DCW Day 1 wrap: asset intelligence as the next discipline, an AI mythbust, and what to catch on Day 2

The most interesting new framing out of Day 1 came from Nick Hutchinson, co-founder of Glider, on the Digital Operations Stage. His argument: the next discipline for the built environment is not better information management - it's asset intelligence. The two are related and distinct. Information management is the discipline of the asset's record - what's there, what's been signed off, what the Golden Thread evidences for Building Safety Act and Gateway 2 purposes. Asset intelligence is the discipline of the asset's decisions - using the asset's own history, condition, obligations and trajectory as the thing decisions are made against. One makes the record auditable. The other makes it actionable.

The reason this matters in the room is that it changes the test you apply to vendors. A CDE that captures and structures information is useful but incomplete; the value layer above is whether that information can actually drive a maintenance decision, a refurb scope, a remaining-life estimate, a procurement choice, or a Gateway evidence pack. The pattern it sets up is interesting. Procore's CDE relaunch on Monday emphasised the record side - connecting data, citing sources, requiring human approval. Hutchinson's argument is that the deeper play sits one layer above, in the agents and models that read the record and recommend the next action. Both layers matter; they're different products and different procurement conversations.

Use this onto Day 2: Take Hutchinson's filter onto the floor. The intelligence test - what decision does this drive, on which asset, with what evidence - sorts serious products from polished demos faster than any other question.

At 15:00 on the Inspire Stage, Vicki Reynolds (Technical Director of Digital Estates at ONE Creative environments) and Dan Rossiter (Built Environment Sector Lead at BSI) ran the "Would I AI To You?" session - a format borrowed from the TV show, deployed seriously. They presented each other with AI-related scenarios from the built environment and tried to sort fact from fiction in front of the audience. The point of the format is that it forces the speakers to be specific. Vague "AI is transforming construction" claims fall apart under cross-examination; concrete claims (we used this model on this kind of document and got this kind of result) survive it.

If you missed it, the practical takeaway is the format itself. It's a useful exercise to run with your own team - pick a real AI use case and ask whether the headline claim survives a deliberately sceptical retelling. Most don't, which is precisely the point. Reynolds and Rossiter's framing pairs naturally with Hutchinson's asset-intelligence test: one tests whether a claim is real; the other tests whether the underlying product drives a decision. Together they're a clean two-question filter for the rest of the event.

For your team after DCW: Run a "Would I AI To You?" exercise on the two strongest pilot ideas in your pipeline. Whichever survives is the one to fund.

A quick scan of the rest of Wednesday that landed. Nemetschek's James Chambers ran the workforce panel - craft expertise, digital skills and hybrid collaboration, framed around the practical question of how a building company actually staffs itself for the next decade. Stephen Boyd MBE and Andrew Green presented "Universal data spine" on the Digital Operations Stage - the engineering case for a shared, interoperable data backbone across infrastructure systems. Tektome's Naoki Kitamura on the Tech Stage at 15:15 addressed the brutal demographic question: scaling BIM quality when experience is retiring faster than it can be replaced. Paul Drayton (Laing O'Rourke) hosted the afternoon main-stage discussion on what performance is genuinely possible through effective design-phase collaboration, with WSP, BDP and AECOM. And Suzanne Hill (AI for SMEs) ran "Will an AI-enabled construction industry still need main contractors?" on the Inspire Stage - a deliberately provocative title for a panel actually about how project structures shift when agents take on work currently done by admin and junior PMs.

If one thread ran through them all, it was succession and continuity - how do you keep a body of work, knowledge and decisions intact as systems get smarter and senior people retire. Asset intelligence is one answer. The workforce panel was another. The "Universal data spine" talk was a third. The mood music wasn't the breathless tech hype that DCW used to host; it was a calmer, more grown-up conversation about what to actually build now to be in good shape for 2030.

The thread: Day 1 was about continuity - how to retain knowledge, decisions and value as the tooling layer accelerates. Bring that lens to anything you walk into today.

