Buildots Cuts Delays by Up to 50%: AI Finally Sees the Superstructure

Buildots has made superstructure tracking generally available after a year of live-site testing. Drone and 360-degree camera imagery is now processed into structured progress data linked directly to the project BIM, closing the visibility gap that has historically relied on physical site walks and incomplete snapshots. This week also saw the frontier model wars intensify: OpenAI, SpaceXAI, and Meta all released new models with aggressive pricing. Anthropic's $50bn Fluidstack build-out, announced last November, has sites coming online through 2026. The construction sector is moving from watching AI from the sidelines to building on it.

The AI infrastructure boom is reshaping construction demand. Data centres are now the second most in-demand construction sector globally, with over 70% of markets reporting tight contractor capacity. Turner & Townsend's latest report flags the skilled-labour pool as the headline risk for delivery teams. Meanwhile, on the software side, Procore is adding AI to BIM coordination, Build is selling completed due diligence work instead of licences, and CMap is putting natural-language querying over project financials. The pattern is clear: AI is moving from the lab into the delivery workflow.

Buildots has launched superstructure-tracking capability, extending its AI-driven construction-monitoring platform into the structural phase of projects. The feature, which has been in beta testing with customers on live sites for over a year, is now generally available.

Tracking progress during superstructure has historically relied on physical site walks, an approach that leaves significant portions of a project unmonitored due to safety restrictions, access constraints and the sheer scale of structural works. The result is that critical structural decisions are often made without objective progress data. The new capability uses imagery captured via drones and 360-degree cameras, which Buildots' AI then processes into structured data linked to superstructure elements and quantities within the project BIM. This gives teams automated progress analysis across the structural phase using the same platform they already use for underground utilities and fit-out, maintaining a continuous data record across the project lifecycle without switching systems.

Read the full story

OpenAI publicly launched GPT-5.6 on 9 July in three tiers: Sol ($5/$30 per M tokens), Terra ($2.50/$15) and Luna ($1/$6). OpenAI claims Sol tops the Artificial Analysis Coding Agent Index at 80, 2.8 points above Fable 5, and calls it its strongest cybersecurity model yet. GPT-5.6 is also now the preferred model in Microsoft 365 Copilot, a notable twist in the same week Microsoft was reported to be swapping its own MAI models into Copilot to cut costs. For AEC firms buying AI by the token, the pricing and performance gap matters: Sol delivers competitive coding performance at a fraction of the cost of alternatives.

Read more →

SpaceXAI (the rebranded xAI) launched Grok 4.5 on 8 July, its first joint model since the $60bn Cursor acquisition. Trained on tens of thousands of GB300s and priced at $2/$6 per M tokens with roughly 2x token efficiency and 80 tok/s throughput, it targets office work directly with Word/PowerPoint/Excel plugins. This is a serious price/performance challenger for AEC firms buying AI by the token. EU access is expected within days.

Read more →

Meta released Muse Spark 1.1 through a new Meta Model API, its first paid model offering and a pivot from open-weights. The model excels at agentic coding and computer use with a 1M context window and parallel subagents, priced at $1.25/$4.25 per M tokens—aggressively below OpenAI and Anthropic. On Meta's JobBench, Muse Spark scores 54.7 vs Opus 4.8's 48.4, delivering strong performance on agent tasks. For AEC teams evaluating cost-per-capability, the pricing and benchmark matter.

Read more →

Anthropic rolled Claude Cowork out to web and mobile (Max beta first): start a task at your desk, check it from your phone, and scheduled tasks now run with no device online. Anthropic's own usage data says 90% of Cowork use isn't coding; business ops and content creation dominate. That's the work around the work that fills a project manager's week. For delivery leaders, the practical implication is that AI agents are now available for the background work that doesn't fit in one sitting.

Read more →

Procore adds AI to BIM coordination issues and natural-language reporting: Shipping 21 July, Procore is adding AI that drafts BIM coordination issue titles and descriptions from text, photos or linked docs, plus natural-language queries for ad-hoc reports.

Build raises $8.5m seed for AI infrastructure due diligence: London/New York startup Build applies multi-agent AI to site sourcing, technical due diligence and power assessment, claiming 95%+ reduction in DD timelines across 100+ projects in 15 countries.

CMap adds AI chat interface to professional services automation platform: CMap, PSA software used by architecture, engineering and consultancy firms, added an AI chat interface for natural-language querying over project financials, resourcing and pipeline data.

