AI in delivery: between promises and reality, how to move from gadget to operational value?
Artificial intelligence is everywhere. It’s discussed at conferences, in innovation reports, and in strategic presentations. Yet, in the day-to-day work of IT teams and digital projects, its real impact often falls short of expectations. Why such a gap? Is it a question of technological maturity, governance, or simply approach?
Cosmetic AI: a hidden cost
Decorative chatbots, isolated automations, projects difficult to maintain… AI is sometimes reduced to a technological showcase, without any real integration into operational processes. The result: costly investments, but few tangible gains in terms of performance, quality, or delivery times. What if the real challenge wasn’t adding AI, but integrating it where it truly matters?
Vibe-coding and code assistants: acceleration or chaos
AI tools promise to reduce development time by a factor of five, or even ten. Yet, the reality is more complex. As Google’s DORA report highlights, AI amplifies the efficiency of already mature teams, but can plunge less industrialized projects into chaos.
The massive production of code generated by AI then becomes difficult to absorb, test, and deploy securely. So, how can we avoid the trap of superficial innovation?
AI in the service of delivery: an overlooked lever
Did you know that 30 to 40% of the time and budget for digital projects is spent on tasks that are not very distinctive, but essential? Formalizing requirements, writing user stories, testing, documentation… These are all steps where AI, when properly integrated, can make a difference. By verifying the completeness of requirements, assisting in the writing of specifications, or generating large-scale test scenarios, it can significantly reduce design and validation time. Productivity gains of 30 to 40% have already been observed in some upstream phases. So why not make it a cornerstone of your delivery strategy?
Human in the loop: the essential balance
Contrary to popular belief, the best AI isn’t the one that impresses in demonstrations. It’s the one the end user doesn’t notice because it’s seamlessly integrated into business processes. Attempts at mass automation have shown their limitations, highlighting the need to keep a human in the loop to ensure reliability and risk management. How do we find the right balance between automation and human expertise?
Governance, security and sovereignty: the pillars of ROI
Gartner estimates that up to 85% of AI projects fail to meet their objectives due to a lack of quality data, appropriate governance, or integration with the information system. Not to mention the often underestimated operating costs, particularly those related to inference.
Cybersecurity is also a major challenge: language models represent a new attack vector, making input security essential. Finally, not all use cases require large, general-purpose models. Fine-tuned, on-premises open-source models offer a more controlled and sovereign alternative. How can we reconcile performance, security, and sustainable ROI?
Towards a pragmatic and business-oriented AI
Truly useful AI is neither visible nor spectacular. It is integrated into the information system value chain, resulting in more reliable, faster, and more secure applications. It is not added at the end of the chain, but rather permeates the entire lifecycle of the digital product. The next wave of transformation will not come from a more impressive AI, but from a more responsible, pragmatic AI, deeply rooted in business realities.
To go further
These questions, and many others, are at the heart of the opinion piece by Philippe Audibert, CTO of DATASOLUTION. It’s an uncompromising, but above all pragmatic, analysis aimed at transforming AI into an operational lever for securing digital projects, improving software quality, and creating sustainable value.
