On June 2 and 3, OutSystems took over the RAI in Amsterdam for ONE 2026 under a banner that left little room for ambiguity: Where AI meets enterprise impact. If last year the conversation was still about whether generative AI belonged in serious enterprise development, this year that debate was simply over. The whole event was organized around a single premise: that the "agentic future" is no longer a roadmap slide but something teams are shipping into production today. The numbers OutSystems brought to the stage backed that up. Its own 2026 State of AI Development report found that 96% of the organizations surveyed are already using AI agents in some form, a striking shift from experimentation to execution in the space of a single year. ONE 2026 was, in effect, the company planting its flag on that shift.
The platform got agentic
The headline announcement was the Agentic Systems Platform, powered by what OutSystems is calling the Enterprise Context Graph, the layer meant to give agents the context they need while preserving the governance, security, and control that enterprises can't compromise on. Around it sat a cluster of capabilities aimed squarely at builders: an Agent Workbench for assembling agents, and Mentor, the AI woven through the development experience. The framing from OutSystems' leadership was consistent throughout: this is about moving past AI demos and into high-consequence, mission-critical systems. As CEO Woodson Martin has framed it, writing software and building AI systems are now effectively the same discipline, and the harder problem is no longer adoption but standing up an architecture that can coordinate agents safely at scale. For those of us who build on the platform, the takeaway is concrete: agents you can build, govern, and run in one place, rather than stitched together across half a dozen tools.
What builders were actually digging into
Keynotes set the direction, but the breakout and expert sessions are where you see how mature something really is, and this is where ONE 2026 was most encouraging. The technical track read like a tour of the problems serious AI engineering is hitting right now: multi-provider LLM routing inside OutSystems, native vector storage landing in ODC, modernizing legacy systems onto ODC, agentic AI applied to end-user support, advanced engineering patterns, security-first architecture, accessibility done properly, and a deep dive into the new mobile UI framework. The signal underneath all of it is that the platform is absorbing AI primitives (vectors, routing, agents) natively, rather than leaving them as bolt-ons. That's the difference between a vendor reacting to AI and a vendor building for it. It was also striking how much of the strongest technical content came from the community itself. A good share of these sessions were delivered by OutSystems Champions and MVPs, not only OutSystems staff, which says something healthy about the depth of the practitioner bench.
What we're taking back to Bool
If I had to compress ONE 2026 into a sentence: the agentic shift is real, but the differentiator won't be the models. It will be disciplined engineering, and the people who know how to lead it. The platform is clearly built to let us ship agents into production with the controls enterprises need; the harder part is the judgment, the architecture, and the engineering discipline to keep those systems safe once they're live. It is the argument I took to the stage myself. In my session, When AI Learns Samba, I ran two models on a single phone with everything on device, a drum-pattern classifier paired with a small on-device language model, to make exactly that point: the win came from the engineering, not from reaching for the largest model available. The full session is on demand [here]
We left Amsterdam with a clear roadmap and a longer to-do list. Both are good problems to have.
About the author
