Brickbait
Thinking...
I originally wanted to write a piece about the future of my professional space—construction software. But, extrapolating ahead, I don’t know if that’s really going to work.
Let me try and explain why.
(Cue the usual AI doomerism; “it’s coming for all of us”. No surprises there.)
Which game are you playing?
Briefly zooming out to the scale of the wider tech sector might help give a sense of the direction we’re moving in. It seems like the big frontier tech companies are operating at a speed and level of sophistication well above that of most tech startups. They’re playing chess and we’re playing chequers.
Figma’s stock price should act as the canary in the coalmine. Once an unquestionable SaaS darling, Figma was a modern and indispensable UI design tool that was on the cusp of acquisition by Adobe for a (then) reasonable-sounding $20bn.
Now, its stock price lies in tatters. How could this be?
Earlier this year, I was a heavy daily user of Figma. I knew all the keyboard shortcuts and could speak in detail about the few limitations the platform still had. I had repeatedly optimised my component library so that I could rapidly draft beautiful and maintainable UI designs for the tech startup I was at. Nowadays, months go by without me even bothering to open it. I could just as easily forget that it exists.
The entirety of my design process takes place within Claude’s desktop app now. Not only can it rapidly generate designs exactly as I want them, but they are often fully working and interactive prototypes that are built using native web technologies, not Figma’s proxies of them. The machines have become tasteful and capable design collaborators.
In the short space of a few months, this workflow (now available in Codex, among others) has not only supplanted the likes of Figma, but is also rapidly cannibalising other, more modern competitors within the space (Base44, Lovable, etc). Last year, few people saw this coming.
Software on demand
Back to construction. I suspect 2026–27 will see many of the tech startups in my space gradually begin to wither and fail. Looking back, it’s possible that most of them simply started too late to make a huge difference to our sector. Many of these will disappear quietly—and a handful will not. But SaaS economics is what it is and the prognosis looks bleak.
What’s coming to replace them could well be software, geared to whatever the user wants to be doing, generated on demand. This isn’t theoretical: Google has already previewed exactly this with its experimental Disco browser. It’s powerful, and its single-use disposable software may well be the harbinger of things to come. There are already libraries of open-source and ready-made components (like That Open Company) just waiting to be assembled into apps.
These apps will come built with only the essential functionality you’ll need to carry out your work:
- If you’re reviewing BIM coordination against contractors’ drawings and your studio’s internal standards, the app will come ready-built just for you with these built in. It will understand how you work, know your past projects.
- If you’re undertaking land value analysis, it will have already scraped land values from the local area, giving you a map and the key numbers you need to make your initial decisions.
- If you’re applying for planning permission, it will know the local and national planning law down to every detail, analyse recent approvals and highlight any notable exceptions to the rules.
- I’ll assume that most manual CAD and BIM work is a foregone conclusion at this point.
Before long, the AIs might just skip most of this; they will simply make proposals and ask the user to review/sign off on them, akin to the viewer-only/commenting subscription tiers we currently reserve for clients. Looking at OpenAI’s latest work on Intelligent UIs, this type of workflow may be all we really need to resolve our designs.
Initially, we may see a brief spike in work and custom software development, as historically low supply meets decades of pent-up demand. We all know firms that are using embarrassingly antiquated tech stacks and still following process like it’s the 1990s. But in the wider timeframe, I see the frontier labs viewing 99% of us as middlemen carrying out work that exists in the moment between now and the stable end state. And nobody likes a middleman.
Endgame?
Eventually, it seems that AI will learn how to do the reviewer’s job, too. Reinforcement learning from human feedback has massively improved the latest generations of models, and manual review is ultimately just an RLHF task. Automating review doesn’t even seem like an unreasonable idea anymore.
Will it be able to make beautiful buildings that everyday people like?
Possibly, yes—you could interpret public engagement and expressed sentiment from social media posts as another form of RLHF. We could start seeing the built equivalent of clickbait slowly emerging over the next five or so years.
Shall we call it ‘brickbait’?