Construction AI: Autodesk puts a voice-driven RFI writer on your phone, and a housing giant bets on cameras

Autodesk's July 2026 Forma construction release landed this month with more than 70 product updates, and most of them are the usual housekeeping: sheet callouts that link across collections, project photos bundled into handover packages, big exports split into sensible zip files. Worth having, none of it front-page. The one that earns its place is Quick RFI Create on Mobile. The RFI author opens the app, speaks the problem in plain language, and Autodesk AI fills in the relevant fields for them to check before the RFI goes into the review workflow. On the phone, on site, boots still muddy.

That's a small feature with a real payoff, because the RFI is where a good site manager's day quietly disappears. You spot the clash, you know exactly what's wrong, and then you lose twenty minutes typing it up properly so the design team can act on it. If the AI drafts that for you off a spoken sentence, you've bought back the twenty minutes. The same release also puts Rooms and Area Detection into public beta, using AI to read location polygons straight off a drawing so an admin isn't tracing them by hand. This all sits on the Forma platform Autodesk assembled when it folded Construction Cloud in back in March, and it's the visible edge of what Autodesk's Mike Haley has called the base of an agentic ecosystem, the Assistant layer they've been building toward across Revit and the rest.

These are vendor-described features, so treat the pitch with the usual salt. The AI writes a draft; it doesn't understand your job. Run Quick RFI Create on a live project for a fortnight, read every draft before it goes out, and see whether it saves time or just moves the checking to the other end. My hunch is it's genuinely useful, precisely because it's narrow. It isn't trying to run the project. It's trying to fill in a form while you're still on the scaffold, and that's a job worth automating.

The procurement filter: if you're already on Forma, turn Quick RFI Create on for one team, one job, and measure RFI turnaround before and after. Don't roll it out on a promise.

On 2 July, L&Q signed a three-year agreement to deploy OpenSpace's visual-intelligence platform across its active development portfolio. L&Q isn't a startup dabbling. It's one of the country's largest housing associations, more than 105,000 homes, a landlord with a development programme big enough that portfolio-wide oversight is a genuine problem. The stated reasons are quality assurance, dispute resolution and construction information management. Read that again. That's the golden thread, described in the language a development director actually uses.

Here's why it matters beyond one contract. Reality capture, the ceiling-and-hard-hat 360 cameras that record what was actually built as it was built, has spent years being sold as a nice-to-have for progress photos. What L&Q is doing is treating it as compliance infrastructure. When a defect claim lands three years after handover, the argument isn't he-said-she-said any more. There's a timestamped visual record of what went in behind that plasterboard, and who was on site the week it did. For anyone who's sat through a dispute trying to reconstruct a site sequence from a WhatsApp thread and a few blurry phone snaps, that's the whole game.

And watch where the tech is being proven. OpenSpace announced on 1 June that it had passed 1,000 data centre projects, and its CEO Jeevan Kalanithi told Commercial Observer in July that half of those came in the last twelve months, as the AI race entered what he calls its physical phase. The firm quotes 41% fewer claims and five times more quality issues captured on that data centre work, which are vendor figures and worth checking against your own experience, but the direction is the point. The tools get hardened on the hyperscale jobs, where the schedule pressure and the money sit, then they move down onto ordinary housing sites. L&Q is the down-onto-housing part of that story, happening now.

For your board pack: the next time a defect or delay claim costs you six figures to defend, ask what a shared, timestamped visual record of the build would have been worth. That's the number that justifies reality capture, not the progress photos.

Put the two together and there's a pattern worth naming. Neither of this week's items is a clever chatbot you have a conversation with. One is a form that fills itself in from a spoken sentence. The other is a camera that remembers what got built. Both are narrow, both are boring, and both attack the exact places where construction leaks time and money: the RFI that eats a site manager's afternoon, and the dispute that can't be settled because nobody kept a record. That's where AI pays for itself in this industry, on the unglamorous edges, not in the headline.

The standing discipline hasn't changed. Check the AI's draft before it leaves your hands, because a voice-generated RFI with the wrong location reference is worse than no RFI at all. And own your record, because a portfolio of site footage is only an asset if you can get it out and read it when the platform contract ends. The data centre pipeline that's driving so much of this is itself a reminder of how fast things move and stall at once: mid-July reporting suggests 30 to 50% of the global data centre capacity planned for 2026 could slip on power and equipment, so the sector proving these tools is also the one most likely to teach you how a fixed date becomes a moving one. Buy the tool that works on your job today. Watch the ones the hyperscale builders are hardening, because they're coming your way next.

A practical step: pick one leaky process this quarter, RFIs or defect records, put one narrow AI tool against it on a single live job, and measure the before and after. One process, one job, real numbers. That beats a platform-wide rollout on a vendor's promise every time.