November 2022: the insight that changed our direction
ChatGPT was released at the end of November 2022. Many saw it as an entertaining conversation partner. Tuomas Palenius saw something else.
Tuomas had built integrations his entire career. Soon after the release, he realized that within a few years AI would automate coding – not partly, but completely. And if coding gets automated, so do integrations.
So the question was not whether the change would happen. The question was who would build it for integrations. We decided it should be the people who know integrations best: us. We needed an AI-native integration platform of our own.
Consulting and AI pull in opposite directions
Our background is at Devikone, a consultancy specializing in integrations, so we saw the conflict up close. In the logic of consulting, the best outcome is the longest possible project. In the logic of an AI product, the best outcome is that the customer gets a working integration as quickly as possible.
We saw it in our own work. AI-assisted development compressed a quarter's worth of Apache Camel work into weeks. Our clients were happy. Our hourly-billing margins were not.
You cannot serve both at once. We had to choose, and our values made the choice easy:
The customer gets an integration at exactly the right price, as quickly as possible – without compromising on quality.
That sentence is the foundation of FastHub. It is why we don't bill by the hour but by connectors and service level: the price follows the scope of the integration, not how long it takes to build.
Three versions and a moving target
In the early days, we were aiming at a moving target. AI evolved faster than any product roadmap, and every leap changed what was possible.
The first version was a structured design tool. Integrations were described in a structured form, and prompts were supposed to turn that description into a working implementation. It was the right answer for the AI of its time, which could write code but needed a tight framework.
The second version arrived when AI agents became possible. We built our own agent platform that coded integrations on top of a graph engine we had developed ourselves, using Apache Camel's EIP patterns. The agents were already doing real work, but an integration could only be as versatile as our own engine allowed.
The third version overcame the limits of both. FastHub is now an AI-native platform that writes pure Apache Camel. There is no proprietary layer in between to set a ceiling: an integration can be of almost any complexity.
A foundation built for the future
Twice we built a layer between the AI and the code, and twice the AI outgrew it. The third time, we removed the layer in between. Now every step forward in the models makes FastHub better instead of obsolete.
Apache Camel is the industry's established open-source integration framework. The integrations FastHub builds are therefore standard code that any Camel developer can read – not a closed black box. AI does the work, but a human approves the plan before it is implemented and the release before it goes to production.
FastHub is built in Turku, Finland, entirely on open-source technologies and hosted 100% in the EU. The moving-target phase is behind us.
