The next wave of AI doesn’t recommend. It acts. And for an industry built on compliance, player protection and real-time decisions, that changes everything.
In the wider tech industry, agentic AI is moving fast. Inside gambling, the mood has been more cautious. Fincore CPO Dominic Le Garsmeur tells iGaming Expert why he thinks that is about to change.
The wider tech industry is moving fast on agentic AI and MCPs (model cost protocols). The gambling industry looks like it’s significantly behind that curve. Why is that, and does it matter?
I’d push back on the framing slightly. We’re not behind on AI. We’re one of the more mature industries on predictive AI. Churn models, risk scoring, real-time responsible gambling monitoring: in various forms, operators have been doing this for years. Where we’re behind is agentic AI. Making the leap to autonomous platforms.
There are three reasons for that gap. The obvious one is regulation and the need to get responsible gaming 100% right. An autonomous action that touches a player’s account or gives them a bonus they shouldn’t be getting has compliance consequences that a wrong answer from a chatbot simply doesn’t. The second is architectural. Most platforms are monoliths where intelligence and execution are tightly coupled, there’s nothing clean for an agent to plug into.
The real danger is mistaking justified caution about autonomy for permission to ignore the architecture. Those are not the same thing.
When you look at how most gaming platforms are architected today, what do you see?
Mostly monoliths. Or distributed monoliths pretending to be a services-based architecture. When new intelligence services are added, there’s often a messy coupling with the operator’s platform. People are trying these new services out, but they’re not doing it in a way that gives them a clean separation between intelligence and execution. In the long run, you need the right architecture to have the ability to switch models, switch suppliers. Most platforms don’t give you that.
How has your thinking about platform architecture changed over the last two or three years?
We decided to stop competing on intelligence and start competing on execution. Our internal shorthand is: models will keep improving, we want to be the operating system they execute through. What drove it was watching MCP go from a niche protocol to a vendor-neutral standard adopted across industries. That told us the integration problem was being solved and the frameworks to support it were maturing.
If an AI agent were trying to operate across a typical operator’s stack today, PAM, bonus, payments, RGS, what would it actually encounter?
Rough borders between systems. Different authentication, different data shapes, no shared notion of ‘the player’ across systems. It would find plenty it could read and very little it could safely write. No audit trail. No guardrail layer built to enforce system rules.
In practice, it would behave like a very capable intern with no badge access, no idea who to ask, and just enough reach to accidentally breach a deposit limit. That’s the thing people get backwards: the agent isn’t the hard part. The environment is.
Genuinely agent-ready, what does that actually mean?
Three things: a clean data contract so the agent can see what’s true, a safe execution interface so it can act without bypassing compliance, and observability so you can prove what it did and why. The technology is maybe 40% of it. The other 60% is organisational. Who owns the agents? Who’s responsible for what they can do? Most of the genuinely hard questions aren’t engineering questions at all.
Where does MCP sit in Fincore’s roadmap?
At the centre of it. Our collaboration with Sportradar, where Fincore was selected as the execution layer for their VAIX AI player recommendation engine, pushed us to start thinking about this whole area early. We’re now rolling out MCP interfaces across our product lines, starting with read-only services so we and our customers can get comfortable with the technology.
The biggest part of the roadmap isn’t building the interface, that’s becoming part of the day job now. The biggest part right now is building the supporting infrastructure: the guardrail platform, the observability layer. It’s not the sexy part. But getting this right gives us a foundation for a long time to come, not just a demo.
What are operators actually asking about, and what should they be asking that they’re not?
They ask what your model can do. What they don’t ask is: what happens when the model improves? What happens when there’s a better one? If your architecture forces a rebuild every time the brain gets better, you’ve built the wrong thing and signed up to a maintenance bill that compounds.
What surprised you most about building for an agentic future?
A year ago it was all about the models. Benchmarking the latest from OpenAI, Anthropic. Models are powerful now, even complex iGaming tasks don’t need frontier models. What actually matters is well-structured data contracts and well-designed interfaces. Most models can do a great job as long as you give them the right tools. The model you use doesn’t matter that much. The tools do.
Is the industry’s caution justified, or is it becoming an excuse for inaction?
It’s both. And the discipline is being honest with yourself about which one you’re doing on any given day. Caution is absolutely needed when enabling agents and giving them access to your systems. But the time to prepare is now. If you’re not getting your platforms ready, you’re going to be in for a shock when you and the regulators are ready.
What would you say to a CTO who’s watching all of this but hasn’t moved yet?
This is happening. You need to move carefully and thoughtfully, but you need to be getting your people, platforms, and processes ready fast. You’re either driving the train or you’re standing on the tracks. And that rumbling is getting louder.
