The render is beautiful. Now what?
Your kids have moved out, or a parent is about to move in, or the family has outgrown a house you have loved for fifteen years. The place that once worked has stopped working.
So you did what most people now do first. You opened an AI tool, described the house you wish you had, and less than a minute later you were looking at an image of a kitchen opening onto a north facing deck, with morning light falling exactly where you wanted it.
That was a useful evening. It was not a plan.
If you came to this page expecting an architect to tell you these tools are rubbish, you will be disappointed. They are not. They are genuinely good at some things, and they have changed the first few weeks of a renovation project for the better. What they do not do is the part that determines whether your project gets approved, gets built, and costs what you expected.
Here is the honest version of both.
What AI is actually good at
Giving you vocabulary. Most people know what they dislike about their house long before they can describe what they want instead. Generating twenty variations is a fast way to find the words for it.
Getting a household to the same starting point. Two people can argue about "open plan" for a decade and mean completely different things by it. Looking at the same image is a shortcut through that.
Testing a feeling cheaply. Wondering what the room would be like with the ceiling lifted at one end? You can see a version of that in seconds rather than waiting for a sketch.
Asking better questions. Used as a research tool rather than a design tool, it is useful. More on that below.
None of this is trivial. Clients who arrive with a folder of AI images alongside a few pages torn out of the lookbooks are usually easier to work with, because they have already done the hard work of noticing what moves them.
What the image does not know
1. Your site
An AI render assumes a flat, empty, generic block. Yours is not.
Much of Melbourne's south eastern suburb belt sits on reactive clay. Under AS 2870, the Australian Standard for residential slabs and footings, a site is physically classified from A through to P based on how much the soil moves with moisture. That classification drives your footing design, and footing design drives a meaningful slice of your budget before you have chosen a single tap. A soil test is a physical act performed on your actual land by a person with a drill rig. No model can do it from a description.
The same goes for the fall across your block, where the sewer branch runs, and what is buried along your boundary. If there is an easement or a water authority asset near your proposed extension, that is not a detail to sort out later. Melbourne Water is explicit that a council or building surveyor cannot issue a building permit for work near its assets without written approval first. An AI does not know there is a drainage easement down your side boundary. Your title does.
2. The rules, which are local, and which move
This is the failure mode almost nobody anticipates, because the tool never flags it.
Two examples from the last eighteen months in Victoria:
On 8 September 2025, Clause 54 of every Victorian planning scheme changed under Amendment VC282, introducing new residential development standards and a deemed to comply assessment pathway for single dwellings.
On 1 July 2026, domestic building insurance in Victoria was replaced by the Home Warranty scheme, administered by the Building and Plumbing Commission, with the threshold for eligible work lifting from $16,000 to $20,000.
A model trained before those dates will answer questions about both with total confidence, and it will be wrong. It will not tell you it is out of date, because it does not know that it is.
Then there is the local layer on top. If your property sits in a heritage overlay, which a great many period homes across the inner and middle south east do, demolition and works you would reasonably assume are routine require a planning permit, assessed against the specific heritage citation for your area. Two houses on the same street can sit under different controls. Navigating that is the whole point of the permits and plans stage, and it is where a project either moves or stalls.
3. Whether it can be built, and for what
Anyone can draw a nine metre unsupported span. Someone has to price the steel that holds it up, and someone has to work out how that steel gets into a backyard with a 900mm side access.
The gap between a design that looks buildable and a design a builder will price without loading it with contingency is where renovation budgets quietly disappear. Buildability is not an aesthetic judgement. It is the accumulated knowledge of what trades charge for, what councils query, and which clever detail will cost you three weeks and a five figure sum to resolve on site. You can see how that resolves in practice across the completed projects, where the interesting decisions are usually the ones you would never notice.
4. Who is accountable
In Victoria, you cannot represent yourself as an architect or provide architectural services unless you are registered with the Architects Registration Board of Victoria under the Architects Act 1991. Companies and partnerships providing architectural services must be approved by the ARBV as well.
