CAD when software is cheap to build

At a glance TL;DR

• AI is lowering the barrier to building and changing design tools, opening CAD to many more developers and approaches.
• Bespoke architecture involves discovering what should be built. Shared models give people and agents somewhere to develop and test those decisions together.
• Revit could remain useful while changing profoundly. That does not settle whether Autodesk will continue to dominate the working environment around it.

Creative freedom

CAD is entering a Cambrian explosion. AI is lowering the cost of building and changing software, opening the tools of architectural practice to many more developers, experiments and ways of working. Applications that once seemed like permanent fixtures are becoming much more contestable.

The consequences reach beyond having an assistant operate Revit. When software becomes cheap enough to build around a particular problem, the problem no longer has to fit the software quite so obediently. A practice can explore a tool that reflects how it designs. A small team can attempt an alternative working environment. An open community can develop possibilities that would never make it onto a major vendor’s roadmap.

The development effort that once kept so many alternatives out is becoming a weaker barrier. Existing geometry engines and open building-information formats provide foundations; agents help people build on them. Independent CAD projects already describe using AI to write much of their software while providing ways for agents to operate it. FundaCAD’s project documentation offers one small, concrete example in parts modelling.

This changes which ideas are worth attempting. A smaller market can justify an application, and a recurring problem within one practice can justify a tool. More specialised applications can appear, compete and combine. Some will extend familiar platforms; others will question why those platforms are organised as they are. The same pressure reaches the other software a practice uses to produce and manage its work.

How long, then, can CAD designed around human input survive before it becomes unrecognisable? And does a proliferation of alternatives mean companies such as Autodesk lose their position?

Architecture gives those questions an interesting complication: much of the work is about discovering what should be done.

A useful spectrum runs from deterministic tasks, where the rules and desired result can be specified, to ambiguous work, where even the definition of a good result is still developing. Updating a known set of dimensions sits towards one end. Deciding how a building should meet an awkward site, accommodate conflicting needs and feel to inhabit sits towards the other.

Repeatable buildings offer more scope for generation in a single pass. If the brief, site conditions, assemblies and acceptable solutions are already sufficiently resolved, much of the remaining work is applying them. A standardised scheme can therefore move much closer to something an agent produces from a compact instruction. The apparent simplicity of that instruction rests on decisions already made.

A bespoke building contains many decisions that have yet to become instructions. A client discovers what matters by seeing a proposal. A section reveals that two reasonable ambitions cannot comfortably coexist. A material choice changes the character of a room and prompts a different understanding of the brief. Information emerges through designing; the finished answer was never fully contained in the initial request.

More capable agents can participate in that discovery. They can propose, compare and help resolve uncertainty. But where people’s intentions develop through engagement with the design, producing a complete model quickly does not finish the conversation. It gives that conversation something more substantial to work with.

This is a strong reason to expect shared modelling environments to endure. People need somewhere to see what is being proposed, understand relationships and intervene precisely. Agents need access to the same evolving project. Plans, sections and models provide a common working surface for both.

Something recognisably like Revit could have a considerable life ahead of it in that role. A designer might reshape a space directly while an agent carries the implications through related drawings. Elsewhere, the agent might propose several arrangements for comparison. The balance could change repeatedly within the same project, as questions become settled and settled decisions are reopened.

Direct manipulation remains useful because drawing is itself a way of thinking. Moving a wall can be faster and more expressive than describing its intended position. Looking across a plan and section can expose a relationship that nobody thought to ask about. An environment that supports conversation, delegation and immediate manual adjustment gives that exploration more room.

Autodesk is already moving in this direction. Its September 2026 Revit technology previews introduced connections through which agents can modify models, including a limited set of operations with review of each change. That is an early step towards a different relationship with an existing application. Autodesk’s announcement

Over time, the familiar application could change substantially. Commands organised for a person to execute one at a time could give way to larger intentions carried out by agents. Interfaces could adapt to the question being worked on. New tools could be created around a particular design difficulty, then revised as the team understands it better. A familiar name and model might remain while the experience of working changes almost completely.

The case for a shared modelling environment is stronger than the case for any particular company owning it. Existing Revit projects, integrations and familiar ways of working give Autodesk advantages. Cheaper software makes it easier for competitors to offer alternatives around those same projects, especially where information can move between applications without losing its meaning.

Competition may also change what dominance means. Revit could remain widely used while more design exploration, automation and interaction happen through tools supplied by others. Alternatively, Autodesk could make its environment the most useful place for that wider ecosystem to develop. New entrants can start with different assumptions; incumbents can bring agents into environments where substantial work already happens.

For a practice, this makes adaptability increasingly valuable: being able to keep useful project information, introduce better tools and choose when to draw, delegate or develop a new operation. The name on the application matters less than how much freedom it gives the people doing the work.

The coming abundance of software expands who can shape the instruments of architecture. Revit may live a long time through that change. The expectation that a practice must organise its thinking around a largely fixed set of tools has a much less convincing future.

Sources

1. FundaCAD project documentation Retrieved 29 October 2026
2. Autodesk Revit MCP Server Tech Preview Launch, 10 September 2026 Retrieved 29 October 2026