During a webinar, experts explained how to develop a mobile app in 3 weeks (from iteration to code, including specifications and mockups). This was achieved using an agentic factory. Discover the key moments you shouldn't miss!
In the age of AI agents, is it possible to build a mobile application in just three weeks? The suspense is short-lived: yes. During a webinar organized by Smile, our experts revealed how they met this challenge to present a working prototype at VivaTech 2026. The team took the opportunity to explain what an agent factory is and why it's the essential foundation for your tech projects. They also discussed the future of roles such as developer, system architect, and product owner.
What is an agentic factory? How can it help organizations design a mobile application in record time?
First, the agentic factory is a value chain of five specialized AI skills—specification, design, development, testing, and security—that operates in continuous iteration and is improved as needed based on project requirements and company ambitions. The human element remains central, as it is a real person who configures the various skills, adapts them to the project, improves them, and validates them.
Next, it's important to clarify that while the agent-based factory is a basic concept, the development method chosen to carry out the project is crucial. For Fred Bovy, Engineering Manager at Neopixl , it's common to use vibe coding to quickly obtain results such as a prototype or a demo to present at a trade show. However, vibe coding doesn't allow for managing technical debt or considering the product's long-term viability. “At some point, we'll need to evolve. We'll go through a phase of assisted coding,” explains Fred Bovy. “The developer will be supported by an agent. The ultimate source of truth will remain the source code, which will be understood by the AI and allow the developer to produce better and faster results.”
For the challenge SMILE set itself – creating a mobile application in just three weeks – the Spec-Driven Development (SDD) methodology was chosen. “Everything is based on structured specifications, which requires more rigor, but is ideal for industrialization and achieving real gains with AI,” explains Fred Bovy.
A use case combining innovation and personalization
To validate the approach, the team designed an immersive application for Speed Vegas, a company offering track driving sessions in the United States. In three weeks, the team managed the entire product lifecycle: ideation, specifications, design system, mockups, and code, explains Deborah Delin, Digital Experience Consultant at Smile.
On the menu: personalization to make the experience immersive, facial scanning, and augmented reality.
“The experience begins two days before arriving at the track. On a premium homepage, users can finalize their booking and find the details. We've added upsell options with available extras. There's a customization section: users put their name on their racing suit, scan their head to determine their helmet size (which they can then personalize by adding stickers), place it on a table, and view it from different angles. After confirming their preferences, users receive safety information, including best practices and reminders of proper track etiquette. Once the checklist is complete, everything is ready to go to the track. We also offer a post-race experience: users can access a 'collection' section where their cards recount their various experiences during the race,” explains Deborah Delin.
The 5 lessons to be learned from this challenge:
- Data quality remains the essential prerequisite.
- A systematic human review is necessary during the design phase to eliminate inconsistencies.
- Each generated line of code must be checked.
- The management structure must be very well organized to guide the actions of the agents.
- The multiplier effect: the higher the quality of the input elements, the better the result. Don't launch an agent factory on a poorly defined project.
The impact of AI on tech jobs
Is AI replacing humans? Experience shows the opposite: human expertise is becoming even more critical.
Less than 1% of the code was handwritten by the developers. Their role was to correct the code chain and instruct the agents to prevent errors from recurring. With each iteration, the overall quality of the development improved.
“The role of the architect is changing, just like that of the developer, the QA engineer, or the Product Owner. The architect no longer just designs the application system: they design the chain that will produce the application. Experts are being promoted, not replaced,” concludes David Renoux, Director of Digital Experience at Smile.
To watch the webinar replay, go here !
