OutSystems recently conducted a survey of over 1,900 IT professionals on the state of AI development.
We wanted to highlight a few stats that are especially relevant for those learning and building with AI development on the OutSystems platform:
We will go deeper into these findings on May 13 in a global webinar with Tiago Azevedo (OutSystems CIO), Kris Lande (OutSystems CMO), and other industry leaders, focusing on how teams are approaching AI development in practice.
š Register here: Insights from the state of AI development 2026
Curious to hear from the community: which of these technical challenges is currently a bigger focus for your team, building orchestration layers or legacy integration?
Let us know in the comments below, and we hope to see you at the webinar!
Iād argue that these two challenges are more connected than they initially appear. In real-world AI implementations, orchestration becomes significantly harder when agents have to interact with legacy systems that were never designed for autonomous workflows.
Legacy integration is often the bigger constraint because it determines what the AI can actually access, modify, and trust. You can build a sophisticated agent, but if the underlying ERP, CRM, or internal systems have limited APIs, the orchestration layer quickly becomes full of workarounds.
From what Iāve seen in AI projects at Technource, a practical approach is to establish a reliable integration layer first, then introduce human-in-the-loop checkpoints around higher-risk actions. This gives teams clearer boundaries around what the agent can handle autonomously and when human approval is required.
For teams moving beyond AI prototypes, I think the bigger question is whether to modernize legacy systems first or build an abstraction layer that allows AI workflows to work with them without replacing everything.