That registration exists so there is a person, with professional indemnity and a licence at risk, standing behind the drawings. An AI output has nobody behind it. When something is missed, the cost of the miss lands on you.
5. The brief itself
This is the largest one, and the least technical.
The part that is not technical at all
The brief you arrive with is usually not the brief you need.
"We need a bigger kitchen" is very often "we cannot have people over without one of us being stranded alone at the sink." Those are different problems. The first has an expensive answer involving a rear extension. The second sometimes has a much cheaper one involving moving a wall and reorienting a bench. The only way to find out which you have is to ask, and then to keep asking.
An AI tool will take your first sentence at face value and design exactly what you asked for. That is the whole problem. It is very good at answering the question. It has no reason to interrogate whether it is the right question.
Two people in a household rarely want the same thing, and they rarely say so directly.
A good deal of early design work is listening to the same room described in two slightly different ways by two people who live in it, noticing where those descriptions diverge, and finding the design that gives both of them what they actually meant. That is a conversation, and it takes more than one sitting. It is why the feasibility study runs across three sessions rather than one.
Knowing what to leave out.
Generative tools are additive by nature. Ask for more and you get more. A considered design is mostly a sequence of decisions about what not to do, and those decisions are the hardest to make alone, because every one of them feels like a loss until you see the result.
Judgement when the number comes back.
At some point a builder returns a price and it is higher than you hoped. The value in that moment is not more options. It is someone who has been through this many times telling you which three things to protect because they are the reason the project works, and which six to let go because you will not notice them in a year. That is experience under constraint, and it is the opposite of a generation problem.
Advocating for you.
With council, with the building surveyor, with the engineer, with the builder. Someone has to make the case for your design and hold the line on it across four months to more than a year. That is what the project management stage actually buys you, and the relationship is as much the service as the drawings are.
What to actually do with your AI images
Do not throw them out. They are useful. Use them properly.
Keep every one you liked and bring them to your first meeting. They are evidence of taste, which is hard to describe and easy to show.
Annotate why. "This one, because of the light coming across the bench" tells an architect something. "This one" tells them almost nothing.
Do the same for the ones you hated. Knowing what to avoid narrows a design faster than knowing what to chase.
Do not brief from a render's dimensions or materials. They are decorative, not derived from anything. Treat the image as a feeling, not a specification.
Use the tool for questions, not answers. "What are the common ways people bring light into a south facing living room in a period home?" is a genuinely good prompt. "Design my living room" is not.
So, do you still need an architect?
Not for everything, and it would be dishonest to pretend otherwise.
If you are doing a like for like bathroom, a deck, or a straightforward internal refresh with no structural change, a good builder and a competent draftsperson may be exactly the right call, and cheaper.
Where an architect earns the fee is where the variables multiply:
A period or heritage home where the façade needs to stay and the new work has to sit behind or above it
Anything that changes the structure, the roof line, or the orientation of how the house is used
Tight, sloping, or awkwardly oriented sites where the constraint is the design problem
Dual occupancy and subdivision, where planning drives everything
Any project where the real question is still whether to extend or to move
That last one matters most, because it is the question underneath almost every enquiry. An AI tool will happily design an extension for you. It will never ask whether you should be building one at all. If that is where you are, the renovate or relocate calculator is a better first ten minutes than another render.
The next step, and what it costs
The reason most people put off the phone call is not the design. It is not knowing what the whole thing costs, and not wanting to find out in a way that commits them to anything.
That is what a feasibility study is for. Three working sessions and a cost estimate, so you finish with a realistic picture of what your site allows, what your brief would cost to build, and whether extending or relocating is the better decision for your household. It is a fixed fee, and it is deliberately the smallest possible commitment that still gives you a real answer. The full outline of how it works is on the services page.
Bring your AI images. They are a good place to start the conversation.