I Used an AI Architect to Design a Subdivision. The Result Was Surprisingly Useful.

I spend a lot of time looking at land parcels and trying to figure out what can be built on them. How many units will fit. What the setback requirements are. Whether the numbers work.

This process traditionally involves engaging an architect for a preliminary feasibility sketch — $2,000 to $5,000 and two to three weeks. For properties we don't end up purchasing, that's dead money.

So when I came across an AI-powered architectural design platform that claimed to generate compliant residential development concepts in seconds, I was sceptical but curious. I tested it on a real parcel in Cranbourne — a suburb where we transact regularly — and the results were genuinely interesting 1.

This is not an advertisement. I have no financial relationship with the company. I'm writing this because the technology is reaching a point where property investors need to understand what it can and can't do.

What I tested and how

I uploaded a site plan of a vacant Cranbourne lot — actual dimensions and boundaries from a recent survey. I also uploaded a reference image showing the architectural style I wanted (modern townhouse).

Two prompts:

1. Generate a residential development layout within site boundaries, following Victorian planning standards, with maximum feasible units.
2. Create a construction time-lapse animation.

The image result appeared in approximately six seconds. The video took thirty seconds 2.

The output included:

• Site layout showing three residential units
• Approximate unit positions relative to boundaries
• Driveway access and parking configuration
• Rendered perspective view
• Time-lapse construction animation

The aesthetic quality was impressive. If you showed it to an investor or bank, it would communicate the concept clearly.

I should provide some background on why this matters commercially. Our typical workflow when sourcing development-potential properties for clients involves assessing 30-50 sites per month. Of those, roughly 5-10 will be physically inspected by our field team. Of the inspected properties, 2-3 will receive a formal architectural feasibility assessment. And of those, 1 might actually be purchased.

At $3,000-$5,000 per feasibility sketch, the cost of eliminated properties adds up quickly. Over a year, we might spend $50,000-$75,000 on architectural assessments for properties we don't buy. Any technology that reduces this dead cost while maintaining quality filtering is commercially significant.

The AI tool I tested has the potential to cut that cost by 60-70% by eliminating non-viable sites at the concept stage before engaging a human architect. That's $35,000-$50,000 in annual savings for our operation alone — savings that translate directly into lower fees or higher service quality for clients.

I should note that the quality of AI-generated design output depends heavily on the quality of the input. A blurry satellite image with approximate boundaries will produce a vague, unreliable concept. An accurate survey plan with precise dimensions, orientation, and topography will produce a concept that is directionally useful.

For this test, I used a recent professional survey — the same survey that would be submitted with a development application. The AI had accurate boundary data, correct orientation (critical for solar access calculations), and precise dimensions. The output quality reflected the input quality.

This is an important caveat for anyone experimenting with these tools. The technology works best when fed professional-grade inputs. Using a screenshot from Google Maps will produce a result that looks impressive but is dimensionally inaccurate — and inaccurate dimensions in development design lead to non-compliant buildings.

The implication is that AI design tools don't eliminate the need for professional surveying — they depend on it. The survey cost ($2,000-$3,000 for a boundary and feature survey) remains a necessary pre-condition for any useful design output, whether generated by AI or by a human architect.

What the AI got right

The unit count was plausible. Three units on this lot size is consistent with what a human architect would propose given GRZ zoning requirements 3.

The general layout — units maximising frontage, shared driveway, north-facing living areas — reflected sound design principles.

The speed is the obvious advantage. Six seconds versus three weeks. Even requiring professional refinement, having a preliminary concept in seconds transforms the feasibility assessment process. Instead of paying $3,000 for a sketch, I can generate a concept for free, assess viability in minutes, and only engage the architect for sites that pass initial screening 4.

For small-scale developers doing multi-site assessments — which is exactly our workflow — this is a genuine efficiency gain.

Let me put the speed advantage in commercial context.

Last month, our team assessed fourteen potential development sites for a client looking to subdivide a large block and build two townhouses. Of the fourteen sites, we physically inspected seven, and of those seven, we commissioned architectural feasibility studies on three.

The total cost of the three architectural studies was $11,500. The total elapsed time from commissioning to receiving the final reports was seventeen days. Of the three studies, one concluded that the site was not viable (the sewer connection would have cost $45,000, destroying the project economics). That study cost $3,800 and contributed nothing to the outcome — pure dead money.

If we'd been able to screen that site with an AI tool first — identifying the sewer issue within minutes rather than weeks — we'd have saved $3,800 and seventeen days. Multiplied across a year of development assessments, the efficiency gain is substantial.

The key limitation today is that AI tools cannot access utility infrastructure data. They can generate a visual concept, but they can't tell you where the sewer runs, whether there's a gas main under the driveway, or how far the nearest electrical transformer is. These are the factors that actually determine development viability — and they require human investigation.

So the optimal workflow today looks like this:

1. AI screening (minutes): generate a concept, assess whether the site geometry supports the intended development
2. Utility investigation (1-2 weeks): confirm service connection feasibility through Dial Before You Dig and direct utility authority enquiries
3. Professional design (2-3 weeks): engage an architect for the sites that pass both the AI screen and the utility check

This three-stage funnel eliminates non-viable sites at the cheapest possible stage, reserving expensive professional resources for sites with genuine potential.

