Construction AI: Autodesk shows a 2027 roadmap in Las Vegas, while the agents you can buy today are already sprawling ungoverned

Autodesk spent its University conference this week, 15 to 17 September at the Venetian in Las Vegas, previewing a set of agentic AI features across its three clouds: Forma for architecture and construction, Fusion for design and manufacturing, and Flow for media. Chief executive Andrew Anagnost pitched the whole thing as "project intelligence" stitched across the products. For construction the notable bits sit in Forma and Revit: a growing collection of construction agents aimed at manual review, plus Drawing Change Analysis, Drawing Compliance Review and a preview of the next Autodesk Assistant for AEC. Civil 3D also becomes the newest Forma Connected Client.

Read the announcements and a phrase repeats: previewed, not yet generally available, planned to begin rolling out in 2027. What that means on site is that the agent Autodesk showed reading your drawings and checking your compliance is, for most firms, a thing you'll see in maybe a year and a half. The direction is right. Checking drawings and cutting manual review is exactly where the time goes on a job. But a preview isn't a tool you can put in front of a QS next Tuesday, and I'm not sure a 2027 date does much for the person drowning in change reviews this autumn.

Here's the aside an editor might cut. The interesting story isn't the demo, it's the shape of the buy. This is the biggest name in the industry telling its customers the good stuff is coming, later, inside the same platform they already pay a lot for. That sets up the whole rest of today's brief.

Set the Autodesk preview next to what's already on sale and you get the clearest case yet for the category we've been naming: challenger construction software, the Monzos of construction software, AI-first platforms doing the actual work rather than the incumbents' feature list. constructionAI went into production back in March, built by a construction professional who says he's never written a line of code, and pointed squarely at the 98 per cent of UK firms that got priced out of enterprise software years ago. It sits at the same job as Autodesk's preview, reading drawings, checking the work, doing the admin, and you can start on it now.

Then look at what the incumbent buy actually costs. Procore's model charges a percentage of your construction volume rather than a flat fee, so the bill climbs as your turnover climbs. Users on the review sites report renewal rises of roughly 10 to 14 per cent a year, and one reviewer running a six-person remodelling firm said Procore's own team told them the product "wasn't designed for companies their size." Those are user-reported figures, not audited, so treat them as the direction rather than the decimal. Procore has been busy on AI too, adding "Digital Coworker" agent packs back in July. But an agent bolted to a platform priced as a tax on your growth is still a platform priced as a tax on your growth.

That's what the category is about. Not who has the cleverest demo, but who lets the person doing the paperwork start this week, see the price, and leave with their data if it doesn't work out. When the biggest name in the room answers "2027," the smaller lot answering "this afternoon" isn't a scrappy underdog story. It's just the better buy.

While the vendors show roadmaps, the quieter story is what firms have already switched on. A run of enterprise security reporting this month put hard numbers on it, and they're not comfortable. One widely cited figure has 88 per cent of organisations reporting a confirmed or suspected AI-agent incident in the past year, against just 24 per cent that have a dedicated AI security governance team. These are survey and vendor-reported numbers, so the exact percentages deserve a raised eyebrow, but the pattern lines up with what people are seeing: agents get spun up faster than anyone can track them.

The mechanism matters for construction because of how these tools connect. Every time a team wires an AI assistant into a SaaS app over the Model Context Protocol, the plumbing that lets agents reach your systems, it opens a new and often invisible path to data you're meant to be guarding. There's already been a live supply-chain incident, a booby-trapped MCP server that silently copied every email it processed to an outside address. Think about that against a Building Safety Act job, where the Golden Thread is supposed to be a controlled, evidenced record. An agent quietly reaching across your document store isn't a hypothetical governance essay. It's the sort of thing that turns up in a Gateway conversation you didn't want to have.

Picture the door to your site office. You wouldn't leave it open and hope. Wiring an agent into your project data is the same door, and right now most firms haven't checked who's got a key. The comparison only goes so far, but the point holds: know what the agent can reach, and know who can shut it.

Put the three together and a single habit falls out. The incumbents are previewing 2027 while the challengers ship this quarter and the agents already in the building run with almost no oversight. So the discipline is old and dull and it works: pay for the tool that does the job now, not the one promised at the next conference, and before you connect anything to your project data, write down what it can touch and who owns the switch.

The procurement filter: For any AI you're weighing this month, ask three questions and take the plain answers. Can I see the price without three sales calls? Can I start without a six-month rollout? Can I take my data and leave? If the answer to any of them is no, you've learned something more useful than the demo told you.