Large old trees provide essential habitats for birds and many other species, yet they are rapidly disappearing from many landscapes. While artificial habitat structures have been trialled, their design rarely captures the morphological complexity of natural habitats. This limitation stems from both challenges in extracting relevant features from natural forms and the difficulty of developing cost-effective systems that can be reproduced at scale. This paper addresses this gap by presenting FloaTree, an experimental example of a human–machine design workflow to generate, optimise, and construct tensegrity structures derived from AI-generated visual abstractions of large trees. We developed a parametric workflow that translates such AI-generated polyline abstractions into X-module tensegrity configurations, refined through structural optimisation and represented via connectivity matrices. Iterative prototyping, from small-scale tests to an eight-module pavilion, validated the structural and constructability aspects of this workflow and culminated in the winning entry of the 2024 IASS “Design Competition and Exhibition of Innovative Lightweight Structures” in Zurich. The results demonstrate that tensegrity structures, typically confined to artistic installations or used with limitations as surrogates for other typologies, can be designed for packability, transport, and rapid low-tech assembly to enable their potential application in artificial habitat structures. The project also advances tensegrity design methods through a novel human-machine workflow and a visualisation technique based on connectivity matrices. It shows how the analogue and digital domains can co-exist in design workflows alongside emerging forms of human–AI collaboration. While ecological performance requires future field testing, the significance of this work lies in reframing tensegrity not only as an experimental artefact but as a transferable design framework integrating form abstraction, structural logic, and constructability, thereby suggesting broader applications for computationally optimised yet low-tech structures in disturbed landscapes.