Meta commits $9bn to first Canadian data centre; 3,000 construction workers at peak: A 1GW facility in Alberta costing approximately $9bn, a 2-3 year build employing 3,000+ construction workers at peak, is more evidence that hyperscaler capex is now a core construction-demand story.

Anthropic commits $50bn to US data centres in Texas and New York: Anthropic announced a $50bn investment in American AI infrastructure with custom data centre facilities built with Fluidstack to meet Claude demand, confirming Anthropic is now a major infrastructure client in its own right.

Five stories worth reading before you get back to delivery:

Anthropic's Orchestrator Patterns: ~92% of Fable's Performance at ~63% of the Cost

Anthropic published patterns for pairing Fable 5 as planner/advisor with cheaper executor models, showing how the advisor setup retains ~92% of Fable's SWE-bench Pro score at ~63% of cost. Read →

Anthropic Launches Reflect: A Personal Claude Usage-Insights Dashboard

Reflect shows users their Claude usage patterns over 1, 3, 6, or 12-month windows, including topics, habits, quiet-hours and break nudges, plus suggested skills. Useful for teams thinking about how staff actually use AI. Read →

Anthropic Finds 'J-Space': A Hidden Internal Workspace Where Claude Reasons Silently

Anthropic interpretability research identified an emergent internal 'global workspace' in Claude where multi-step reasoning happens unwritten. Editing it flips answers, wiping it breaks reasoning, and deception-related concepts fire there before an answer appears. The strongest interpretability result of the year so far. Read →

Cursor Publishes Two Years of Usage Data: 40% of Developers Accept AI Commits Without Review

Cursor's usage report shows power users generate 10x the median lines of code, 90% of tokens are input tokens, and the share of developers accepting AI commits with no manual review jumped from 10% to ~40% in a month. For delivery leaders, the QA implication is clear: verification, not generation, is becoming the bottleneck. Read →

OpenAI Launches GPT-Live: Full-Duplex Voice That Listens While It Speaks

OpenAI's GPT-Live voice models power a new ChatGPT Voice that listens and speaks simultaneously, with interruptions, backchannels and live translation. Rolling out globally as the default voice for Go/Plus/Pro users, with API access to follow. Hands-free AI that actually converses opens up site and field use cases. Read →

Alongside GPT-5.6, OpenAI launched ChatGPT Work, an enterprise workplace companion that gathers context from your team's tools and turns scattered notes and drafts into finished spreadsheets, docs, slides and live dashboards. Available on desktop for all plans now, with web and mobile rolling out. If you run a PMO, this is aimed squarely at your status packs.

Try it now →

The European Commission cleared the Code of Practice on AI-content transparency. From 2 August, Article 50 of the AI Act requires machine-readable watermarks on AI-generated audio, image, video and text for EU users, chatbot disclosure, and visible deepfake labels. Deployers have no transition period. Providers with systems already on market have until 2 December 2026 for machine-readable marking. The sign-up deadline for the code is 22 July. Anyone publishing AI-assisted content to EU audiences, including marketing and bid teams, has three weeks.

Learn more →

OpenAI, Anthropic, Amazon, Microsoft and Bank of America, convened with Gina Raimondo, launched RAISE US, a $500 million-plus initiative with a target of $1 billion to retrain workers displaced or reshaped by AI. First pilots are running in Arkansas, Connecticut, Maryland and Utah. With Turner & Townsend flagging data-centre construction labour as the most constrained market globally, funded AI reskilling is now a delivery issue, not an HR footnote.

Learn more →

Oxford expert, Atif Ansar discusses the transformative impact of AI on project management [Podcast]

Before starting a big task, ask the model to sort your brief into knowns and unknowns and interview you on the gaps before it builds anything. Five minutes of forced clarification up front prevents the expensive rework loop. It works for cost plans, reports, programmes and code.

Source →

BuiltWorlds returns to Chicago for its annual Construction Tech Conference, bringing together construction leaders, tech vendors, and innovators to explore emerging technologies reshaping the sector. With AI infrastructure booming and data centres becoming the second most in-demand construction sector, this year's agenda covers digital innovation, AI integration, and the tech stack for modern delivery.

Learn more →

Till next time,

_{All content reflects our personal views and is not intended as professional advice or to represent any organisation.}