AI Agents in Production: Takeaways from the LangChain Meetup in Mexico City

Direct answer

The bottleneck in moving an AI agent from a local script to real production is almost never the model — it’s the architecture around it. That was the central takeaway from a recent LangChain meetup in Mexico City, “Agents in Production,” where two sessions — one on designing agents, the other on operating them at banking scale — converged on the same point from different angles.

What was covered at the meetup

The event, organized by Adrián Sánchez Castro and the local LangChain community, featured two main talks:

• Carlos Alberto Haro (Lead AI Engineer) presented “Fantastic Agents and Where to Find Them,” a practical framework for designing intelligent agents around three pillars: knowing (understanding context and available data), doing (executing concrete tasks), and deciding (choosing between alternatives with judgment, not just fixed rules).
• Alejandro Zarate (Chief Data and AI Officer at Santander Mexico) broke down what it actually takes to operate agentic systems at banking scale: the underlying architecture and the real lessons of shipping them to production in a regulated environment, where a mistake isn’t just a bug — it’s a compliance problem.

The question left on the table

The meetup closed with a direct question: how ready is your current tech stack to move AI agents from local scripts to full-scale production? It isn’t rhetorical. In practice, that transition tends to trip on three things a test script never needed: structured data the agent can rely on without guessing, error handling that doesn’t leave the agent improvising silently, and a way to measure what the agent is doing once you’re no longer watching it run in your terminal.

Why it matters for the ecosystem in Mexico

Beyond the talks, what stood out in the networking session afterward was a community that prioritizes execution and shared learning over hype: people building agents in production, not just talking about them. It’s a sign the AI conversation in Mexico is maturing from “what’s possible” to “what holds up with real users and a real budget behind it.”

Source: this recap is based on the original LinkedIn post about the event — see post.