Investors are moving faster than the AEC software industry. As AI systems start to absorb professional skills at scale, are AEC’s software, labour, and business models quietly heading for structural change?
Yesterday, Autodesk’s stock dipped. So did Bentley’s. Not because of earnings, but because AI software company Anthropic announced something that had nothing to do with AEC: packaged legal skills built directly into Claude, the AI assistant.
Investors connected dots that most of our industry hasn’t yet. A general AI system just upskilled itself in a professional domain and shipped that capability to millions overnight. If law, why not BIM model coordination? If contracts, why not specifications? If due diligence, why not design review?
For the markets, this wasn’t incremental progress. It was a category shift — AI stepping out of the toolbox and into the job description.
Things are moving so fast that what we assumed we’d never see in our lifetimes is now happening over lunch. And most of AEC is still arguing about whether to upgrade from Revit 2022.
The industry reaction matters for AEC, even though AI tools in architecture and construction are only just appearing and remain far from mainstream. Most practices are still experimenting at the edges — testing copilots, generative design, automation in narrow workflows. Safe bets. Contained experiments. Innovation theatre.
But investors aren’t pricing where AI is today. They’re pricing where capability appears to be heading, and how fast it could scale once it clears the hurdles of trust, liability, and usability. They’ve done this maths before. They remember what happened to Kodak, to Blockbuster, to every industry that assumed disruption would politely wait for them to be ready.
Investment firms are now actively warning long-horizon software investors about near-term risk exposure from AI. Not because companies are weak — but because even well-built, mature software businesses can now be structurally destabilised at speed. In an AI acceleration environment, durability no longer guarantees survival. Firms that spent decades building defensible positions can, in theory, be wiped out in a single capability shift. The risk profile of “safe” software has fundamentally changed.
The concern isn’t that software disappears overnight. It’s that its entire value proposition starts to rot from the inside. Much of the existing software economy is built on selling licences and services to support human-led processes. As AI skills improve, the link between time spent in tools and value delivered doesn’t just weaken — it breaks. Investors see a future where firms buy fewer specialist applications because more work can be executed directly from intent. It’s like going from a hammer and chisel to a 3D printer overnight.
Why pay for a tool when you can describe what you want and have it done?
For AEC firms, the immediate signal is trajectory, not replacement. AI is beginning to move into our space — and it’s not starting at the edges. Startups are targeting the high-value tasks: coordination, clash detection, optimisation, reasoning across constraints. The unglamorous, expensive, expertise-heavy work that has traditionally justified complex software stacks and senior salaries.
Even if adoption is slow, the direction is unmistakable: expertise is migrating from people into systems. The twenty-year veteran’s pattern recognition? Trainable. The coordinator’s ability to spot downstream clashes? Automatable. The specification writer’s institutional knowledge? Extractable.
This isn’t science fiction. It’s a funding deck being pitched right now in San Francisco.
The knock-on effects are already visible if you know where to look. Firms are becoming more cautious about expanding headcount — not because demand is falling, but because output is decoupling from labour. The old equation was simple: more projects meant more people. That equation is starting to wobble.
At the same time, practices are starting to question whether adding more software licences really adds proportional value. Especially when emerging AI tools can sit across multiple workflows and automate not blindly, but with judgement. The per-seat model starts to look quaint when one system can do the work of ten seats.
Insourcing. Get comfortable with the word
With AI-assisted development, firms can increasingly build targeted tools and workflows in-house — tuned to their own standards, their own data, their own hard-won lessons. No more waiting eighteen months for a vendor to half-implement a feature request. No more bending your process to fit someone else’s idea of how you should work.
The question that should keep software vendors awake at night: who understands the AEC problem better — the customer who lives it, or a product team that’s never set foot on a job site?
This isn’t about becoming software companies. It’s about taking back control. Vendors have held the keys to capability for decades. That lock is starting to look pickable.
For traditional software vendors, the implication is existential. Much of their value has come from monetising friction — interfaces, workflows, the coordination tax of getting humans to work together through tools. They didn’t sell outcomes. They sold process. And they charged handsomely for every click along the way.
As AI matures, that layer doesn’t just come under pressure. It gets bypassed entirely. Software won’t vanish — but value will migrate toward platforms that enable execution, integration, and real-world responsibility. The winners won’t be the ones with the most features. They’ll be the ones who figure out how to stay useful when the user can just speak what they want.
The disruption in AEC is still early, uneven, experimental. This messy, fragmented, uncoordinated corner of the world is hard to tame, so there’s plenty of room for denial. Plenty of time to pretend this is hype, that construction is different, that regulation will save us, that clients will never trust AI with real decisions.
Investors have already done the maths. Capital doesn’t wait for consensus. It prices in probability — and acts.
The question is not whether your software vendor survives the next five years. It’s whether they’ll still have anything to sell you that you can’t develop in house with apps on demand. And whether they’ve figured this out yet.
For those who want to track this shift rather than be blindsided by it, we maintain the AEC AI Directory — a live catalogue of AI tools entering the built environment space.
If this topic interests you, I also invite you to join us in London at NXT BLD 2026 on 13 / 14 May, where we will continue our BIM 2.0 charge, our support of start-ups and will dig deeper than any other conference on the Agentic future of AEC.
