With new AI models released almost every week, companies struggle to turn their investments into measurable profitability gains. Solutions.
The frenetic pace of innovation in artificial intelligence constitutes a major paradox for companies. New models are released almost every week, with an estimated useful life of between 12 and 18 months.
Let’s take the concrete example of the market leaders in 2026. At Anthropic, since the start of the year, there have already been several major releases: Claude Opus 4.6 (February), Claude Sonnet 4.6 (mid-February), Opus 4.7 (April), Opus 4.8 (end of May) and Claude Fable 5 / Mythos 5 (beginning of June). At Mistral, we observe the same sustained pace with Mistral Small 4 (March), Voxtral TTS and other multimodal and agentic variants in the first half.
Even if a company were to immediately adopt the new state-of-the-art model, the time required to configure processes, validate results, train teams, integrate the system into the existing architecture, and deploy at scale is easily 6 to 12 months. There would then only be around 6 months of useful life for the model before a new, more efficient version would make it obsolete. This infernal cycle leaves very little time to make the investment profitable.
The productivity paradox applies to AI
This observation relates directly to Solow’s paradox formulated in 1987: “We see computers everywhere, except in productivity statistics”. Nearly forty years later, AI reproduces the same pattern. It is omnipresent in speeches, trade shows, annual reports and marketing demonstrations. However, tangible gains in productivity and especially profitability remain modest or take a long time to materialize in the majority of traditional sectors.
To this first paradox is added that of Jevons. An improvement in efficiency (more efficient models, less costly for inference) ultimately leads to greater overall consumption of resources (computing, energy, data), because it makes AI accessible to a multitude of new uses. Companies are multiplying experiments, agents and analyses, causing costs to explode without net profitability following suit.
The necessary complementarity IA + H
AI alone does not create lasting value. Its true power lies in its complementarity with humans. Models generate often impressive outputs, but they always require business judgment, validation, contextualization and final accountability by people. Real profitability only emerges from a mature hybrid AI + Human process. However, this process requires time: team learning time, industrial running-in time (feedback loops, measurement of mixed KPIs) and integration time into existing systems.
A renewal rate that is too high prevents this maturation. Projects often never reach industrial scale: by the time the company configures, validates and deploys, a new major version is already available, making the previous work partially obsolete. Companies then remain stuck in a logic of perpetual Proofs of Concept or partial deployments.
The real reasons for adopting AI
Why are companies massively adopting AI despite this double paradox? Several reasons, both rational and emotional, explain this behavior:
- Strong intuition: leaders clearly perceive that AI has enormous potential to create value. The illusion of the prototype (which seems perfect in demonstration) maintains this conviction.
- The quest for a new source of economic growth: in a context of slowing overall productivity, AI represents a major hope for a rebound.
- Inspiring extraordinary cases: Elon Musk and other leaders have demonstrated the creation of very high quality and very large scale processes using AI, sparking a ripple effect.
- Accessible quick-wins: each company can quickly experience significant gains (automation of repetitive tasks, content generation, coding assistance) which offer rapid and visible returns.
- The novelty effect and FOMO (fear of missing out): we perceive the potential before even having measured it.
- Competitive pressure and investor expectations, which demand an AI-first posture.
- The price of licenses.
- The relative ease of launching low-risk use cases (copilots, content generation, internal chatbots).
The urgent need for AI governance in business
Faced with this observation, AI governance becomes a decisive strategic issue. The companies that will succeed in creating real added value will be those that know how to:
- Adopt a “process-first” rather than “model-first” approach: stabilize a mature hybrid process before moving on to the next version.
- Implement rigorous value governance: measurement of business KPIs (productivity, reduced risk, customer satisfaction, improved margin) and not just technical ones.
- Invest massively in human factors: continuing training, clear definition of responsibilities, continuous improvement loops.
- Design modular architectures that facilitate transitions without rebuilding everything.
The double Solow-Jevons paradox is not inevitable. It simply reveals that AI is not a magic bullet, but a technology for organizational transformation. Its frantic pace imposes a new discipline on managers: knowing how to voluntarily slow down on certain critical processes to better accelerate on the creation of sustainable value. AI will truly be everywhere the day it clearly appears in companies’ operational results and income statements.
