Iris ten Teije argues that the traditional CI pipeline model, designed for a single frozen software artifact, is obsolete due to advances enabling real-time, personalized software adaptations within the running environment. Her vision promotes a new paradigm where adaptive, individualized software versions coexist safely alongside a canonical codebase, overcoming previous limitations and ushering in a scalable, trustworthy era of personalized software delivery.
In her talk, Iris ten Teije challenges the long-standing software development paradigm centered around the CI pipeline, which was designed to move a single, frozen artifact of software from build to run environments. This pipeline was built on the assumption that producing software was expensive and risky, necessitating a one-version-for-all approach to ensure reliability, reproducibility, and rollback capabilities. However, this model inherently limited software personalization, as maintaining multiple versions for different users was financially and technically unfeasible.
Iris argues that this foundational assumption is no longer valid. Advances in technology have drastically reduced the cost and complexity of producing correct software changes, enabling real-time, adaptive modifications that can occur in the same environment where the software runs. This shift dissolves the traditional separation between development and distribution, allowing software to be personalized and adapted live for individual users without the need for separate builds or deployments.
The demand for personalized software has existed for decades, exemplified by enterprise professional services, customizable developer environments, and tools like Excel that empower users to create their own workflows. While previous attempts at personalization relied on pre-defined segments or feature flags, the new approach enables truly adaptive software that evolves uniquely for each user. Iris envisions a system where a canonical “stem” codebase exists alongside immutable, isolated divergences tailored to individual users, ensuring safety and reversibility of changes.
Iris acknowledges the challenges of this new model, including defining the source of truth when no single artifact exists, ensuring correctness and desirability of adaptations, balancing autonomy with developer control, and coordinating updates across millions of personalized versions. Her team’s approach focuses on merging intent and outcomes rather than code, maintaining clear boundaries between divergences, and building trust in autonomous systems that can adapt software responsibly without constant human oversight.
In conclusion, Iris frames this transition as a fundamental shift akin to past technological revolutions in software delivery. The pipeline is not failing because it is broken, but because its underlying assumptions no longer hold. As software production becomes as cheap and dynamic as running it, the future lies in delivering the right version of software to each user individually, with the safety and transparency needed to make this scalable and trustworthy. This vision, embodied by her company Differ, represents a new era of adaptive software tailored to the needs of anyone, anywhere.