What's shipping
Blacktown City Council's DAISY AI assistant is processing over 450 questions weekly about residential development applications, according to council data from its January 2026 soft launch. The tool has supported 60+ applications so far - modest numbers, but the council calls early engagement "encouraging."
DAISY answers questions about local planning rules, application requirements, and documentation for low-density residential projects like house alterations and granny flats. It's trained on Blacktown's Local Environmental Plan, Development Control Plans, and NSW State Environmental Planning Policies.
Why this matters
More than 80% of Blacktown's development applications come from individual residents, not professional developers. The council says many submissions arrive incomplete, delaying approvals. If AI can front-load accuracy, it could compress timelines across NSW's ~60,000 annual applications - the government estimates "hundreds of thousands of days" in potential savings.
Blacktown is one of 16 NSW councils testing AI planning tools under a $5.6M state initiative, with $2.7M allocated through Early Adopter Grants. Blacktown received $190,000; others like Newcastle got $390,200 for joint projects. City of Sydney is running a parallel pilot for Complying Development Certificates.
The fine print
DAISY runs on Microsoft Azure with no third-party model training - a nod to data sovereignty concerns common in government AI deployments. The council warns users that the AI "can make mistakes" and that staff make all final decisions. This isn't automation replacing assessors; it's triage.
Notably absent: accuracy metrics. The council is tracking "usage data and interaction levels" but hasn't published error rates or how often DAISY's guidance holds up under formal assessment. That data will determine whether this scales beyond the 12-month trial window.
What to watch
Whether AI can handle planning's messy edge cases - heritage overlays, flood zones, neighbor objections - or just the straightforward stuff. If it's the latter, the efficiency gains narrow considerably. The real test comes when councils publish outcome data: did AI-assisted applications actually move faster, or did they just shift workload upstream?
History suggests councils handle 85% of NSW's new dwelling approvals. If 16 pilots show meaningful time savings without quality drops, expect rapid rollout. If not, this becomes another "interesting experiment" that didn't ship at scale.