Enterprise integration has always had a people problem hiding inside a technology problem.
The systems that need connecting (Jira, ServiceNow, Salesforce, Azure DevOps) are well understood. The APIs exist. The data models are documented. And yet, for most organizations, getting two platforms to talk to each other reliably still takes weeks, sometimes months.
In some cases, business stakeholders report waiting years, not because the work is impossible, but because the person who understands what needs to happen isn’t the same person capable of making it happen.
That gap between business intent and technical execution is the problem Exalate has been working to close. Not with a shortcut, and not by simplifying away the complexity, but by rethinking who should be in the room when integrations are built.
This is the story of how that vision became a product, what it took to get there, and where it’s heading next.
The Problem Was Never the Technology
To understand why Exalate’s AI initiative matters, it helps to understand the friction it was designed to address.
Exalate specializes in complex integrations: the kind that involve custom sync logic, cross-company data flows, and configurations that need to stay maintained as both systems evolve.
Until recently, that complexity demanded a technical expert at the lead. A developer or solutions engineer who could translate business requirements into working Groovy scripts, field mappings, and conditional rules.
The problem, as Bruno Dauwe, Exalate’s Product Manager, describes it, is that this person is almost never the one with the most at stake. “What we discovered over time is that the people who are the actual stakeholders are non-technical people. They were somewhere in the background. Maybe they discussed requirements with the technical person, but the technical person was at the forefront. And for us, this felt wrong. You should be working with the people who have the most stakes.”
The result was a persistent bottleneck. “We have customers and stakeholders who have been waiting for two years for an integration because their IT department is overloaded and not prioritizing this. If instead of being a passive victim of this situation, you can become an active participant because you can start defining the integration yourself, just describe it as a human, and then AI, instead of the technical person, turns this into a working integration.”
That framing shaped everything that followed.
Where the AI Experiment Began
Exalate’s first move into AI started with a focused internal experiment, in collaboration with an AI specialist partner, to identify where in the business AI could create the most value.
Two areas stood out immediately.
The first was documentation. The second was configuration: specifically, the scripting process that sits at the heart of how Exalate builds integrations.
As Bruno puts it: “The script being a technical activity, while it should actually not be so technical. That’s where we thought we could be improving most.”
These two friction points became the foundation of Exalate’s AI initiative, and they spawned two distinct tools that would eventually be brought together: Aida and AI Assist.
Two Tools, One Vision
Aida came first. The concept was straightforward: rather than requiring users to search through static documentation, the documentation itself became source material for an AI.
“Instead of browsing to your documentation, the AI just assembles information for you and creates on the fly a dynamic documentation page exactly catered to your question,” Bruno explains. “We had very quick positive feedback on that.”
The harder challenge was scripting.
The quality of AI at the time wasn’t yet sufficient to take the leap directly to zero technical knowledge. So the team took a deliberate step in between.
AI Assist was built around that insight: accept plain language descriptions of what the integration should do, and generate the corresponding Groovy sync logic.
A user could describe: “Sync high-priority Jira work items to ServiceNow incidents, map statuses across both platforms, and include internal comments only,” and receive a working, reviewable script.
The advantage Exalate had here was structural. “Being 100% script-based allowed us to move really quickly forward,” Bruno notes, “because LLMs and code generation are in the heart of what we are good at.”
The Hard Part: Getting to 90%
Launching AI Assist was the easy part. Making it reliable enough to trust was not.
Early versions achieved an acceptance rate of around 60%. Roughly 4 in 10 generated scripts were rejected, either because they were inaccurate, incomplete, or needed significant manual revision. At that level, the tool was interesting but not yet valuable.
The solution was the AI Professors system: a structured, human-guided improvement loop built around Exalate’s own solution engineers. Rather than relying on automated self-improvement, the team gave solution engineers the chance to curate the script library and refine the prompts, sorting priorities that shaped its responses.
The decision to keep humans in that loop rather than letting the model improve itself was deliberate and came from direct experience.
“We did experiments with a self-improving system,” Bruno explains, “and our partner showed us that this, on paper, looks good, but the solution starts improving, and at some point, you cannot see why it was improving itself in a certain direction anymore. If that direction starts going the wrong way, there’s nothing you can do except reset the whole system and recalibrate. By keeping a human in that loop, you can keep directing the improvement intentionally.”
Crucially, this process never involved exposure to actual customer data or live configurations. Improvements happened at the model and prompt level only, ensuring no new security exposure was created through the improvement loop itself.
The results were steady and measurable. Acceptance rates climbed from 60% to 90%, the threshold the team had defined internally as the minimum bar for genuine value.
“Reaching 90% was the ultimate milestone,” Bruno says. “That’s where we said we have something in our hands that is really valuable. If 90% of the time the system is right, there is no doubt this is the future.”
