The generative AI redefines our interfaces. With the MCP protocol, it becomes interoperable and could replace the smartphone as the main entry point in our digital uses.
Is it still just something to write about artificial intelligence?
From the understanding of LLM technology to the profound transformations of the many tasks in the service trades, from the speed race led by the GAFA to our questions about the challenges of European sovereignty, from the first impacts on the labor market to the mediatized cases of companies turning away after being proclaimed “IA First”: many has been said.
Depending on our profession or our appetite for technology, everyone observes the impact of generation models (language, image, videos, computer code) and gradually transforms its uses.
Three years of product development, uses and an economic model to consolidate
Since the launch of Chatgpt in November 2022, a product development phase has succeeded a technical development phase.
At first, it is the increase in the quality of the models, the increase in the volume of training data and the enrichment of external sources (web navigation in particular) which allowed our prompt always imprecise to generate ever more qualitative outings, and in new forms (images, texts, videos).
The year 2025 is that of the year of products. On the foundations of these models, the main world technological champions have declined these incredible advances to simplify their uses and diversify their channels of use. Other companies, often a lower size, have come to lean on these models to hyperspecialize them (the sale AI, the AI of customer service, the AI of IT developers, etc.) or create technological bricks allowing you to optimize them for a precise task (optimization of the RAG in particular).
The issue is now that of the economic model: in 2024, Forbes pointed out that each dollar won by Openai cost him $ 2.35. Google or Meta, more established companies, finance these activities on the important profits carried out by their other services. The real cost of the productivity gain obtained by the increase in production in the service professions is therefore not yet known, which creates risks as to the ability to retain existing customers.
In short, will you be, or your business, always ready to pay for content production assistance if the price of this subscription (which constitutes more than 75% of operating income) double or triple?
From generation to action, the next LLM border:
The next step is that of the interoperability of AI models between them and between other computer products that we use daily to go from research to action. Today, an AI can help you choose a restaurant. Tomorrow, she could reserve this restaurant for you, or make a bank transfer, and come back to you when information is missing to accomplish his task.
A critical step in this capacity of the models to connect with each other and to other systems was crossed on April 22, Google, Openai and Anthropic agreeing on a common protocol of exchanges and interactions between artificial intelligences: the MCP.
Behind this decision of technical appearance, it is the ability of these models to navigate with all of our digital tools which is now open, allowing language models to become the gateway to our services, where the Internet /Mobile browser were so far the essential interfaces of our digital lives. Until today, the challenge for a company was to be well referenced on Google, or to see its application available on the App Store. Tomorrow, a new issue should be the ability of your service to be accessible by the most used language model (s), through the MCP protocol.
From computer to smartphone, what tool for a new generation of interfaces?
To find a site on the Internet, the browser accessible from a computer remained the most used medium until 2013. At the end of 2016, this trend is reversed worldwide, more than 50% of internet traffic is now driven by smartphones, including 90% by mobile applications.
Generative artificial intelligences have already started to take market share to what even little was a breathtaking fortress, namely Google Search. But the development of the Context Protocol model as a standard of connection between digital services could allow the arrival of new products which would become the input doors of our daily interactions.
Several start-ups have already taken the plunge by offering “Wearables AI”, which aims to replace your smartphone. Their adoption is for the moment a failure but many analyze OpenAi’s decision to buy “IO”, the start-up of the ex-Apple Jony IVE star designer as the confirmation that this trend is durable and could allow LLM to strongly anchor their technologies in our daily lives, once the judicial setbacks of the Californian ecosystem passed …
Replace the smartphone or enrich it, a technical challenge and product:
What forms could these new miniature tools that play the essential role played by our smartphones? The most likely track today is that of coexistence more than replacement.
The calculation capacities necessary for treatments by AI are indeed important. To avoid too strong latency and therefore an inconvenience for the user, part of these calculations must therefore take place directly in the product. This is for example the case of improved Siri versions in the iPhone. How to reconcile the size of the device and calculation power to allow new experiences of consumption of digital services? What consequences on the development of IT solutions, in particular on user interfaces? This is undoubtedly the questions that will occupy ecosystems in the coming years.