What the AI got wrong (and why you still need professionals)

The AI produced a concept that looked correct. But "looks correct" and "is correct" are different things in planning law.

Issues I identified:

• Setback accuracy: AI doesn't calculate from specific boundary lines. It approximates based on training data [5].
• Easement awareness: The lot has a 3-metre sewer easement. The AI placed a unit partially within it. Council would reject this immediately.
• Building height: AI defaulted to two-storey without checking zoning height limits.
• Car parking: Victorian planning requires specific spaces per dwelling. The AI's parking appeared to fall short.
• Services: No consideration of sewer, water, or electrical connection points [6].

The tool generates a concept. Not a compliant design. Using AI-generated designs for a development application without professional review would result in rejected applications and wasted fees.

There's a broader question here about the future of architectural services in residential development that I think is worth addressing.

I'm not suggesting AI will replace architects. The creative, compliance, and coordination roles that architects fulfil cannot be automated with current technology. But I do think AI will change what architects spend their time on.

Today, a significant portion of an architect's billable hours goes into preliminary concept development — producing the initial sketches and layouts that clients use to evaluate feasibility. This is relatively mechanical work: apply setback rules, maximise floor area, position units for solar access, ensure car parking compliance.

AI can do this faster and cheaper. Not better — faster and cheaper. The quality isn't there yet for formal submissions, but for the screening-and-concept phase, it's more than adequate.

What this means for architects is that the value proposition shifts from concept generation to design refinement, compliance navigation, and construction documentation. The parts of the job that require human judgement, local knowledge, and professional accountability become more important, not less.

For property investors and small-scale developers, the practical implication is that feasibility assessment becomes faster and cheaper. You can evaluate more sites, test more configurations, and reach go/no-go decisions in hours rather than weeks. That speed advantage matters in a competitive market where good development sites get snapped up quickly.

I'll be monitoring this technology closely and testing new platforms as they emerge. If a tool reaches the point where it can produce council-ready concept plans — with accurate setbacks, compliant parking, and proper services placement — that would be a genuine disruption to the pre-DA phase of residential development. We're not there yet. But we're closer than I expected.

How to actually use this technology today

Despite limitations, AI design tools have clear immediate applications:

Rapid site screening. When we assess ten properties a week, running each through AI before engaging an architect saves time and cost. Non-viable sites get eliminated in minutes 7.

Client communication. A rendered concept is dramatically more effective than verbal description.

Concept iteration. Trying different configurations — three large vs four small units, single vs double storey — takes seconds versus days.

What AI cannot replace: formal architectural drawings for DA, engineering certifications, Building Surveyor sign-off, planning compliance assessment, and on-site builder judgement 8.

The technology is a tool, not a replacement. Used correctly — as screening and communication aid — it's genuinely valuable. Used incorrectly — as substitute for professional design — it's a fast track to rejected applications.

I'm Yan Zhu. I test every new technology that might give our clients an edge. This one has genuine potential. But like every tool in property development, it works best in the hands of people who understand its limitations.

If you've got a site and want to know what can be built — properly — reach out. We'll give you the AI concept and the professional assessment. You need both 9.

One final observation about the state of AI in property development more broadly.

The design tool I tested is just one application of AI in this space. Other tools are emerging for automated property valuation (already in production at several banks), construction cost estimation (useful but still imprecise), and planning regulation interpretation (promising but legally risky).

The common thread across all of these tools is the same: they're useful for preliminary screening and directional guidance, but they cannot replace professional judgement for formal decisions. A bank's automated valuation model might give you a reasonable estimate, but your formal valuation for lending purposes still requires a human valuer. An AI cost estimator might flag that a project will cost approximately $450,000, but your actual construction contract will be negotiated by humans based on site-specific conditions.

The investors who will benefit most from these tools are the ones who understand this distinction: AI as a screening layer that makes human expertise more efficient, not AI as a replacement for human expertise.

We're integrating several of these tools into our workflow, always as supplements to — never substitutes for — the professional assessments that protect our clients' capital. The technology is exciting. The hype around it is dangerous. As always, the truth is somewhere in between.

I want to make one more practical suggestion for property investors who want to experiment with AI design tools.

Before you use an AI concept for any decision-making purpose, stress-test it against one simple question: can a human professional independently verify the key assumptions within 48 hours?

If the AI concept shows three units on a site, call the council planning department and ask whether three units would be permissible under the current zoning. That phone call takes five minutes. If the answer is yes, the AI concept has provided useful directional guidance. If the answer is no, the concept is misleading regardless of how professional it looks.

The danger with AI-generated design concepts is that they look authoritative. The renders are photorealistic. The layouts appear professional. A buyer or investor who doesn't understand planning regulations could easily mistake a concept for a compliant design and make purchasing decisions based on development potential that doesn't actually exist.

This is why I've categorised this article under 'Scam / Warning' rather than 'Technology.' The tool itself isn't a scam — it's genuinely useful when properly contextualised. But the potential for misuse is significant, particularly if AI design concepts are used in marketing materials or sales presentations without appropriate disclaimers.

As AI tools become more accessible and their outputs become more visually convincing, the gap between concept and compliance will become an increasingly important area of investor education. The investors who understand this gap will use AI as a productivity tool. The investors who don't will use it as a decision-making tool — and some of them will get hurt.

Be in the first group.