Because you already have AutoCAD and Revit skills, I would not choose a replacement by rendering quality alone. Take one completed project and run it through each candidate workflow: import or rebuild the survey/base plan, record existing conditions, develop grading and drainage intent, create planting and material schedules, make a client revision, produce a 3D view, calculate quantities, and export the construction set. That test exposes handoff friction much faster than a feature list. Many practices keep accurate CAD/BIM for the base and construction documents, then add specialized landscape or real-time visualization tools only where they save time. First decide whether your revived work will be residential concepts, planting design, irrigation, or permit/construction documentation, because those need very different depth of terrain, plant data, collaboration, and local standards. Disclosure: this account belongs to a garden-planning software project, so I am deliberately not recommending our own product here.

I see so many things done with AI now. They are slick, very professional-looking and people love them. They are usually bad designs with no sense of scale and not-infrequent hallucinations, but the bar for nice-looking presentation is now much higher than it used to be.

That is the risk I would make the test project expose. I would add four pass/fail gates: the scale and dimensions survive every revision; measured, observed and assumed data stay visibly distinct; the software never invents grades, plant performance or regulatory clearances; and the final quantities and construction drawings can still be checked in the CAD/BIM source. AI can raise the presentation bar, but a polished view should be treated as a hypothesis until it passes those checks. If a workflow cannot preserve scale or show where a number came from, I would reject it however convincing the render looks.