PlanAssist and the First Assessment Before the Planning Decision

An assessment before the assessment

On 31 October, England’s PlanAssist trial expanded to 15 councils for 12 months. The government says early testing in Camden, Barnet and Dorset cut time spent on key planning tasks by 58%. The tool brings applications, local policies and site constraints together, then gives trained planning officers an evidence-backed initial assessment. Officers still review applications and make the decisions. ^[1]

That phrase—“initial assessment”—interests me. Initial sounds temporary, almost disposable. Assessment sounds considered and consequential. Put them together and we get something that is officially preliminary but practically persuasive: a first organized account of what matters. PlanAssist does not grant permission for your loft conversion. It may, however, help construct the version of your loft conversion that the decision-maker first encounters. ^[1]

The demonstrated change is straightforward: software now compiles planning evidence and supplies an initial assessment during a live, multi-council trial. Officials will use it alongside their normal process, and the government intends—subject to this testing—to make it available to every English council by the end of 2027. Availability is the intention; successful national performance has not yet been demonstrated. ^[1]

The frame has weight

We tend to locate power at the end of a process. Who approved it? Who rejected it? Who signed the page? Yet judgment begins earlier, with selection. Which policy appears relevant? Which constraint is prominent? Which objection is summarized first, clearly, or at all? A person may retain the final word while inheriting the machine’s first grammar. The decision stays human; the decision-shaped space may not. ^[1]

This is where the word “assist” needs assistance. Help can expand agency by gathering scattered material and returning time to an overloaded officer. Help can also narrow agency if a polished assessment becomes the path of least resistance. I am not claiming that PlanAssist does this. The published announcement reports time savings, but it does not provide an independent error analysis or evidence that final decisions became more accurate or consistent. ^[1]

A useful warning comes from a very different setting. WIRED reported cases in which hikers relied on chatbot guidance that was wrong or unsafe. A human-computing researcher described “cognitive surrender”: people can stop thinking critically when an apparently expert system speaks confidently, especially outside their own expertise. Planning officers are trained professionals, not lost hikers, so the comparison proves nothing about PlanAssist. It does identify a human tendency worth testing rather than dismissing. ^[2]

Time saved, judgment spent?

The case for PlanAssist is substantial. Routine householder applications can occupy attention that might be better spent on difficult developments. The government says householder cases account for nearly 70% of applications, and it aims to reduce routine processing times from eight weeks to four. For a family waiting to build an extension or a shopkeeper improving a premises, administrative delay is not abstract. It costs time, money and sometimes nerve. ^[1]

Refusing useful automation also has a hidden cost. Public servants repeat clerical work. Applicants wait. Complex cases compete with simple ones for scarce attention. If PlanAssist reliably collects the correct policies and constraints, an officer may gain the breathing room required for actual judgment. Scepticism should not become a fondness for queues. Bureaucratic friction is not automatically democratic friction; sometimes it is merely friction. ^[1]

But speed changes incentives. Once an institution expects faster decisions, saved time may not remain available for deeper review; it may simply become the new production target. The officer could gain a better tool while losing the practical freedom to start again. Applicants benefit from quicker answers. Councils may stretch limited capacity further. The hidden cost falls on whoever needs the first assessment challenged—and on the officer held responsible for an account they did not fully construct. ^[1]

Read beneath the signature

The sensible test is therefore not whether a human remains “in the loop.” A loop can be large enough to contain a person and still leave no room to move. What matters is whether the officer can inspect the underlying evidence, notice omissions, revise the assessment and take the time required to disagree. The announced trial offers useful safeguards: officers remain responsible, PlanAssist is not mandated, and it will run alongside normal processes while performance is tested at scale. ^[1]

For an ordinary applicant—or a neighbour affected by an application—the practical habit is to question the frame, not merely the verdict. If you can review an assessment, check whether the relevant policy, site constraint and substance of your evidence appear accurately. Ask for an omission to be corrected. A confident summary is still a summary. The useful question is not only, “Was AI involved?” but, “Could the responsible person see beyond what it assembled?”

PlanAssist may prove to be careful public-service automation: quicker paperwork, clearer evidence, more time for difficult cases. I hope it does. Yet the trial should measure disagreement as seriously as agreement—what officers changed, what the system missed, and whether people had room to challenge its frame. The first assessment need not become the final decision. Keeping that distinction real, rather than merely grammatical, is the work ahead. ^[1]