Podcast Summary: Just Now Possible – "Building AI Employees for Hospitality: How AITropos Takes Orders Where Customers Already Are"
Host: Teresa Torres
Guests: Santi Marciori (CEO, AITropos), Juan "Juanu" Aedo (CTO, AITropos)
Published: April 30, 2026
Overview
In this episode, Teresa Torres speaks with Santi Marciori and Juan Aedo of AITropos, a company building AI-powered employees for the hospitality industry. The conversation dives deep into how they identified their core problem—streamlining order-taking for restaurants and hotels—leveraged AI agents and tooling, overcame technical roadblocks, and validated their product in the market. They share candid stories about their innovation process, technical strategies, and what it takes to create AI systems that deliver real operational value where customers already are: messaging apps.
Key Discussion Points and Insights
1. Why Hospitality? Origins and Problem Selection
[01:19, 08:41]
- AITropos was created to deliver tangible operational improvements in hospitality, focusing not on replacing humans outright, but on enhancing efficiency where the human touch is less critical.
- Both Santi and Juan have deep roots in hospitality tech (hotel/restaurant systems, POS, PMS), seeing firsthand where businesses struggle—especially in quick-service restaurants (QSR) and places where operational efficiency is critical, not the "human touch".
- Their initial explorations spanned hundreds of ideas before honing in on order-taking as the high-impact, unsolved problem.
Quote:
"We are actually building AI employees for the hospitality industry, and we're trying to generate real operational impact. ...It's not just a bot, not just a chatbot, but it has a lot of tools and many integrations."
— Santi Marciori [01:19]
2. The Human–AI Balance in Hospitality
[03:08, 04:14]
- Clear separation between where AI employees make sense (e.g., fast food, poolside orders, routine service) versus where human interaction is desired (luxury hospitality, complex service).
- Not about total replacement, but filling service gaps where humans aren’t needed or available.
- Realized success depends on being able to blend AI service with seamless hand-off to human staff when necessary.
Quote:
"We're not replacing where a human should be, but where the technology is in the middle, in between humans."
— Juan Aedo [04:14]
3. Zeroing in on Order-Taking as the Core Use Case
[07:10, 11:17]
- After much experimentation, identified order-taking as both the hardest and most valuable feature to nail—especially for QSRs and restaurants.
- Story recounted of building a hardware device and waiter assistant app before pivots led them to customer-facing order bots.
- Order-taking involves messy, non-deterministic human input that needs to be mapped onto structured POS requirements reliably and fast.
Quote:
"We started offering this solution to different restaurants...they started asking for this service to be deployed for customers directly. That's how we found out there was huge potential."
— Santi Marciori [12:38]
4. AI From Day One & the Developer Experience
[13:17–14:14]
- AI (LLMs, agents) was core from the outset—never considered a non-AI approach.
- The team is "addicted" to AI-powered workflows (coding, prototyping), so much so that LLM outages leave them feeling lost.
- Acknowledges the dopamine hit of rapid AI prototyping, stressing the difference between “demo-ware” and robust production systems.
Quote:
"I can't remember a time when we didn't use AI anymore. I don't even want to think about that. Gives me the chills."
— Santi Marciori [13:17]
5. The Challenge: From Messy Conversations to Structured Orders
[17:02, 18:12, 19:36]
- Technical challenge was not integrations, but turning unpredictable customer language into correctly structured orders every time.
- Real-time performance is critical; agents must respond quickly and naturally, neither too fast (robotic) nor too slow (frustrating).
- Built for WhatsApp first (Latin America’s “operating system”), but platform-agnostic by design.
Quote:
"The hardest part is being able to translate the non-deterministic world of human conversations and LLMs into structured information so you can feed that to systems."
— Santi Marciori [17:02]
6. Architecture: Iterations and Agent+Tools Paradigm
[21:24, 25:14, 38:55–45:23]
- Iterations:
- Started with waiter-focused hardware → app for waiters → customer-facing WhatsApp agent.
- Key architectural decisions included offloading as much as possible to AI agents equipped with specialized tools.
- Tools & Orchestration:
- The agent accesses a suite of custom tools (add product, check stock, payment link generation, geolocation, knowledge base search).
- Prompts are dynamically constructed, including both static (system-level) and “immediate” context based on conversation state and user input.
- Aggressive use of parallel processing, caching, and fast databases to meet speed requirements.
Notable Implementation:
- Injecting pre-processed context/knowledge into the agent’s system prompt for faster, more accurate tool invocation (anticipates likely next step instead of waiting for agent to react).
Quote:
"We have a prompt composer framework that we implemented which injects fragments depending on configuration...we even do MVPs for features in the prompt, then build the features if they're needed."
