Two years ago I was sceptical about AI design tools. Most of what I saw was Pinterest-style mood boards generated from prompts, with no understanding of how a room actually functions. I was wrong about how fast they’d improve. I was right about the limitations that still exist.

This is what I’m using as of May 2026, what I’ve tried and abandoned, and where I still draw plans by hand.

The tools I actually use

Three tools have earned a place in my workflow:

Modsy AI Studio - the rebranded version after the 2024 acquisition. Strong for residential interiors, particularly where you’re working with stock furniture from major retailers. The integration with IKEA, West Elm and a handful of European brands means the visualisations are accurate to actual products you can buy. Subscription is $89 a month and I justify it because the time saved on client visualisations is significant.

Spacely AI - this one earned its place for the speed of iteration. Upload a photo of an empty room, get six different design directions in under a minute. The quality of any individual render is below what I can do in SketchUp + V-Ray, but for early-stage client conversations it’s invaluable. We can throw away seven concepts and find the one direction worth developing.

Foyr Neo - more of a traditional CAD tool with AI assistance baked in. Used for measured drawings, materials lists, and the kind of construction documentation you actually need to build something. The AI parts are mostly time-savers - auto-suggesting layouts, identifying clashes, generating elevations from plans.

What I tried and dropped

I trialled half a dozen others. The ones I dropped fell into two categories.

The first was the pure-text-to-image tools (mostly built on top of Stable Diffusion or similar). Beautiful renders, completely useless for actual design work. The image of a kitchen the AI generates looks great. It also has 13 cabinet doors of inconsistent sizes, an oven floating six inches off the floor, and dimensions that don’t add up. Fine for inspiration, dangerous if a client thinks it’s a real proposal.

The second was the AI-assisted CAD tools that promised to “design your room for you.” The output was always generic - same trends, same neutral palette, same Scandinavian-coded approach. Anyone using these tools at scale is going to produce work that looks like everyone else’s work.

Where AI still falls short

A few honest observations from using these tools daily for a year now.

Lighting design is still bad. Every AI tool I’ve tested overpopulates rooms with light fixtures and gets the lighting temperatures wrong for the room function. Kitchens lit like art galleries, bedrooms lit like operating theatres. I redo all lighting design by hand.

Material choices are trend-driven. Ask any of these tools for a kitchen and you’ll get fluted oak, brass fixtures, terrazzo benchtop. It’s the May 2026 version of grey laminate and chrome from 2018. Following AI defaults will date your work fast.

Scale and proportion errors are subtle but constant. A sofa that looks right in a render but is actually too large for the space. Rugs that are too small. Pendants hung at the wrong height. You need to verify everything against actual measurements before committing to a design.

Local context is missing. None of these tools understand Australian conditions properly. Suggested fabrics that won’t survive a Sydney summer, finishes that won’t pass NCC requirements, materials that aren’t sold here. I always cross-check against local supplier availability.

How I actually use them in projects

Early stage: Spacely or Modsy to generate concept directions for client conversations. Speed matters more than accuracy here.

Design development: hand sketches, then SketchUp, with AI tools used selectively for furniture suggestions and for testing finish combinations.

Documentation: Foyr Neo or traditional CAD. AI assists but doesn’t lead.

Client presentations: a mix. The AI renders are good for “feel,” but I always include hand sketches and material samples so the client understands these are tools, not finished proposals.

The bigger question - what’s an AI tool for?

I had a useful conversation about this with the team at Team400, who do AI consulting work across a few industries. Their framing has stuck with me: AI tools should make you faster at what you’re already good at, not pretend to be good at things you’re not.

For a trained interior designer, AI tools speed up early-stage iteration, free up time for client work, and let you test ideas you wouldn’t have bothered drawing manually. That’s value.

For someone with no design training trying to design their own home, AI tools produce confident-looking output that might be terrible advice. The tool can’t tell them their layout doesn’t work, their material choices are dated, or their lighting plan will be a disaster. The output looks finished even when it’s wrong.

This isn’t a snobbery argument. It’s a real concern. I’ve had several clients come to me after spending months iterating on AI-generated room designs, only to find that none of it can actually be built as drawn.

What to expect through the rest of 2026

The tools will keep getting better at rendering. Whether they get better at design - actual functional, contextual, considered design - is a separate question. So far the improvements I’ve seen have been in surface-level quality, not in the underlying design intelligence.

For now, my advice is the same: use these tools to think faster, not to think for you. Generate ten concepts, throw away nine, develop the one that has something genuine in it. The judgement is still your job.

The Architectural Review ran a useful piece earlier this year on the divide between AI as a presentation tool and AI as a design tool. Worth reading if you want a longer view on where this is heading.