ArcGIS CityEngine 2026.0 has finally landed, and Python 3 support is out of beta. It’s a solid release with meaningful improvements under the hood. For me, though, it lands at an interesting moment: right as AI-assisted coding tools have properly come of age.
So is CityEngine dying? No: that’s not really the question. It’s clearly still useful, and 2026.0 proves Esri are still investing in it. The real question is whether its role changes now that AI makes building your own custom tooling faster, cheaper, and more accessible than it has ever been.
I’ve been in this game long enough to answer that from experience rather than speculation. Back in 2010, I wrote a post called “Making the rules up as I go along,” about downloading a CityEngine trial and wondering whether procedural modelling could actually earn a place in a small practice’s workflow. Sixteen years on, I’m asking a strikingly similar question: just one rung further up the ladder. Procedural 3D modelling from a geospatial base was transformational for me and for what I do in geodesign. AI-assisted coding feels like the next rung of that same ladder.
I’ve never been an excellent programmer. Coding has always been a challenge for me, and what success I’ve had has come through sheer determination and brute force rather than deep technical understanding: searching the internet for something close to what I need, then copying, adapting, and bending it to my context. The tools I built in Jython for my own workflows, which eventually grew into a full GUI framework as CityEngine Pro Tools, took an eternity to write. Some might say I should have gone back and properly learned a programming language. Fair enough, but I’ve simply never had the time, and building things this way has always got me where I needed to go.
Which is exactly why AI-assisted coding, ‘vibe coding’ if you like, has been genuinely transformational for someone like me. Going from idea to working implementation in fifteen minutes, instead of hours of trial and error, changes what’s worth attempting in the first place. Is the resulting code always clean and textbook-structured? Mostly, yes. And honestly, coding is a personal thing: two people with different backgrounds will write the same tool differently. What matters to me is whether it works and whether it gets me further.
What AI coding has really done for me is open up ideas and workflows that were previously out of reach. Is it taking work away from anyone? As someone who works close to freelance, honestly, no: these are things I simply wouldn’t have built otherwise, and wouldn’t have had the budget to pay someone else to build for me.
Where I’d urge some caution is this: domain knowledge combined with AI is genuinely powerful. What concerns me a little is people outside a domain assuming they now understand it because an AI explained it to them. It’s a pattern our industry has seen before: someone arriving with a shiny new tool insisting it’ll save you time, followed by the rest of us explaining, patiently, that’s not quite how it works here. AI just makes that pattern faster and more convincing.
So what?
I understand CityEngine deeply, and I’ve had more ideas for improving my working life with it than I can count. Do those ideas reach Esri’s R&D team in Zurich? Sometimes. Do they always turn into changes? Not always: fair enough, we simply have different priorities, and that’s fine.
What Python 3 support has done is open up a whole world I’m still exploring. I’ve already built tools that integrate deeply with CityEngine: tools that analyse not just reports, but the 3D models themselves. There are moments where it genuinely feels like I could build my own procedural modelling tool from scratch with these AI tools.
But that’s the thing: I don’t think I’d want to. CityEngine’s procedural engine is the product of years of dedicated development, and it still does things a weekend of AI-assisted scripting simply can’t replicate. What’s changed isn’t whether I need CityEngine: it’s what I can now build around it. And that combination, of CityEngine’s engine and AI-assisted tooling doing the rest, is exactly where things get interesting.
More on that soon I’ll be writing about some new exciting tools I have made!
I agree with you that using AI tools makes things look easy and for users outside the domain that’s big attraction. Like with any other AI tool it can take a lot of effort to correct the output/code and it can be equally frustrating like writing the code from scratch and it does not quite work.
Python 3 if the biggest feature of 2026.0 and I am yet to explore and find a test project. I hope we will see some more examples from Zurich team and the community and the expansion of user base.