Construction AI: the productivity case gets formal, and Google I/O lands tomorrow

EY released "The intelligence layer: how agentic AI can connect the infrastructure industry" - a report produced with the University of Cambridge (Laing O'Rourke Centre for Construction Engineering and the Centre for Smart Infrastructure and Construction), NVIDIA and Futurity Systems. It reviews more than 100 construction productivity studies and concludes that agentic AI is the most credible route to closing the sector's chronic productivity gap. Headline framing: agents act as an intelligence layer connecting siloed BIM, ERP, scheduling and sensor systems, surfacing decisions in real time rather than waiting for human triage. The report puts global infrastructure investment need at US$140tn by 2050, against a current US$64tn funding gap, and reiterates the well-established 15 per cent rework-cost baseline.

For UK firms, the Cambridge endorsement is the part that matters. This is the report you can put in front of an infrastructure client board, a procurement panel or an internal finance director without having to defend the source.

Houzz published its first UK State of AI in Construction and Design report this month. Early adopters are reporting savings of around 3 hours/week and a £23,000 annual productivity uplift per practitioner. The highest-adoption use cases are administrative work (79 per cent of pros overall, 89 per cent construction, 88 per cent design), sales and marketing (52 per cent overall), and planning and design tools to visualise projects and generate floor plans (33 per cent). Three in five respondents believe AI will transform the industry within five years.

The single-number claims are self-reported and should be quoted with the appropriate caveat. The directional signal - that small-to-mid UK practices are now reporting measurable benefit - is consistent across multiple UK surveys this quarter.

UK construction robotics startup All3 closed a $25m seed in late April led by RTP Global, with SuperSeed, Begin Capital, s16vc and VNV Global participating. The pitch is end-to-end: AI generates building designs tailored to a specific site, robotic factories produce custom parts off-site, and mobile robots - including Mantis, a four-legged autonomous platform with interchangeable tools - handle installation, fastening, finishing and inspection on site. All3 claims potential for up to 50 per cent reduction in development time, 30 per cent cost reduction and 25 per cent lower carbon impact, with initial commercial deployment on active sites in Germany.

For UK off-site, MMC and modular contractors, this is the home-grown alternative to the US/Asian humanoid pipeline. The fundraise is recent enough to be a leading indicator of where UK venture money on physical AI is going, and the technology choice (legged robots + AI design) is meaningfully different from the dominant humanoid bet.

Google I/O 2026 kicks off at the Shoreline Amphitheatre on Tuesday 19 May with the main keynote at 18:00 BST, and the developer keynote at 21:30 BST. Confirmed agenda topics include Gemini, Android 17, Chrome, Cloud and "more". Expect Gemini 3.5 Flash and Pro to be formally unveiled (leaked checkpoints surfaced last week), VEO 4 / "Omni" video model territory to be clarified, and Android XR for smart glasses to land ahead of Samsung's Galaxy Glasses at Unpacked in July.

For construction-tech leaders making AI tooling decisions this week, the practical recommendation is simple: hold any non-urgent AI commitments until Wednesday evening. The shape of the Google offering for the next twelve months will be substantially clearer by then - and several of the leaked items would directly change the right answer for video, multimodal and Android-tablet deployment choices on site.

xAI launched Grok Build on 15 May - an agentic CLI for professional software engineering that puts the company into direct competition with Anthropic's Claude Code and OpenAI's Codex. Early-beta access is via SuperGrok Heavy ($300/month) with an introductory deal at $99/month for the first six months. The technical pitch is credible: a plan mode that produces a written plain-English plan before touching files; up to eight concurrent specialised sub-agents that plan, search docs and write code in parallel; and native compatibility with the Anthropic skill format - drop ~/.claude/skills/ into Grok's working directory and the skills work unmodified. It also discovers MCP servers and exposes them as tools.

New cybersecurity-benchmark results circulated this week show Anthropic's Mythos model out-performing OpenAI's GPT-5.5 on capture-the-flag-style exploit tasks - the same offensive-security workload class that underpins Anthropic's Project Glasswing and OpenAI's Daybreak defender initiatives. The signal worth taking away is not the single head-to-head result; it is that frontier evaluations are moving from synthetic puzzles to genuine offensive-security workloads, and the gap between the leading models on these tasks is now measurable.

For construction firms, this matters in two ways. First, the "AI defender" tier each frontier lab is offering is being shaped by these evaluations - and your eventual access to enterprise-grade defensive AI depends on which lab's ecosystem you sit in. Second, the same capabilities that score well on defender benchmarks are, structurally, the capabilities that score well on offensive ones. Tighter agent permissioning, audit trails and least-privilege design remain the unchanging operational priority.