Lewisham Council, backed by MHCLG's PropTech Innovation Fund (round 4) and architecture practice RCKa, ran an AI-plus-GIS "Small Sites Finder" across the borough — trained on typical infill, backland and side-street parcel shapes within freehold boundaries, using Ordnance Survey data — and surfaced more than 3,000 sites with capacity for roughly 9,747 homes. Identification accuracy ran 31–76% depending on site type, and the council is explicit that officer review stays essential. The signal isn't Lewisham's number; it's that land discovery — historically dependent on officer memory and manual review — is becoming a structured, repeatable, borough-scale pipeline. Public-sector land intelligence, at the very front of the lifecycle. UK Construction Online →

TestFit — the feasibility platform behind about 3,200 deal evaluations a week — shipped MCP support, letting an assistant like Claude or ChatGPT drive the platform in plain language: pull the parcel, check zoning and FEMA floodplains, set setbacks, test townhome / garden / podium schemes, run a pro forma. In a watched first-look demo it saved 33 schemes on a live Tempe site for about a dollar in AI spend. The design choice is the story: the assistant only directs TestFit's deterministic engine (900-plus constraints at once) — it never generates the geometry or the numbers. CGO Laura Paciano: "We didn't build TestFit on AI. We built the thing AI needed to run on." This is the direct answer to last month's "LLMs can't do geometry": keep the model as controller, keep a verified engine doing the spatial math. Caveat — single first-look from an enthusiast newsletter; desktop-only, one week live. AI for CRE Collective →

In a candid interview, CSI CEO Mark Dorsey defended Dynamic Standards — the effort to turn MasterFormat, UniFormat and OmniClass into continuously-updated, machine-readable data that BIM, digital twins and AI-assisted spec tools can consume — while owning a botched rollout: members learned of new subscription pricing from a leaked partner website, not from CSI. "That was a mistake. We own that." A governance and pricing review is underway (member pricing now cited at $399 per firm, down from $699); a free two-level lookup remains. As specifications stop being static documents and become data, whoever controls that classification substrate — and how they price access — becomes AEC infrastructure. Dorsey's own caution is the honest part: "AI-generated work is only as reliable as the underlying information." Walls & Ceilings →

Buildots launched an Intelligence Lab publishing free construction benchmarks drawn from aggregated, anonymized global project data — an attempt to replace decades-old frameworks with current metrics. Early findings from lab head Amir Berman: across the dataset only 48% of active activities are on track in an average week (commercial 44%, industrial 43%); the final 20% of any activity eats 27% of its duration, meaning late-stage stalls are structural, not incidental; and data-center builds show a 20–50% gap between planned and delivered MEP work. Named adopters Ledcor and Pomerleau run it portfolio-wide so lessons compound across jobs. For an industry the digest keeps flagging as benchmark-starved, this is the gap starting to fill — vendor-run, but published openly and framed as a gut-check, not a sales sheet. Daily Commercial News →

Munich Airport (Germany's second-largest hub) is using Celonis process-mining to build a real-time digital twin of operations, starting with baggage handling — a system where one small disruption cascades into delays and cost. The model does predictive capacity management (pre-processing bags in lulls before peaks hit), tracks each bag's full process chain from check-in to carousel, and auto-alerts staff the moment thresholds break. CCO Jan-Henrik Andersson: "Celonis gives us the context we need to see how our airport actually runs in real time… and achieve meaningful outcomes from our AI deployments." Note the definition of "twin" here — not a sensor-and-geometry model of the building, but a live map of how work actually flows. For FM and operations, an asset's processes are becoming as instrumented as its physical plant. Computer Weekly →