Google DeepMind announced an expansion of PlanAssist on October 31. The Gemini-based assistant supports the processing of applications for changes to homes. The UK government also confirms a move from three participating planning authorities to 15 in England, with testing over 12 months. This expands an experiment in actual administrative work; it does not establish a nationwide service with equally proven results everywhere.
The system checks documents and applicable constraints and prepares a referenced report that an officer can edit. A qualified planning professional retains the final decision. The connection between source material, recommendation and human judgment is the significant part of this approach.
### What the reported time saving measures
The official announcements report a 58% reduction in the officer time devoted to assessment tasks during early piloting. This is a result reported by the project's participants; SIExpo has not independently tested it. It does not automatically measure every day between an application's submission and the final decision.
That distinction matters in practice. Reducing the working hours needed for an internal check can still leave overall waiting time dependent on missing documents, objections or another step. Applicants care about the final deadline, while the administration also cares about staff capacity. Tracking these outcomes separately avoids turning a useful improvement into an excessively broad promise.
Consider a hypothetical application for a home extension. An assistant would be useful if every concern could be traced to the relevant source. A polished summary alone does not establish that the correct version of a local policy was used. An inspectable reference lets the officer compare the recommendation with the document, rather than treating confident language as evidence.
### A larger trial must examine quality too
Our editorial interpretation is that transferability is the next important test. A system might help with well-prepared applications while requiring more correction when drawings are unclear or documents conflict. Average task duration therefore needs to be considered alongside the proportion of outputs requiring changes and the reasons for those changes.
A practical evaluation would distinguish time spent on initial reading, checking references and correcting mistakes. It would also compare cases of similar complexity, rather than placing the easiest applications in one group and difficult ones in the other. These are editorial evaluation criteria, not additional results already demonstrated by the project.
Human oversight must also involve actual work: inspecting the specific grounds for a recommendation and being able to reject it. Formal confirmation without that scrutiny would weaken the distinction between an assistant and an accountable authority. Useful automation should make the reasoning easier to trace, as well as make the text faster to produce.
PlanAssist's expansion is a significant example of moving from a demonstration toward a broader administrative pilot. The confirmed development is its increased scope. Whether that produces sustainably faster work with comparable accuracy across different councils remains a question for the next evaluation.
Editorial check: 2026-11-01T19:30:20+02:00 (Europe/Sofia).
