DevCon 2025, Developer Guide, AI, AI, and AI

I am in Amsterdam today, participating in the DevCon 2025. I came here directly from Berlin, starting at 2 o'clock this morning. Long day, much coffee, much excitement and adrenalin. Here are some other topics I need to share:

DevCon 2025 Amsterdam

I am in Berlin on the way to the Autodesk DevCon 2025 in Amsterdam. My last Autodesk and my last month of employment at Autodesk. Looking forward to seeing what the new exciting times will bring for me personally and for the world at large. Above all, I'm looking forward to meeting you here today and tomorrow!

Shared Projects for Multi-Version Add-In

A thread that We already mentioned the Revit API discussion forum thread on the optimal add-in code base approach to target multiple Revit releases. It was resurrected by Jeff JeffroH Hpad Hornsby, who suggests:

I'm surprised more developers aren't utilizing shared projects. Shared projects seem to offer the greatest flexibility, in my opinion. Maybe not too surprised since shared projects came out less than 10 years ago. The AutoCAD API adheres better to established API design principles, particularly in not breaking existing code (changing signatures, etc.), and works really well with shared projects for multiple year releases. Centralizing your code in a shared project and using each year's project solely for build configurations, references, build events, and related setup seems easier and more efficient in my opinion. To demonstrate, I created the RevitSharedProject example of a Revit shared project. I took the 2020 ExtensibleStorageManager and SchemaWrapperTools (referenced by ExtensibleStorageManager) projects from the SDK samples and made them ready to build for 2020-2025 by adding conditional compilation symbols to separate code changes between the years. The only change is you will have to change build output path for the 2025 projects as the relative paths converted to full paths. To add a 2026 project:

This suggestion sparked some controversial discussion, with other developers disagreeing with this approach, so check out the original thread for the full view.

Many thanks to Jeff for raising this.

Revit 2026 Developer Guide

This year's Revit API Developers Guide is now available.

Revit API and SDK Documentation RAG

We also already mentioned converting Revit API help file to RAG for LLM and the corresponding Revit API discussion forum thread on convert Revit SDK documentation for local RAG (aka use it with a LLM).

Chuong Ho shared some new input on this as well:

I have done some with CHM files, cf. [RevitAPIDocGen][https://github.com/chuongmep/RevitAPIDocGen).

The revit-api-chms repository contains many Revit API CHMs for processing.

The CHM to Markdown converter provides another open source oiption for processing them.

Thanks to Chuong Ho for sharing this.

Revit-IFC DeepWiki

I recently mentioned that DeepWiki analyses all GitHub repos One nice Revit related example is provided by the Autodesk revit-ifc project, producing the revit-ifc DeepWiki.

RevitGeminiRAG

Most of the following is AI related, often in conjunction with the Revit API.

Ismail Seleit on linkedin shares: Really excited to share something I've been tinkering with: RevitGeminiRAG is now out on GitHub! This project started with a simple question: How can we give LLMs full access to the Revit API so they become specialized Revit software developers and interact flawlessly with Revit? RevitGeminiRAG is a Revit plugin that allows users to interact with their Revit models using natural language queries processed by Google's Gemini models. It leverages a RAG system to provide the LLM with relevant context from the Revit API documentation, enabling it to understand and respond to Revit-specific requests more effectively.

Yet Another Revit MCP

Paulo Giavoni on LinkedIn: Today I try a quick visualization technique for a client's Revit model using MCP Claude AI. Would this be the end of BIM modelers?

A2A and MCP Towards Implicit Programming

Nine minutes with Nate B. Jones (links, substack) on MCP, A2A, and the Beginning of the End of Explicit Programming:

My note on MCP: How I Think About MCP: A Practical Guide to Tool Use in AI Agents takeaways:

Quotes:

Summary:

Google’s A2A protocol and Model Context Protocol (MCP) mark a profound shift in AI architecture, moving us from deterministic software to dynamic, agent-based systems. MCP enables AI to understand and use tools without explicit instructions, while A2A allows agents to discover and collaborate with one another. This paradigm demands rethinking state management, security, and optimization strategies. It’s not just a technical evolution—it’s a redefinition of how we build and scale intelligent software. We are entering a world where we delegate to intelligence itself, not just code.

AutoGenLib On-The-Fly Coding

More automatic code generation tools – Vibe coding on steroids:

AutoGenLib is a Python library that automatically generates code on-the-fly using OpenAI's API. When you try to import a module or function that doesn't exist, AutoGenLib creates it for you based on a high-level description of what you need.

M$ and Google Code 30% AI-Generated

Already today, AI writes 30% of Code for both M$ and Google.

AI Masters Geoguessing

o3 beats a master-level geoguesser player, in spite of attempts to trick it by adding fake EXIF data.

Geoguessing

Neuronal Biology and AI Algorithms

AI can also help us better understand the functionality of real live neuronal biology, by learning algorithm of biological networks.

Induced Atmospheric Vibration

Finally, an interesting physics question that has nothing whatsoever to do with either Revit or AI, discussing what took down the Iberian power system and explaining What is "Induced Atmospheric Vibration"?