AI agents promise to automate complex work, but without a persistent state they quickly lose reliability. In this interview, Rookoo co-founder Jeroen Borloo explains why conversational AI breaks down in multi-constraint workflows, how state-first agent architectures prevent drift, and why human oversight remains essential in high-stakes B2B transactions.
1. You’ve written openly about experiencing impostor syndrome even after more than 20 years of professional experience. How has that persistent self-doubt shaped the way you lead Rookoo, and did it influence your decision to build a product designed to reduce anxiety and cognitive overload for frontline teams?
Jeroen Borloo: We’ve also reflected internally on impostor syndrome, including a post by our designer Jared titled “Imposterception.”
Nonetheless, impostor syndrome forces you to stay awake in your own story. It keeps you curious and reminds you that you’re constantly learning. As long as it doesn’t paralyze you, it can actually be a strength. When we started Rookoo we all came from years of being employees and suddenly decided we wanted to build a company from scratch. That requires a healthy dose of self-reflection and the willingness to say “I don’t know yet”.
Inside Rookoo we try to make that normal. You’re allowed to make mistakes as long as you learn and you share the learning. That mindset comes straight from watching frontline teams struggle at Clarabridge and Qualtrics. These teams were always chasing their backlog while juggling a mountain of expectations. The turnover was brutal. We wanted to build something that gives them a bit of breathing space so they can focus on their work with more headroom and less anxiety. Creativity flows when you’re not drowning. That’s what we want to support.
2. Before Rookoo, your team spent years at Qualtrics and Clarabridge turning messy human language into structured insight. With Rookoo, you’ve inverted that logic by using language to execute work, not just analyze it. What was the hardest conceptual shift in moving from understanding customers to letting software actively act on their behalf?
Jeroen Borloo: Understanding customers is still the hardest part. You start with a set of assumptions about what customers want and then you discover that only seventy to eighty percent behaves the way you expect. You need an operating framework that can deal with that other 20 to 30%
We had to move away from deterministic flows and build something that can navigate context freely inside clear boundaries. That also means guiding customers through options instead of assuming they know what’s possible. Once you understand the intent and the constraints, the execution becomes the easy part. Getting that understanding right is the hardest part.
3. Rookoo avoids the term “chatbot” and positions itself as a “digital colleague.” From a systems perspective, what is the defining technical difference between a conversational interface and an agent that can reliably reason across long, constraint-heavy workflows?
Jeroen Borloo: A chatbot talks, a digital colleague works alongside its human peers.
Behind the scenes our agent runs procedures, executes steps, pulls data from CRMs, checks availability, validates requirements, sends internal approval requests and keeps the whole flow consistent. It reasons about what should happen next based on the state of the conversation and the rules of the venue or operator.
So whilst still ‘chatting’ it also acts upon what is being discussed.
From a systems perspective the main difference is that we separate the conversation layer from the decision and execution layer. The conversational interface is just one client of an underlying agent that works on top of a structured state. That state contains everything needed to reason about a booking. Things like dates, capacities, pricing rules, customer preferences, internal constraints and the history of previous actions.
The agent decides which tool or workflow to call next. For example check_availability, generate_proposal or request_internal_approval. Those tools sit behind an abstraction layer so the same agent logic can work across different backends like HubSpot, Odoo or a custom event system. Each tool call reads the central state, performs a domain action and writes the result back into that state so the agent never loses track when constraints change mid conversation.
On top of that we add validation, guardrails and idempotency. Inputs are checked against schemas and business rules before they are accepted. High risk actions require explicit confirmation or human sign-off. We tag operations with correlation IDs so retries do not create duplicate bookings. That combination of structured state, tool abstraction and controlled execution is what turns a chat interface into a digital colleague that can stay reliable across long complex workflows.
4. Event and experience bookings are fundamentally multi-variable problems. When a customer changes one constraint mid-conversation without relaxing others, large language models often lose track. How does Rookoo’s architecture maintain state consistency and avoid constraint drift in real-world negotiations?
Jeroen Borloo: We use a multi-agent setup with a shared state that acts as the single source of truth. Every requirement, every change and every action is logged in that state. That same state is synced to the CRM so the customer view stays aligned with reality.
Each agent specializes in part of the journey. One handles intake, one handles inspiration, one checks availability, one prepares a quote. They all work on top of the same state so nothing drifts when the customer updates one detail.
And because we’re active across channels like chat, email and WhatsApp we always know what was said last. We believe the source of truth should start with the customer’s intent, not with a backend system that often lags behind.
5. In B2B transactions, hallucinations are not just errors but potential legal and brand risks. Rookoo keeps humans in the loop by generating provisional orders rather than fully autonomous commitments. Do you see human oversight as a temporary safeguard, or as a permanent design principle for commercial AI agents?
Jeroen Borloo: For simple requests like meeting rooms with predictable setups I can see full automation working soon. For larger events the stakes are higher and the nature of events is too dynamic. Last-minute changes, supplier constraints and creative decisions require human judgment.
Human oversight is a key part of the design. The agent can do the heavy lifting and prepare everything for confirmation but the final step remains with a person who understands the nuance of that specific booking. It’s a balance between speed and responsibility.
6. Rather than replacing existing systems, Rookoo positions itself as an intelligent layer on top of CRMs and ERPs like Tripleseat, HubSpot, and Odoo. What were the hardest challenges in building an agent that behaves consistently across very different backend environments?
