The AI Tool Landscape

A list of AI tools is out of date the day you write it. A map of families is not, because the families change far more slowly than the products inside them. In 2023 the hot image model was one name; by 2026 it is several others - but it is still an image model, doing the same job in the same place in your process. If you learn 'image models turn text and reference images into pictures, and here is how I judge and use them,' you have learned something durable. If you only learn 'I click these buttons in this app,' you have learned something with a shelf life of months.

This is the same instinct behind Lesson 0.1: the unit of skill is the workflow, not the tool. The tool landscape sits one level below that. So rather than memorising products, we sort the whole field into six families by the job they do:

1  LLM assistants ....... work with language + reasoning
2  Image models ......... make and transform pictures
3  Modelling / BIM AI ... help build geometry + layouts
4  Visualization AI ..... turn models into convincing images
5  Analysis AI .......... predict performance + read data
6  Automation / agents .. chain the others into pipelines

Hold those six in your head and you can place almost any new tool you meet - and, just as importantly, spot the gap a tool is trying to sell you into filling when you do not actually have it.

It helps to notice that the anxiety is partly manufactured. Vendors and feeds profit from the feeling that you are behind, so they amplify every release. But a working designer does not need to have used the newest model to be excellent with AI, any more than a photographer needs the newest camera body to take a great photograph. What you need is a stable understanding of what each family does and the judgement to direct and check it. That understanding is what this course builds, and the map below is its foundation - a picture calm enough to keep on your wall for a year without it going stale.

LLM assistants are the everyday workhorses: ChatGPT (OpenAI), Claude (Anthropic) and Gemini (Google) are the three most designers will meet, with several capable open alternatives. They work with language and light reasoning - summarising a long code document, drafting a spec or an email, interrogating a brief, brainstorming, restructuring your messy notes into a clear scope. They are the subject of all of Module 1, and they are the best place to start because they need no special setup and reward clear thinking over technical skill.

Image models turn a text prompt (and often a reference image) into pictures. Midjourney is prized for its aesthetic polish; Stable Diffusion is open, local-capable and endlessly customisable; Adobe Firefly is built into Creative Cloud and trained with commercial use in mind. For a designer these are concept and mood engines - widening early exploration, testing a material palette, restyling a room in seconds. The deep dive lives in Module 3 and in the sibling course Generative AI for Architecture & Interiors. A word of honesty that recurs across this course: image models render plausible pictures, not buildable ones. A gorgeous AI facade may be structurally nonsense, and the columns may not line up floor to floor. That is fine for ideation and dangerous if mistaken for a design.

How do you choose within these two families? For LLMs the three leaders are close enough that the honest advice is to try two on your real tasks and keep whichever you think alongside more naturally; they differ more in feel than in raw ability, and capable open-weight models exist if you need to run one privately. For image models the split is clearer: Midjourney for fast, beautiful concept imagery; Stable Diffusion when you need control, local running or customisation; Firefly when commercial-use clarity and Creative Cloud integration matter. In every case, judge a tool on your work, not on someone else's demo reel - the outputs that go viral are curated from many failures you never see.

Modelling and BIM AI help you build and arrange geometry. This is the youngest and least mature family, so temper expectations: it includes text-to-3D (a prompt into a rough mesh), generative layout tools that propose plans against rules you set, and a growing crop of AI plugins for Rhino, Grasshopper and Revit that automate tedious modelling steps or suggest options. Useful for speed at the early, exploratory end; still firmly something you supervise and rebuild properly downstream. Module 4 covers it, and the parametric side connects to Computational & Parametric Design in the Academy.

Visualization AI takes something you have already modelled and makes it look convincing - diffusion applied to a rough viewport or clay render to produce a photoreal image in seconds, plus AI upscalers and detail and style tools for post-production. It is one of the highest-payoff families for practice, because it collapses render times from hours to moments; Module 5 is devoted to it.

Analysis AI points machine learning at performance and data rather than pictures. Autodesk Forma and similar early-design tools give near-instant feedback on sun, wind, noise and daylight; surrogate models approximate a slow simulation (like energy or CFD) fast enough to explore hundreds of options; and general data tools read a spreadsheet of survey responses or a schedule and find the pattern. Because its output looks authoritative - a number, a heat-map - this family demands particular care about accuracy, which Modules 7 and 9 take up.

A useful way to hold these three together: modelling AI helps you make the thing, visualization AI helps you show it, and analysis AI helps you test it. Their maturity in 2026 varies sharply, and it pays to be honest about it. Visualization AI is the most reliable and immediately rewarding - the results are genuinely usable today. Analysis AI is real and valuable but must be treated as fast approximation, not verified simulation, especially anything you would put in a report. Modelling and BIM AI is the least settled of all: impressive in demos, patchy in daily practice, and rarely producing geometry clean enough to build on without substantial rework. Match your expectations to that gradient and you will be pleasantly surprised rather than repeatedly let down.

Automation and agents are the family that ties the others together. Instead of driving each tool by hand, you connect them - via APIs, custom GPTs, prompt libraries, or software 'agents' that can take multi-step actions - so a whole sub-task runs as a pipeline: pull the brief, summarise the code, draft the spec, format the schedule. This is the most powerful and the most easily over-trusted family, because an agent removes you from the loop precisely where you most need to stay in it. Module 8 covers it carefully, and the caution that governs it is the theme of Lesson 0.4.

Now, how do you keep your footing while all six families churn? Separate what changes fast from what changes slowly. The fast layer - model versions, prices, feature names, which product is 'best' this week - churns every few months and is not worth memorising. The slow layer - the six families, what each is for, how to prompt and how to evaluate - changes over years and is where your skill compounds. Stand on the slow layer. Sample the fast one lightly: follow one or two trustworthy sources, try a new tool only when it plausibly fills a real gap in your workflow, and let the rest wash past. You are not falling behind by ignoring 90 percent of the noise; you are staying sane. Module 10.3, Staying Current, turns this into a sustainable habit.

When a genuinely interesting tool does cross your feed, you can size it up in about ten minutes with three questions. First, which family does it belong to - and do I even have a gap there? Second, does it do that family's job clearly better than what I already use, on my kind of task, or just differently? Third, what does it cost me in money, learning time and data risk to adopt it? Most tools fail one of those three and you move on with a clear conscience; the rare one that passes all three earns a trial. That small, repeatable filter is what turns the overwhelming toolscape into something you calmly manage rather than something that manages you.