Henry AI (New York, YC 2024) closed a $16.5M Series A led by FirstMark, with Thomson Reuters Ventures as strategic backer, bringing disclosed capital to roughly $20.8M. The platform now claims 20,000+ client-ready CRE deliverables — offering memoranda, broker opinions of value, underwriting packages — tied to more than $150B in deal value across 150+ firms including Newmark and Berkadia. The number that should catch a developer's eye: at Grandstone Investment Sales, CEO Meir Perlmuter says broker-opinion-of-value production went from one analyst doing three or four a day to Henry covering all of it across seven brokers with no analyst; Henry reports 20%+ of customers have cut overhead as a direct result. The feasibility document that used to justify a junior seat is becoming a prompt. Metrics are company-stated. Upstarts Media →

In Scientific Reports (Nature Portfolio, Aug 1), a hybrid VAE-GAN framework generated early-stage residential landscape schemes conditioned on site boundary, building layout, and density, then scored each in about 6.5 seconds across microclimate, aesthetics, accessibility, and ecological metrics — using an NSGA-III search over a 128-dimensional latent space to surface Pareto-optimal options. Tested on three parcels in Changsha, China, the optimized schemes scored 6.2% above professional human-designed benchmarks, with the widest margin on the most constrained site. Read it as a signal, not a verdict: it is a single-author, single-climate study on landscape morphology — not building massing — with no independent replication yet. But it is a concrete, quantified claim that generative search now competes with the designer on a multi-objective early-stage problem, and the authors themselves frame it as a "divergent-thinking accelerator," not a replacement. Scientific Reports →

Architosh's Anthony Frausto-Robledo read Nemetschek's H1 2026 results (July 31) for the AI strategy under the numbers. The group behind Graphisoft, Allplan, Vectorworks, and the RISA and Frilo structural tools grew its Design segment to €142.9M (+9.9%), but the sharper signal is Bluebeam Max — pitched as the group's first comprehensive agentic AI suite for construction — and Architosh's read that the winning playbook is a premium AI tier bolted onto an existing subscriber base. For firms, that reframes the AI-budget question: the authoring and documentation tools you already license will increasingly meter their best AI behind a higher tier, and — Architosh's own caveat — a cooling construction market slows who says yes. Independent named-analyst coverage, not vendor copy. Architosh →

Komatsu and Washington, D.C.-based AIM Intelligent Machines (CEO Adam Sadilek) entered a strategic partnership (International Mining, July 31) to run autonomous bulldozers and hydraulic excavators — with AIM's fleets, per the announcement, already operating at U.S. customer jobsites. The integration ties Komatsu's Smart Construction planning and terrain platform to AIM's physical-AI layer so machines determine methods, plan routes, and execute earthmoving on their own; crucially, AIM's autonomy is retrofittable to existing fleets rather than a new-equipment purchase. A Japan rollout is slated for 2027 via Komatsu subsidiary EARTHBRAIN. The pattern worth watching: a major OEM endorsing "own the intelligence, keep the iron you already have" — the same retrofit logic behind this summer's solar and earthmoving robotics plays. No named third-party contractor or productivity figures yet. International Mining →

Computer Weekly (July 29) went inside King Abdullah Financial District in Riyadh, where CITO Ramez AlFayez describes 1.6 million m² and more than 100,000 assets monitored in real time, feeding a central platform that integrates AI systems including IBM Maximo for predictive maintenance and energy optimization across a 48-initiative roadmap spanning digital twins and IoT. The operative shift is cultural as much as technical — "predictive maintenance can replace reactive intervention," the same move Northern Arizona University's facilities team made at campus scale last week, here at district scale with a named executive on record. It is a smart-district story, not a single building, running on Vision-2030 capital most owners will not have. But the operating model — one connected data layer, problems caught before occupants feel them — is the post-occupancy target the rest of the industry is circling. Computer Weekly →