In September I published State of BIM 2026, a market briefing for people building, funding and buying AEC software. It ended with five things to watch over the next 18 months. Each came with a test, the specific event that would show it had moved.
The five were whether a BIM 2.0 tool reaches LOD 350 on a real project, how the major vendors respond to open IFC, who prices AI correctly first, who holds a building's data after handover, and whether a genuinely AI-native way of authoring buildings emerges.
Two months in, more has moved than I expected. Qonic added doors, windows and MEP to its components in October and launched its own AI agents in November. Snaptrude launched its agent at the end of October. Autodesk started charging for data leaving its platform.
Almost all of it is launches and pricing. None of it is yet the evidence the tests asked for.
1. LOD 350 on a real project
The test was whether Qonic or Snaptrude delivers a real project at LOD 350, the level where switching away from Revit becomes a practical question.
On paper Snaptrude has gone past it. Snaptrude Agent claims to take a project from site analysis to a construction-ready LOD 400 model, with structure, MEP and drawings, then back into Revit with existing families intact. In October Qonic extended the Components system it launched in August, adding doors, windows, furniture and MEP terminals that adjust their sills, reveals and lintels when they're moved, and previewed wall creation. Qonic says that's the point where a complete building can be modelled in the browser.
Neither company has named a project delivered that way. The caveat from September still applies. The edge cases that made Revit hard to replace turn up on the hundredth project, not the first.
Autodesk is in the same race from the other side. At AU, Andrew Anagnost suggested Forma could reach comparable architectural capability in about two years, according to AEC Magazine.
Where it stands: Closer on capability. No reference project.
2. The majors and open IFC
The test was whether any of the majors commits to IFC-first development or works with the open-source IFC stack.
Nothing on IFC. The openness the incumbents are adopting is MCP, the protocol that lets AI agents read and change a model. Autodesk open-sourced a Revit MCP server in June and made MCP part of its AI plans at AU. Graphisoft released Archicad MCP in beta this month, with an assistant inside Archicad and a public server that lets tools like Claude Desktop and Cursor batch-edit the model, each change only after the architect approves it. Qonic launched its own MCP server in November, and IFC export is one of the things an agent can do through it.
The access comes with a meter. Since 6 October Autodesk has charged for data exported through its AEC Data Model API, one Flex token per 5 MB beyond a monthly allowance. As Martyn Day put it, "an open protocol can still lead into a metered platform".
Where it stands: Moving, but not the way I framed it. The open layer is agent access rather than a file format, and it isn't free.
3. Who prices AI correctly first
The test was the first significant AEC product to charge for AI separately from access, in a way customers can understand. I expected a challenger to get there first.
All three challengers have. Motif launched with active-user pricing and AI charged separately, as I wrote at the time. Snaptrude's plans now come with AI credits, 400 a month on Pro and 1,000 per seat on Max, with top-ups. Qonic AI runs on tokens, with an allowance on paid plans and packs from €50 that don't expire.
Autodesk is starting with data and machine compute rather than AI. Alongside the Data Model API charge, it told AEC Magazine that every automation machine needs its own named-user seat. The commercial terms for its next Assistant are due in 2027. That looks like a seat with a meter on top.
Where it stands: Moved, challengers first. "Correctly" is the open part. Neither Snaptrude's pricing page nor Qonic's launch says what a typical task costs, so a practice can't yet budget for a busy month.
4. The lifecycle data relationship
The test was the first large contract, for a government estate, a health system or a big portfolio, that names a digital twin platform as its operational data layer for ten years or more.
I haven't found one. The movement is on the vendor side. Autodesk completed its $3.6bn purchase of MaintainX in August, the largest acquisition in its history, and put it alongside Tandem in a new operations division. The briefing mentioned the deal, but not under this signal, and this is where it belongs. It gives Autodesk a relationship with the people who maintain assets, which the briefing said design-tool companies lack. Much of MaintainX's data comes from factory floors, though, not buildings.
Where it stands: No owner contract yet. The vendors are moving faster than the owners.
5. An AI-native authoring architecture
The test was a demonstration, not an announcement: agent-native authoring on a real project, end to end, with no incumbent tool in the chain.
There have been more announcements. Motif Design launched in September. Snaptrude Agent works from site to drawings. Qonic AI puts agents inside the model, so you can ask it to find every window without a fire rating and give them one. Autodesk previewed an agent-first Assistant at AU, with rollout from 2027.
No one has shown a real project delivered that way without Revit or Rhino somewhere in the chain. Snaptrude's pitch ends with the model going back into Revit, families intact. That's sensible positioning and the same Catch-22 as before. The route to adoption still runs through the tool being replaced.
Where it stands: Plenty of claims. No end-to-end demonstration.
Each of the challenger launches also asks practices for the same input, which none of the five signals anticipated. Snaptrude Agent produces drawing sets to a firm's standards and builds missing components from manufacturers' data sheets. Motif's agents dimension the way you tell them to. Qonic AI checks and fixes properties against a rule stated in a sentence. The tools are converging. What will separate the results is the standards each practice can hand them.
I'll score the five again in February.