— Juan Aedo [45:23, 47:50]
7. User Workflow: Seamless Conversation-Driven Order Experience
[31:16, 35:58, 36:25, 38:13]
- Customer simply opens WhatsApp, messages the restaurant's contact, and interacts conversationally.
- No app download required—the agent can recommend, answer questions, suggest pairings, handle customizations, and send images or PDFs of menus.
- All ordering (including payment) happens within—or just outside (for payment)—the messaging app.
Quote:
"We just have a technology that basically converts the audio into text. But the idea is giving the full conversational experience of DMing through WhatsApp that you have with your friends."
— Juan Aedo [38:13]
8. Evaluating, Testing, and Human-in-the-Loop Quality Control
[57:55, 58:55, 59:56, 63:10]
- The core KPI: Percentage of order items interpreted correctly. This trumps all other metrics.
- Pre-deployment, thousands of test interactions are auto-generated (with customer-simulating agents), checked for accuracy by another agent, then reviewed by team/freelancers.
- In early onboarding, a human actively monitors/audits all customer conversations, gradually handing off to automation as the agent reaches reliable performance.
- Even error correction is seamless—if an agent makes a mistake, the human team can "take over" the chat; future errors are fed back into system improvements.
Quote:
"Whenever we find things that are off, we take over and we correct them by hand...Once it happened that we didn't catch the error...Since the customer gets a confirmation ticket, he saw the address was not right and called the restaurant—it was quite an easy fix."
— Santi Marciori [61:10]
9. Lessons in Conversational-Driven Product Development
[49:00, 53:00, 54:11]
- The team practices "conversational-driven design"—mapping out product features and flows based on the way human conversations unfold.
- Emphasis on quick learning, experimentation, and applying lessons from a spectrum of AI and software disciplines as new options (e.g., new LLMs, context strategies) emerge.
- They condense business logic, natural conversation, and the quirks of every restaurant into agent orchestration.
Quote:
"Our goal is to deliver a high quality service for guests and restaurant customers. And we actually do that, of course with a lot of AI. But sometimes we need humans to jump in."
— Santi Marciori [05:02]
10. Next Steps: Scaling to Global Markets
[66:24]
- Having proven product-market fit in Argentina, AITropos aims to scale into Mexico, the US, and Spain, with a focus on decreasing onboarding times and automating more of the setup.
- The aspiration: Be the go-to AI-driven order layer for any hospitality business worldwide.
Quote:
"Our goal is to scale it. We want restaurants all over the world to be able to use this tool."
— Santi Marciori [66:24]
Memorable Moments and Quotes
- On the dopamine rush (and risk) of prototyping with AI:
"One of the best things about AI is it gives you some very quick dopamine hits. Because making a prototype, it's awesome. But that is good and bad at the same time...you have no idea what you are getting into." — Santi Marciori [22:41]
- On the balance of technical optimism and realism:
"We were super optimistic about this...but I honestly had some doubts along the process. We spent a few months working on it and still had an unacceptable error rate because we wanted to make this perfect." — Santi Marciori [22:57]
- On the evolution of product as AI models improve:
"What’s fun about this space is...you build something and then time passes. And even if you do nothing, your product gets better." — Teresa Torres [29:37]
- On reducing onboarding times:
"It used to be three months. And as we improve this, the time for onboarding is reducing, which is basically our biggest challenge." — Juan Aedo [64:29]
Timestamps for Key Segments
- [01:19] — What AITropos does and the meaning of "AI employees"
- [03:08, 04:14] — Human/AI balance in hospitality and the philosophy behind their solution
- [07:10, 11:17] — Deciding to focus on the order-taking problem
- [13:17–14:14] — Early use of AI and how it's woven into their workflow
- [17:02] — Technical challenge: mapping conversation to structured orders
- [21:24, 25:14] — Product iterations from waiter assistant to customer-facing WhatsApp bots
- [31:16, 35:58] — User experience: ordering food via chat and the full flow
- [38:55–45:23] — Under the hood: agents, tools, orchestration, prompt hacking
- [57:55, 58:55, 59:56, 63:10] — Evaluating agent performance and quality control processes
- [66:24] — Next steps and scaling ambitions
Conclusion
This episode is a masterclass in identifying real-world problems, iterating toward true product-market fit, and building robust AI systems that operate reliably in the wild. Juan and Santi share deep technical and strategic insights—grounded by war stories and the realities of serving both humans and businesses in a high-stakes environment. If you're building conversational AI or interested in the intersection of AI and operational workflows, this episode is packed with actionable lessons.
Listen to the full episode for even deeper technical details and candid founder stories!