I just wrapped a month‑long, in-person AI clinic, and I’ll be honest: I left both energized and overwhelmed.
As an industrial engineer focused on multi‑unit foodservice design, I’ve spent years optimizing flow, labor, and space. With all the hype around AI, I wanted to understand it better. I know our process, and in turn our designs, could be improved or accelerated with the right tools. I still don’t know exactly how. My takeaway: AI won’t replace the fundamentals, but it could change how quickly we can get to a scientific answer.
Where can AI make a difference? In short: automating the repetitive parts of our analysis and doing some reasoning as it goes. Our work starts with observation. We capture video and go back to the office to calculate the time it takes to execute each task and study the overall process of the culinary team. Computer vision can help with these calculations by automating time studies from video footage or tracking a car through the drive‑thru end‑to‑end to calculate service time. That’s tedious work for engineers and represents a natural opportunity for AI to have an impact.
Demand planning represents another place AI could help. A straightforward regression analysis of historical sales demand is something we calculate as part of our prototype design process, but an AI application could do that too with some basic instruction.
Upon determining demand, we translate that into capacity, which helps shape equipment and storage needs as well as line sizing. This is where small, purpose‑built apps could help, and other tools that create full blow applications with little or no programming knowledge, ready for implementation. Imagine a simple portal to manage equipment specs and menu items, a dashboard that turns assumptions into capacity constraints, and a “what‑if” sandbox to test menu or station changes. Instead of spreadsheets and scattered PDFs, you have a living model the whole team can use.
Layout generation may represent the biggest impact here. Instead of sketching a handful of options and building one, AI could produce dozens of layouts tied to KPIs such as transaction targets, menu mix, code compliance, and then auto‑simulate each to stress‑test throughput and labor efficiency. While that promise is intoxicating, here’s a sobering dose of reality: none of the tools we reviewed come close to that end‑to‑end promise. It sounds great, but right now it’s more pipe dream than an AI product or set of AI products combined.
We already create multiple layouts as part of a rigorous design process, and we build discrete‑event simulations to test them but asking a general‑purpose AI tool to read a plan, infer all the constraints, and wire it seamlessly into a simulation is not realistic. The link between a layout and a simulation is full of assumptions about recipes, batch sizes, crew choreography, equipment sizer and capacity, utilities, and queue logic. AI won’t intuit that without guardrails and structure.
What did spark hope for me was the idea of AI agents, specialized digital co‑workers that each do one job well, hand work off, and stay inside boundaries we define. One agent could draft multiple layout or station variations from parametric templates and set objectives not from thin air, but from a controlled library. Another could run compliance checks against health, fire, ADA, and ventilation rules, flagging violations as you iterate. A third could interact with discrete event simulations, swapping in queue strategies or crew counts and comparing time per order, walking distances, and employee utilization. A fourth could assemble a costed package, equipment, utilities, and construction costs.
To be clear: no product on the market today performs these end‑to‑end workflows out of the box or can even come close. But the clinic opened my head to the possibility of stitching capabilities together in a way that’s useful and realistic for design in a not so distant future.
So, I’ve told you about what AI can’t do – at least not yet. How about a brief discussion on what it can do as of now. It can generate a semi‑automated loop. Use computer vision to accelerate time studies and drive‑thru tracking. Use forecasting AI agents to automate design target calculation. Design apps with AI can translate sales and checks inputs into equipment capacity requirements. Keep layout generation human‑led but parametric, blocks with rules for clearances, adjacencies, and utilities, so exploring many options is accelerated. Connect those options to a standard simulation template where only a few inputs change (menu mix, crew count, routing) to get layout performance metrics.
What won’t change is the basic understanding of what makes a new restaurant become more efficient: studying the restaurant operation, understanding the process and the flow of things, linking items to equipment usage and putting all that together in a layout that optimizes efficiency. What will change is the speed and tools with which we turn those observations into better spaces, better flows, and more economical builds. It’s the industrial engineers dream, do more with less. If AI helps us get there faster, and with fewer cycles spent on grunt work, that’s worth the time we spend learning it.
