Over the past year, our team has built and deployed multiple generative AI-powered applications for the insurance industry: real products used by underwriters and analysts every day.
To reflect on what that journey has actually been like, we sat down as a team for an internal panel. No slides, no sales pitch. Just developers, architects, and analysts talking honestly about where generative AI shines and where it gets complicated.
Participants:
Danilo Martins
Solution Architect and Tech Lead
Sebastian Acevedo
Generative AI Lead / Backend Architect
José Muñoz
Scrum Master & Project
Manager
Sundeep Ahuja
Senior Analytics & Financial Services Expert
Matias Munchmeyer
Associate Director (Moderator)
“What Excites You Most About Using AI in Insurance?”
Sebastian Acevedo (Generative AI Lead):
“For me, insurance has always felt very static. Slow. Tedious. What’s exciting is breaking that. Things that used to take days or weeks can now happen in minutes. That’s a real cultural shift.”
He pointed out that AI doesn’t just speed things up; it changes behavior. Instead of reacting to client inputs, teams can become proactive: assessing risk earlier and starting conversations rather than waiting for them.
Jose Munoz (Scrum Master):
“What excites me is being the bridge. Seeing technical capabilities translate into something that actually improves how people do their jobs. Answering a hundred questions in minutes instead of days, that’s powerful.”
Danilo Martins (Solution Architect):
“Generative AI is exponential. It multiplies everything: productivity, ideas, output. And the interesting thing is that we now use AI to build AI. That feedback loop didn’t exist before.”
“And What About Generative AI Feels Daunting?”
Sebastian:
“The speed. Honestly. What we built at the start of the project already feels historical. New techniques appear every month. Sometimes every week.”
Danilo:
“That’s why grounding matters. It’s very easy to be seduced by speed and volume, but generative AI is still bad at very broad, complex tasks. Objectives must be clear, or you end up with a lot of output and no direction.”
He shared an example: even top models struggled to replicate a decades‑old C compiler. “It’s a reminder that AI is impressive but not magic.”
“What Challenges with Generative AI Do Business Leaders Often Underestimate?”
Jose:
“Data governance. Almost always. We deal with missing fields, inconsistent formats, regional differences, and unclear data ownership. And when you add AI to that mix—especially in a regulated industry—you must be precise.”
He emphasized expectation management: aligning what business users want with what data can realistically support.
Sebastian:
“And it’s not just databases, documents too. PDFs, scanned files, tables, images… AI needs structure. If the inputs are messy, the output is half‑baked.”
Sundeep Ahuja (Analytics):
“Data quality is still the biggest barrier. And then comes explainability. Generative AI is great operationally, but when you use generative AI outputs in models or decisions, people need to understand why.”
“Perfect Architecture vs. Weekly Delivery. How Do You Balance That?”
Danilo:
“Every shortcut we take for speed creates technical debt. There’s no way around that. The question isn’t whether debt exists, it’s whether we manage it.”
He gave a concrete example: choosing not to build OCR for rare, scanned documents due to time and budget constraints. “That’s a conscious trade‑off, not negligence."
He also highlighted how fast the ecosystem evolves. Features that once required custom engineering are now native to platforms. “Yesterday’s debt sometimes disappears on its own.”
“What Generative AI Innovations Are You Most Excited About Next?”
Jose:
“Tools that give us time back. If AI helps with development, testing, or documentation, we can focus more on communication and quality. But the trick is choosing what actually adds value, not just what’s new.”
He noted the unique opportunity of working in insurance: “It’s conservative, but data‑driven. That makes it ideal for meaningful modernization.”
Sebastian (on AI‑assisted coding):
“I use it constantly. I am the architect, and AI fills in the specifics. I don’t need to memorize every library anymore, and I can focus on structure and intent.”
“What’s the One Message for Business Leaders About Generative AI?”
Danilo:
“Be grounded and still move forward. The noise around generative AI is enormous, but this is a real revolution. Leaders need focus, not fear.”
Jose:
“Trust specialists. This space moves too fast to track on its own. Our role is turning cutting‑edge technology into something usable and safe for real businesses.”
Sebastian:
“The decision matters. Either you evolve with generative AI, or you stick to old ways. And none of it works without good data.”
Sundeep:
“Strategy first. Before telling developers what to build, leaders need clarity on direction: which problems matter, which use cases are viable, and how risk and regulation fit in.”
Final Thoughts
AI moves fast. Organizations don’t.
That tension showed up in nearly every part of the conversation. Technology is advancing at a pace that makes even recent work feel outdated, while businesses must still operate within real constraints: legacy data, regulatory requirements, and the need for reliable, explainable outcomes.
Bridging that gap isn’t about chasing every new model; it’s about grounded leadership, strong data foundations, and teams that know how to translate complexity into something real.
These teams aren’t chasing the latest trend. They’re translating fast-evolving capabilities into tools that work in production, fit into real workflows, handle imperfect data, and deliver consistent value.
Because in the end, AI transformation doesn’t happen in breakthroughs alone. It happens in the day-to-day work of aligning technology with business reality: one decision, one use case, and one iteration at a time.
And that work happens not in demos, but in conversations like this one.
Click here for another conversation we had with insurance underwriters about generative AI