The exhibitor floor was dominated by three pre-event news drops that landed across the past fortnight, all pointing in the same direction. Procore had the freshest of them, having relaunched its Common Data Environment on Monday with Datagrid-powered agentic AI coworkers built in, and chosen the UK and Ireland for the first wave before broader EMEA. The stand was busy - the conversation worth having there is the timeline and pricing for existing UK Procore customers to move onto the new CDE, and what Datagrid availability looks like once it leaves private beta. Bluebeam Max has been live globally since 19 May, embedding Anthropic's Claude inside Revu via MCP for Smart Overlay, Smart Review, sheet stitching and Magic Markups. The takeoff and markup demos are the ones to ask for, because that's where a UK QS team spends real document hours. And on Monday, Autodesk announced it's acquiring MaintainX, pushing Autodesk firmly into the Operate stage of the lifecycle - maintenance, asset data, inspections, work orders - for the first time.

Read together, this is the strategic story of the week. The three dominant AEC software vendors are racing to own the full lifecycle as the platform for agentic AI, rather than competing on point-tool features. Procore's CDE relaunch puts a stake in the record-and-collaboration layer with embedded agents and citation-backed responses. Bluebeam Max ships agents into the document-review layer. Autodesk's MaintainX bet extends Autodesk's Design / Plan / Make / Build portfolio into Operate. Hutchinson's asset-intelligence framing tells you where this all goes next - the layer above the lifecycle stack, where agents recommend the next action rather than just produce the next record. The defensive question for any UK contractor or consultant procuring AI capability over the next two quarters is whether to lock in deeper with one vendor's emerging lifecycle stack or to keep best-of-breed flexibility.

Worth doing this afternoon: Visit at least one stand from each end of the integration spectrum - one big-lifecycle vendor, one Start-Up Village exhibitor doing something narrower and sharper. Compare their answers to the asset-intelligence test.

While DCW Day 1 ran in London, NVIDIA's GTC Taipei keynote at Computex on Monday unveiled Nemotron 3 Ultra: 550 billion parameters with 55 billion active under a mixture-of-experts architecture, running at 300+ tokens per second, and topping the US open-weights leaderboards at an Intelligence Index of 48. The point is the licence as much as the score. It's genuinely open-weights, which means a serious organisation can run it on its own infrastructure, with project and client data never leaving the perimeter. NVIDIA also announced Vera CPU and the Vera Rubin platform for AI-native data centres (now in mass production), Cosmos 3 as the world's first open physical-AI omnimodel (relevant to robotics and reality-capture pipelines), and the RTX Spark personal AI superchip co-developed with MediaTek (128GB unified memory, ~200 TOPS on-device).

For UK AEC and infrastructure firms, the Nemotron 3 Ultra release changes the "local AI" calculation that's been circulating since spring. The blockers on full cloud-LLM adoption have been concrete: client confidentiality, PI insurance exclusions (covered earlier this fortnight), IP sensitivity on original design work, and the GDPR and data-residency conversations that keep getting harder. Until Monday, the open-weights options at frontier scale were limited and the genuinely frontier models were proprietary cloud APIs. Nemotron 3 Ultra is the first credible mid-frontier open-weights option for an on-prem deployment. The economics aren't trivial - running a 550B-parameter model needs serious compute - but for firms with the volume to justify it, the conversation has shifted from "is this even possible?" to "is it worth it?"

The discipline: If client-data sensitivity or PI carve-outs have blocked an AI pilot, get a quick technical review of what an on-prem Nemotron 3 Ultra deployment would cost and what it would unblock. The answer might still be "not yet"; it should be informed.

A useful contextual data point that landed Monday alongside the Autodesk × MaintainX news. The week ending 1 June saw five ConTech start-ups raise capital, three emerge from stealth, a new $85m fund close at Convective Capital, and two acquisitions print. The largest fresh AI-specific rounds were NavigateAI ($25m for an AI field copilot) and LightTable ($22m Series A for AI preconstruction). The categories getting funded - field intelligence, preconstruction automation, lifecycle ops integration - are precisely the categories DCW exhibitors are demoing on the floor. That's a useful answer to anyone in your business asking whether this is hype: the capital is being deployed in exactly the spots where the demonstrable products are showing up.

The closing discipline is the one I keep coming back to. One workflow. Defended quality bar. Human approval gate. And specifically after DCW, write it up in your own voice for LinkedIn - remember the slop-throttling point from a fortnight ago. Specific named demos, specific named conversations, specific named numbers. That's what travels in the algorithm now and that's what builds the personal credibility that wins the next pitch.

The deliverable for this week: A LinkedIn post by Sunday with two specific stand-name moments and one workflow you're going to pilot in June. Two photos, two questions, one workflow target - same discipline as Wednesday morning.