The Motif team published The Good Enough Intelligence this week. It is the clearest piece I have read on what AI in design tools should actually be doing - and the most uncomfortable for the people building them.
The argument, briefly. The most capable design intelligence is the one that exercises the most deliberate restraint. Not restraint as a limitation, but restraint as a design decision - a continuous judgment about what to resolve, what to surface, what to remember, and what to leave for the architect to work through. Borrowing Edith Ackermann: things to think with, not things that think for you. Borrowing Winnicott: a good enough parent that holds the developmental space rather than collapsing it.
The piece is definitely worth reading in full at this point of time in our industry.
I want to add one thing to it - from the vendor side of the desk, more than a decade ago.
The complaint I couldn't take seriously
Somewhere around 2012, we were selling ArchiCAD into a large practice. We were sitting with the senior leadership and the feedback turned to BIM more broadly. One of them said something I have not forgotten:
"We are extremely disappointed that ArchiCAD doesn't teach our graduates how buildings go together."
I was flabbergasted. To a large extent, I had to hold my tongue. It is a piece of authoring software. It is a tool. The complaint sounded like asking a hammer to teach carpentry. The practice's training programme was the practice's responsibility - not Cadimage's or Graphisoft's.
I still think that is partially true. But reading the Motif piece this week made it clear what the partners were actually getting at. They were not asking the software to be a teacher. They were observing that the software was, quietly, doing some of the teaching anyway - and not the parts they wanted it to do. The defaults, the wall-joint behaviour, the way components clicked together: graduates were absorbing those choices as the way buildings are, rather than constructing their own understanding of why they go together that way.
BIM did not remove the developmental process. But it did reshape it, in ways nobody was tracking at the time. And the practice could feel the result on the other side of the graduate intake - three years in, the people who had only ever worked in BIM had a different shape of intuition than the people who had drafted before they modelled.
The complaint was awkwardly phrased. It was also early evidence of the formation problem the Motif piece names directly.
What AI changes
The Motif essay is careful not to make the lazy version of this argument. The claim that AI tools remove specific skills or jobs is, as Lira writes, "inflammatory and misguided." That is not where the case sits.
The case sits here:
AI tools remove the developmental process through which architects learn and evolve their thinking. CAD did not threaten that either. […] Generative tools that front-run the reframe do.
ArchiCAD shaped how graduates built up an intuition for assembly. It did not absorb the reframe - the moment where the architect looks at a problem, decides it is the wrong problem, and reconstructs what is being asked. That moment is still done by a person, sitting with the brief, the site, the budget, the structural engineer's grumbling email, and several days of working through possibilities that turn out not to work.
A generative tool that produces a credible answer in thirty seconds takes the reframe with it. The architect never inhabits the possibility space the answer came out of. The graduate especially never does.
This is the line in the essay I keep coming back to:
It leaves room for the architect to struggle productively, because that creative struggle is where insight, exploration, and the design of great buildings live.
The 2011 partners were complaining, in clumsy language, that the software was not leaving enough room. Fifteen years on, the room is about to get a great deal smaller, very quickly, unless someone in the tool design process is making that a deliberate choice.
The paradox, and what it means for the people building these things
The Motif piece closes on a paradox that should be required reading for every AEC software founder shipping AI features this year:
Greater capability demands greater restraint. Fuller memory demands greater humility. These are not limitations of the system; they are its design.
This runs against almost every product instinct that has been rewarded in software in the last twenty years. The default move is to do more, automate more, remember more, decide more. The shipped feature is the one that closes the loop without bothering the user. Friction is the enemy. Resolve the conflict silently. Reroute the duct. Adjust the ceiling. Update the model.
The Motif argument is that the friction was the point. Not all of it - coordination friction is genuinely unproductive and AI should absorb that without hesitation - but the friction of working through a design problem is where the architect becomes an architect, and where the building becomes the building.
The harder design question for AI in AEC is not what to generate. It is what to deliberately not generate, when, and why. And then how to surface the things the system did decide in a way the architect can interrogate and overrule.
The measure of the tool is not only what it produces. It is also how it empowers the architect who uses it, and what kind of practice it makes possible across generations.
That is a measurement most product roadmaps in this space are not set up to evaluate. The good enough intelligence is a design problem, but it is also a commercial problem, and a recruiting problem, and an insurance problem. It will be hard to build. The companies that take it seriously now will look very different in five years to the ones that don't.
I'm writing this from London the night before NXT BLD. A lot of the people shipping these tools - and the people buying them - will be in the room over the next two days. This is the conversation I want to be having there.