Jeroen Borloo: It’s still a challenge. Every environment behaves differently, so we built an abstraction layer that exposes domain actions like create a quote or check availability. The agent doesn’t need to know where that action lives. Under the hood it could be Odoo, HubSpot or a custom tool.
We use a library of agentic workflows as templates. Customers or our services team adapt them to their setup and the agent stays consistent. We also expose these domain actions as MCP tools so human users can trigger the same actions with the same logic.
7. The events and hospitality industry isn’t just chasing efficiency. It’s facing chronic staff shortages and burnout. From what you’ve observed with customers, has AI shifted from being a productivity upgrade to a structural necessity for keeping teams operational?
Jeroen Borloo: Absolutely. Many event teams are stretched thin and the work is not slowing down. We’re seeing more customers who don’t look at AI as a nice-to-have but as a way to stay operational without burning out their teams. The industry isn’t the fastest in adopting new tech but the shift is visible. Teams that used to run everything manually now see that an AI colleague is the only way to stay responsive without hiring ten extra people who don’t exist.
8. Many AI companies pursue standardized, high-volume workflows such as hotel room bookings. Rookoo chose the messier category of experiences, eatertainment, and non-standard events. Was that choice driven more by defensibility through complexity, or by a belief that standardized markets were already over-served?
Jeroen Borloo: To be honest, it started from passion. The entire Rookoo team all love the event world. It’s one of the last places where people connect offline in meaningful ways. I’ve been a musician and deejay and organized events for years and I’ve seen how admin kills creativity. The more unique the event, the more chaos in the planning.
That complexity is exactly why there’s so little good automation in this space. Most tools stop at the standard use cases. We wanted to solve the part that nobody wants to touch. Maybe that’s idealistic but if we can lower the cost and effort to create something unique then more people will get to experience it. We truly aim to set the new standard for organizing experiences.
9. Early on, you and your co-founders spent significant time working directly inside customer operations. What did you learn on the ground that you would never have discovered from dashboards or customer interviews alone, and how did those lessons reshape the product?
Jeroen Borloo: We learned that complexity kills adoption. If a tool requires 30 steps before you get any value, nobody uses it. Years in customer experience taught us that software needs to disappear into the workflow. I remember one customer at Clarabridge saying during the onboarding phase: “this feels like homework all over again.”
It should feel like help not homework.
We also learned that frontline teams often build their own shortcuts and workarounds because the official process doesn’t match the real world. I think that shaped how we designed the agent. It had to adapt to the team, not the other way around.
10. Rookoo is built and hosted entirely in Europe, under GDPR and the emerging EU AI Act. Many founders see this regulatory environment as a constraint. Have you found that European AI compliance has become a trust advantage when selling to enterprises and public-sector organizations?
Jeroen Borloo: Even when companies don’t fully understand the details, they know European standards are a lot stricter and that creates trust. In Europe trust is a major part of the buying decision, especially in public-sector or enterprise environments. Being fully EU-hosted and transparent about how we handle data makes the conversation easier. It does come with a slower sales cycle though.
We see a very different pace in the US where companies focus much more on speed and impact.
It does forces us to build the product in a way that stands up to scrutiny from day one, which in the long run is not a disadvantage at all.
11. As large platforms like Cvent or Tripleseat continue to develop their own generative AI capabilities, how do you think about Rookoo’s long-term moat? Is it proprietary interaction data, deep vertical expertise, or your position as a neutral intelligence layer across systems?
Jeroen Borloo: Our focus has been on agentic capability from day one. Most platforms are great at event management but their AI features tend to focus on event day or simple automations. The real bottleneck is the discovery and intake phase and that’s where we’ve invested heavily.
We’ve spent a full year refining how an agent reasons inside messy multi-variable requests and how it collaborates with humans. That vertical depth combined with our neutral position across backends gives us a strong advantage. We know how to deal with the chaos before the booking hits the system and that’s where most revenue is won or lost.
Now we can take everything we learned and put it to work in the next chapter of the product. The planning engine we’re building sits right behind the agent. It understands the operational reality of venues: buffers, room hierarchies, release windows, blackout periods and all the hidden constraints teams normally track in spreadsheets or in their heads.
Because the agent already knows how to collect clean requirements, the planning engine can turn those into real availability, smart slot suggestions and conflict-free allocations. The loop becomes highly integrated: The agent gathers the intent, the engine finds the best option and the agent turns that into a proposal a human can confirm.
It’s much easier to build this layer now because the hardest part, reliable reasoning in a messy domain, is already in place.
12. Looking five to ten years ahead, do you envision a future where AI agents on the buyer side negotiate directly with AI agents on the seller side, completing bookings machine-to-machine in milliseconds? In that world, what role do you believe humans should still play?
Jeroen Borloo: We’re already seeing the first signs how agent to agent could be beneficial for both buyers and our customers. When booking an event, we see lots of “shoppers”, asking for the same information directing it to multiple event venues. Often, they have to wait before they can make an informed decision. Imagine they have an agent, tasked with doing a market study on their requirements, reaching out to venues using a Rookoo agent. Faster and more complete answers increase the chance of conversion, so to earn your seat at the negotiating table, this could be instrumental in a world where buyers now expect this all the time.
After this initial phase, the humans step in. The event industry is a very social sector, where human interaction is still an instrumental part of the experience. AI agents act as the supportive operational layer between buyers and venues, so humans can focus on the part that actually makes an experience memorable.
Editor’s Note
This interview highlights a structural constraint at the requirements layer, where human intent struggles to remain intact as system complexity increases.
