 3% Commodity)

The massive investments by American tech giants, whose infrastructure spending will reach 700 billion dollars next year, sustain a very specific strategic narrative: the idea that companies absolutely must acquire their latest closed-source models to remain competitive.

In reality, in 2026, underlying AI models have become a commodity. The average performance gap between a free (open-source) model and a paid (closed-source) model is only about 3%. Yet, most companies today are only utilizing 0.001% of these tools’ actual capabilities. Today, the true differentiator is not the model being rented from American giants, but rather proprietary data and your organizational context.

Vincent Luciani, Co-founder and Executive Chairman of Artefact, anticipated this shift as early as 2014, back when he had to define the word “AI” at every client meeting. Today, with 2,500 employees, Artefact is Europe’s largest independent data and AI consulting firm, and its conclusion is definitive: without clean data, even the most powerful model is worthless.

“It’s not AI that makes the difference; it’s the quality of the data already in place within the company.”, Vincent Luciani, Co-founder and Executive Chairman of Artefact

2. The “Token Maxing” Trap, or the Illusion of Productivity

The core of the lie lies in a cultural practice that is highly lucrative for Silicon Valley: “Token Maxing.” Tech giants exert immense pressure on developers to overconsume computing units (tokens). Vincent Luciani specifically points to Jensen Huang, the founder of Nvidia, who publicly claimed that a developer paid 500,000 dollars a year ought to consume at least 250,000 dollars worth of tokens. This culture has become so toxic that rumors suggest developers have been fired simply for not spending enough tokens, while startups are offering tools to artificially use tokens to simulate high activity.

For companies in the real economy, the financial consequences are disastrous:

  • Uncontrolled budget explosions: Mature companies like Uber have burned through their entire annual AI budget in just four months.
  • The danger of autonomous loops: Modern AI architectures allow autonomous agents to spin up sub-agents to solve a problem. Left unsupervised, these scripts loop all night long. Vincent Luciani notes the case of a client who received a surprise bill of 150,000 dollars in a single night due to a single, poorly phrased analytical query.
  • The productivity paradox: Silicon Valley hypes up technical teams that produce 7 to 8 times more lines of code thanks to AI. However, an MIT study demonstrates the emptiness of this metric: these companies aren’t generating more revenue, nor are they shipping significantly more products. “Token Maxing” is often nothing more than a massive transfer of wealth from corporate treasuries to Silicon Valley servers.

3. The Illusion of Knowledge and the Decision Bottleneck

This mirage of productivity masks a cognitive decline that Vincent Luciani illustrates through a study conducted on students. When faced with an assignment (whether in history, math, or coding), the group of students with access to AI consistently gets much better grades. However, when the test is repeated a few days later without AI, their performance collapses entirely (dropping from a 10 to a 1) because simple copy-pasting bypasses the mental effort required for information retention.

The exact same phenomenon happens in the corporate world. Placing AI at the beginning of a creative or strategic process encourages intellectual laziness. This is why the operational bottleneck has shifted: it is no longer about producing, but about deciding and validating.

In consulting, for instance, tools like Granola (speech-to-text) and AI scripts can generate a commercial proposal and slides within a minute of walking out of a client meeting. But Artefact forbids its consultants from doing this. AI must be introduced at the very end of the process, solely to formalize deep human thinking done beforehand—otherwise, it yields nothing but a bland statistical average.

“The real bottleneck today is no longer production, but the human validation of decisions.”, Vincent Luciani, Co-founder and Executive Chairman of Artefact

4. The Antidote: “The Harness” and the 4 Data Steps

To escape this profit trap, companies must switch to a strict control framework called “The Harness.” This involves setting up rigorous governance to monitor compute costs, track technical model drift, and restrict data access rights.

Building this harness and making AI truly profitable requires getting back to pragmatic basics, summarized in a 4-step roadmap to clean up data:

  • Verify accuracy at the source: Audit existing tools. In most companies, foundational data is either missing or incorrect (for example, one out of two sales reps forgets to update the CRM after a client call).
  • Unify taxonomy: Ensure words mean the same thing everywhere. If a “customer” means “someone who ordered” in France but “someone who paid” in Spain, the AI will blend inaccurate data.
  • Appoint domain owners: Data spoils instantly. Data must be continuously maintained by leaders who are held directly accountable for their scope (CRM, HR, Finance).
  • Enforce a Single Source of Truth: Permanently ban the dozens of Excel files circulating among teams with different grid versions, and centralize all flows into a single, clean, and queryable database.

It is only when this informational asset is properly structured that AI can be plugged in effectively. It can then break down company silos by horizontally connecting finance, HR, sales, and recruitment.

5. Ignore the Hype: Artefact’s Survival Lesson

This discipline in the face of Silicon Valley’s siren song is what allowed Artefact to survive its own crises. In 2017, when it was just a young 50-person outfit, the company pulled off a reverse merger with Net Booster, a publicly traded advertising agency three times its size (600 employees across 20 countries).

The operation instantly turned into an operational nightmare. Net Booster lost legacy clients at a breakneck pace, the UK subsidiary’s directors resigned the day after the deal, and a surprise VAT tax audit hit the German subsidiary. EBITDA plummeted to 400,000 euros, teetering on the edge of regulatory bankruptcy. On the Boursorama forums, attacks were daily, and the stock crashed to 50 cents.

To pull through, Vincent Luciani and his partners made a radical choice: ignore communications regarding their legacy core business and bet exclusively on enterprise data and machine learning. They cut costs, shut down unprofitable countries, and leveraged the inherited international network to set up data consulting cells.

This rigor allowed them to steady the ship, finalize the company’s delisting from the stock market at 8 euros per share—representing a total transaction of 300 million euros—before being valued at over one billion euros in subsequent rounds with funds Ardian and Cinven.

Watch the podcast on Youtube