At the start of the year, Larroudé’s operating systems were not connecting the way Ricardo Larroudé wanted them to.
The problem was that they were not speaking to each other cleanly. The footwear company’s Shopify e-commerce site, factory data, inventory information, site code and marketing performance data all held useful information, but connecting it was too manual a process. CEO Ricardo Larroudé’s goal was to give the company a clearer view of what is selling, what needs to be made, what materials are available and where manual work can be reduced.
To achieve this, he used AI coding tools to personally write, test and refine a system centered on automated connectivity.
“I wasn’t a coder at all,” Larroudé said. “I understood what that world was, and I learned quickly. But you do need someone who understands the process, or people who understand how [software governance] is put into place. There’s a proposition of law, which goes to a vote. You need to create those technical checks inside.” By “checks,” Larroudé was referring to a review process that tests proposed code and workflow changes before they go live, ensuring the system does not break the website, expose data or make incorrect decisions.
“Now, for every sale on the website, I can link it back to the little button that I bought that went on that specific shoe,” for example, Larroudé said. Larroudé, which does around $100 million in annual sales, is large enough to have complex inventory, factory and website data, but still small enough to allow its CEO to overhaul the tech stack quickly. Larroudé said the project began a few months ago and that the main data-ingestion process, in which the AI pulled the company’s systems into a single, connected structure, took about eight weeks.
For a footwear company, a high level of visibility into its supply chain, production and operations can make a big difference. A shoe is made up of leather, hardware, soles, colors, sizes and factory timing. For example, if one system in the business calls a leather color “199” and Shopify calls it “pink,” the Larroudé AI system is being trained to understand that those records refer to the same product.
The system is built on tools the company already uses or can control. Shopify provides the e-commerce layer. GitHub houses the code for the operating system Larroudé has been building. Google Cloud provides computing power, including Cloud Run, which can scale up when needed. BigQuery is used to process large amounts of data. Larroudé said he also uses Claude, ChatGPT and Codex for different parts of the work, from testing ideas to checking and writing code.
The first clear test for this new AI system came when the brand’s website conversion rate dropped earlier this year. The system identified old Shopify apps, scripts and checkout code that were slowing the site down. Larroudé then used AI coding tools to generate fixes, review them and push them live. After cleaning up the code, he said conversion jumped from 0.3% to 1.7% within 45 minutes. Checkout completion, he said, improved from 25% to 75%.
The setup has not been cheap. Larroudé said he initially spent “dozens of thousands” of dollars on token costs, meaning the fees paid to run large numbers of AI queries, as the system absorbed company data. During the heavier build phase, he said costs reached around $20,000 a month while running through BigQuery quotas. He expects ongoing costs to come down to “a few thousand” dollars a month.
Still, Larroudé said that is less than what he was spending on programmers. He previously had a team of 10 people and was spending “dozens of thousands of dollars” on programming work, including one person whose job involved tagging pages and banners on the website. Other tasks for the programmers included cleaning up old site code, reconciling product data across systems, and manually preparing inventory and planning reports. The new system has changed how he thinks about hiring.
“I reduced staff tremendously, and I only kept the ones that can work agentically,” Larroudé said. “I don’t like using the word AI. It’s like luxury for me. I don’t like that word. I like the word agentic, which means dynamic software. I’m keeping the people that can transition to working with dynamic software.”
By “agentic,” Larroudé means software that can take an instruction, complete a task and improve a workflow, with human review before major changes go live. He said the company still needs technical people, but it needs people who can work with the system, check its output and understand the controls around it.
“It’s not like we’re substituting people,” Larroudé said. “We will just be able to free up people’s time. They still have to be in front of the computer. They’ll have to be inside the metaphorical ‘AI car.’ If the car is going to hit a post, they’re going to have to pull the parking brake.”
The next use case for the company is inventory. Larroudé said the new system can help the company understand what materials it needs to buy, what inventory it already has and what it can make quickly. If the brand has enough black leather in stock, for example, the system can factor that into production planning across multiple shoe styles, instead of treating each product separately.
That could shorten the time between demand and production. Instead of waiting for a full planning cycle, Larroudé said the system can help the team identify which products can be replenished quickly using materials already in stock, reducing overproduction while keeping bestsellers available.
For Larroudé, the lesson is not that every fashion CEO needs to become an engineer. It is that leaders need to understand how their company’s systems, data and decisions fit together. “Every CEO needs to be programming right now, not coding,” Larroudé said. “They need to be programming their company.”
