AI has become essential, carving out a key role in our professional lives. But in 2026, the time for simple testing and discovery is over; it is now time to scale. 

The reality is clear: nearly 80% of POCs will never be deployed. This trend was confirmed during our AI Afterwork event, co-organized with X-PM. 

Many of our guests have integrated AI tools into their daily routines, and some have even launched POCs, but few have managed to scale them successfully. 

So, how do we bridge this gap? What are the most common barriers to AI adoption, and which use cases should be prioritized to maximize the chances of success? 

These are the questions that Ludovic CINQUIN, Founder of WhereWeGo and former CEO of OCTO Technology, Justin DERBYSHIRE, Associate Director of IT, Tech & Digital Transformation at X-PM, and our CEO, Homéric DE SARTHE, addressed during this roundtable. 

Justin DERBYSHIRE: "2026 is a true year of maturity and the emergence of real production-level projects. Until now, we were mostly in a phase of anticipation. Pure players are already well ahead."

Homéric DE SARTHE: "Everyone is talking about AI. It’s the topic on everyone’s lips, yet it is a subject that is currently very poorly understood, if understood at all."

‍

Between rapid evolution and limited adoption: the constraints hindering widespread implementation

‍

Despite the government's ambitions outlined in the France 2030 plan—aiming for 100% of large accounts and 80% of mid-sized companies to be equipped by 2030—the reality on the ground reveals a significant gap. Current figures show an adoption rate of only 9-10% in France, compared to over 25% in Nordic countries like Sweden, Norway, and Finland.

While AI is increasingly embedded in the daily lives of businesses, this revolution is experiencing two-speed adoption due to certain barriers to entry: 

‍

1 - A pace of evolution that far outstrips the pace of absorption

With approximately 3.8 major updates per month, AI is evolving fast—very fast, perhaps even too fast. This frantic pace of progress makes it difficult for companies to adapt quickly enough to "stay in the race," especially considering that a true transformation typically requires 3 to 6 years. 

‍

2 - A desire for immediate results

With its rapid evolution and vast capabilities, AI can create the illusion of a "superpower," and many organizations have paid the price. For example, Salesforce laid off 6,000 employees in the name of AI efficiency, only to end up rehiring half of them. 

This is where many companies go wrong, hoping for significant results within the first few months without taking the time to properly prepare the project or adapt their operations and data.

Ludovic CINQUIN: "As long as you don't change the way work is organized, there are no gains to be made with AI. Just because I can answer 40 times more emails with AI doesn't mean I won't end up sending 50 times more."

‍

3 - Poor data quality

As mentioned earlier, some organizations launch ambitious projects while feeding their (future) AI agent erroneous, outdated, or low-quality data. This leads to the "garbage in, garbage out" phenomenon, which often results in the project being abandoned.

‍

4 - A need for stability and a fear of "risk"

Integrating AI often requires rethinking all processes within an organization, which puts the responsibility on CIOs. 

Ludovic identifies a major obstacle: "No pain, no gain. If you want to gain something from AI, you have to rethink your processes, and that is incredibly complicated. Another thing is that when you have a business that works, taking the risk of breaking everything by introducing AI is a pretty bold move."

This need for stability and security sometimes hinders the adoption of sovereign solutions, as CIOs and companies prefer to turn to Big Tech for their monopoly and reputation for "robustness," thereby creating a dependency on American players. 

Homéric DE SARTHE: "If Europe doesn't have several decacorns today, the responsibility lies with all the technical directors who chose the simplicity and job security of picking American players over European ones, often blindly following the recommendations of the major consulting firms that guide and reassure them. The reality is that without financial resources, investors, or clients, it will never be possible for European companies to scale up and catch up on the lost ground. We have the talent, but we don't have the means."

‍

Obstacles exist, but they are not insurmountable

‍

As we have seen, there are many obstacles in the "AI race" that many companies have embarked upon. 

However, our experts also shared best practices for overcoming them: 

‍

1 - Self-questioning is essential for progress 

In an environment that changes daily, the ability to question oneself and remain curious is fundamental to identifying areas for improvement and the adjustments needed to get a project off the ground.

‍

2 - Support is a factor that should not be overlooked

AI adoption is not just about technical teams; in fact, one of the most common obstacles is the reluctance and lack of knowledge among business teams. A solid change management roadmap is therefore essential and requires a larger share of the budget than IT, with a recommended ratio of 30% technical to 70% support.

‍

3 - Preventing "Shadow AI": When employees take the lead

While more and more companies are adopting AI, so are their employees—often without informing management. This leads some to use consumer-grade AI agents, sometimes inputting confidential information or documents and compromising corporate data security. Clear governance and an internal policy are therefore essential to ensure data security and regulate the use of AI within the organization.

‍

The Evolution of AI: Upcoming Trends to Watch

‍

To successfully adopt AI, it is important to anticipate its future developments. Our speakers highlight three major changes on the horizon: 

For Homéric, this means strengthened corporate governance through strict control of information access based on sensitivity levels, modeled after the banking sector. 

It also involves the hybridization of work between humans and AI agents, who will learn to work together, coupled with the rise of SLMs (Small Language Models) deployed directly into professional hardware, allowing companies to secure their value without feeding external LLMs. 

For Ludovic, the next frontier to cross will be robotics and Large Action Models: the physical equivalents of LLMs, they are revolutionizing robotics with now-mastered movement and stability capabilities.

Ludovic: "Chinese robots are just incredible, and ten times cheaper. Movement, which was the biggest challenge five years ago, is now mastered."

‍

Our Experts' Practical Recommendations for 2026

‍

This roundtable concluded with some sound advice on how to best integrate AI into processes at various levels: 

For Executives 

• Train in the fundamentals of AI to make informed decisions 
• Use AI daily to understand its true capabilities
• Learn to distinguish between hype and real business opportunities

For CIOs 

• Shift from a defensive stance to a proactive approach 
• Support business units instead of hindering initiatives 
• Secure usage rather than banning it 

For teams 

• Master prompting to maximize efficiency
• Prioritize secure company-provided environments over personal accounts
• Participate in pilot phases with resilience, accepting initial frustration

‍

Do you have an AI project and want to give yourself the best chance of success for deployment and scaling? Get support from our experts! Contact us