Every architect who has ever worked in a shared BIM model knows the specific dread of hitting Synchronize with Central and getting told that the lobby stair you just spent forty minutes redesigning is owned by a colleague who left this morning for two weeks in Bali. Not damaged, not deleted, just locked, because worksharing gave one person editing rights to that element and never gave anyone a way to ask for them back short of a phone call to a beach. The whole team can see the model. Only one person can actually touch that wall, and he is not checking Slack.

That kind of bottleneck, one person or one program owning a piece of the model that everyone else needs to work in, is close to the exact seam a new BIM platform is trying to cut out entirely, by making software agents actual participants in the model instead of visitors knocking from outside it.

Motif, the design software startup founded by two former Autodesk executives, announced on September 8 the launch of Motif Design, which it is calling the first agent-native building design software platform. The pitch, according to coverage of the launch and Motif's own framing, is architectural rather than cosmetic: in Motif, the user interface, the API, and the AI agents are all clients of the same underlying data model, so an agent does not sit outside the application sending it commands through a thin integration layer, it works directly against the live project data the same way a human user does. Agent actions are logged and reversible, the company says, and project data stays owned by the firm rather than being used to train outside AI models. Motif is also multi-model on the AI side, routing tasks to different engines, including Claude, Gemini, and OpenAI models, through its own layer for reasoning about objects and spatial relationships, rather than betting the whole platform on one vendor's model.

The founders' resumes explain why this is being pitched as a rebuild rather than a plug-in. Motif was founded in 2023 by Amar Hanspal, formerly co-CEO and chief product officer at Autodesk, and Brian Mathews, formerly Autodesk's vice president of platform engineering, alongside founding employees of Revit itself. That is the team responsible for a meaningful chunk of the software the industry runs on today, choosing to start over on the theory that a legacy desktop application built for one designer at a time was never going to become agent-native by bolting a chat window onto the side of it. The company raised 46 million dollars in seed and Series A funding, the seed round led by Redpoint Ventures and the Series A led by CapitalG, Alphabet's growth fund, in a round announced in January 2025. Hanspal has been consistent about the framing since Motif's first public product: architects are, in his words, still using twentieth century tools to design twenty first century buildings, and a building itself is thousands of interconnected decisions that most existing software does nothing to help anyone actually make.

What Motif Design ships with today is narrower than the full vision, which is itself a reasonable signal that this is a real product and not a slideware demo. The initial target workflows are interior fit-out and test fits, where the object coverage and packaged agent skills are already deep enough to be useful, along with firms doing Rhino-to-documentation work who want to carry Rhino geometry through to construction documentation without dropping into Revit purely as a drafting engine. The platform streams live from Revit and Rhino without exporting or converting files back and forth, and Hanspal has compared its packaged agent skills to AutoLISP, a way for a practice to script and automate its own specific methods rather than accept whatever workflow the software vendor decided was standard. There is no traditional named-user seat structure and no separate metered charge just to view or navigate a model, which matters for adoption in an industry that has spent years resenting per-seat AI add-on pricing on top of already expensive base licenses.

The competitive subtext here is hard to miss and worth naming directly. Autodesk, the company Hanspal and Mathews left, has been building its own agentic layer on top of Revit and its broader Forma platform, using Model Context Protocol servers so its Assistant can act on project data across tools. That is a coherent strategy for a company with decades of installed base to protect, but it is also, by definition, an agent talking to an existing application through an API rather than an agent that is native to the data model from day one. Whether that architectural difference matters to a working project team as much as it matters in a product pitch is exactly the kind of thing that will only get settled by firms actually running both approaches on real jobs over the next year, not by either company's launch copy.

The bigger pattern is the one worth watching past this specific product. The industry has spent the last two years bolting AI assistants onto software that was never built to host them, then working around the seams where the assistant's suggestions have to be manually pulled back into the tool of record. Motif's bet, like a growing number of agentic tools in adjacent parts of this industry, is that the seam itself is the problem, and that the winning products will be the ones where the agent was a first class citizen of the system of record from the first line of code, not the ones that added agents fastest to what already existed. That is the same premise BidForgeAI is building against on the bidding side: an agent is only as useful as its access to a firm's actual project data, not a chat window that has to be told everything twice.

Sources