It also became the trigger for the next phase. “90% was a minimum that we set out to start looking into the next step,” Bruno explains, “the next step being citizen integration.”
Bringing Aida and AI Assist Together
With both tools running in parallel, the team faced a question: should they remain separate, or was there a more powerful version where they worked together?
The answer came from observing how users actually moved between them. Aida was used to understand what was possible and scope the requirements. AI Assist was used to build the integration. They were already working sequentially in practice.
“We had different initiatives running, and we decided, let’s rearrange all of this so they can collaborate together,” Bruno explains. “If Aida needs a piece of script, it just goes to AI Assist. If AI Assist needs to explain something, it goes to Aida. Together they form one answer.”
This orchestration layer was more than a product decision. It was a signal about what AI’s role in the integration process was becoming: not a feature sitting alongside the workflow, but a participant embedded within it, with a defined role and clear boundaries.
Enterprise Readiness Wasn’t an Afterthought
Introducing AI into a product that handles sensitive, often cross-company data required careful thought. “When we started adding AI into our product, we immediately had customers asking: ‘ In our company, this is not allowed,” Bruno recalls.
Exalate’s response was to make the AI layer fully opt-in from day one. Organizations that didn’t want AI involved could disable it entirely with no impact on the rest of the product.
For those who enabled it, a hard boundary was established: no customer data used in AI processing ever leaves Exalate’s managed domain, and none of it is used to train external models.
This sits within Exalate’s broader security posture: ISO 27001:2022 certification, end-to-end encryption, role-based access controls, and full documentation through a public Trust Center.
The principle, as Exalate’s IT Manager, frames it: “AI can handle a lot of the grunt work, but someone still needs to make sure the system is secure and that the integration follows company policies. AI can automate, but we still need human oversight.”
That oversight is built into the product itself. Every AI-generated script is presented as a suggestion, with color-coded highlights showing exactly what would be added or modified. The user reviews, accepts, adjusts, or discards. Nothing applies automatically.
The Bigger Shift: Redefining the Actors
The 90% milestone and the governance foundations were meaningful achievements. But for Exalate, they were always in service of a larger ambition.
This is what Exalate calls citizen integration: a model where business users can describe, define, and manage integration behaviour directly, without needing to write code or rely on a technical intermediary.
Importantly, IT doesn’t disappear from this model. Its role becomes clearer.
“We do not want to eliminate IT out of the picture,” Bruno is emphatic about this. “The stakeholder from the business side is responsible for the shape and behaviour of the connection. IT is responsible for all the guardrails surrounding how it should perform: security, scalability, and performance. You should really consider the AI as just an actor in this story, who also needs to behave, has a certain role to play, and should not go outside of that role.”
The result, as Francis Martens, Exalate’s CEO, describes it, is a shift in what IT leadership looks like: “IT leaders will move from being implementers to strategists. Instead of worrying about every little detail, they’ll be focused on making sure AI integrations stay aligned with business needs.”
AI Alone Isn’t Enough: Why Product Thinking Matters
One of the more candid observations from Exalate’s product leadership is that simply layering AI onto an existing product is the wrong approach.
Citizen integration, combined with some of the more traditional improvements: trust levels, collaborative platform, and refined roles, would make a huge difference.
The reasoning is straightforward. The AI capability (letting a non-technical user define integration logic in plain language) only delivers its full value if the product also supports clear role separation, governance structures, and collaboration workflows between business and IT. Those are traditional product development problems. But they have to be solved in parallel, or the AI capability lands in a product that can’t support the new way of working it enables.
What Came Next
By the end of 2025, Aida evolved from a documentation assistant into a full scripting co-pilot embedded inside the product. AI Assist had been refined through the AI Professors programme to a 90% acceptance rate. The two had been brought together under a shared orchestration layer. Governance controls had been built in from the ground up. And the product foundations for citizen integration were taking shape.
“2025 was developing the core AI capabilities,” Bruno summarises. “And now in 2026 it’s reshaping the product around these AI features, instead of just adding them as a layer on top.”
The early signals from users were consistent: once teams started using AI-assisted configuration, they kept using it. “People who started using it for the first time just keep coming back,” Bruno notes.
Anecdotally, users report completing integrations up to five times faster than before. No intermediary, no queue, no translation layer.
As Francis Martens frames the broader significance: “Integration has historically been the domain of specialists, creating bottlenecks for business transformation. We’re removing those bottlenecks and enabling enterprises to connect their systems at the speed business demands, not at the pace their technical resources allow.”
The story of Exalate’s AI journey isn’t primarily a story about technology. It’s a story about who gets to participate in building the integrations that modern enterprises run on, and a deliberate, methodical effort to make the answer: everyone who needs to.
Want to see Aida in action?
Explore Exalate’s AI capabilities or book a demo to see how AI-assisted integration works in practice.
Recommended Reads:
