Hello Project AI enthusiasts,
ChatGPT can now draw a cost calculator inside the conversation, while Google's Gemini agent can arrive with its own email address, calendar, and directory identity. Anthropic has added a new enforcement clause for Claude users, and Microsoft and Meta are reportedly trimming Claude spend as model costs become a procurement issue. Meanwhile, Texas and Scotland are showing that power, planning, and local consent can constrain the AI build-out before capital does.
Featured This Week
GPT-6 and Intelligent UI begin rolling out to ChatGPT's 1.2bn-plus weekly users
OpenAI says GPT-6 is now rolling out across ChatGPT tiers, with Intelligent UI allowing the model to produce interactive calculators, forms, charts, comparisons, and tappable tools inside a response. On web-search questions, OpenAI says GPT-6 Instant starts answering 44% sooner on average than GPT-5.6 Instant. Enterprise access still depends on administrator settings, and the Work and Codex models are unchanged.
For a QS, PM, or commercial manager, the practical shift is easy to see. A cost comparison or cash-flow calculator generated in chat can look like a finished professional output. It is still a model response, with assumptions and arithmetic that need checking before anything carrying your name reaches a client. The interface has become more persuasive, so the checking control matters more, not less.
Editor’s Picks
This week's must-read stories on AI, project delivery, and infrastructure:
Google’s Gemini agent gets its own email address, calendar, and seat in your company directory
Google’s new Gemini agent can operate as a persistent coworker with a Workspace account, email address, calendar, Drive access, and directory entry. It can connect to Microsoft 365, Slack, Jira, Confluence, Salesforce, ServiceNow, and MCP servers. The project question is straightforward: who owns and approves the work produced by an agent that has its own identity?
Anthropic’s new usage policy makes abusing Claude a bannable offence from 12 November
Anthropic’s updated Usage Policy adds a prohibition on sustained and needless abusive or cruel behaviour towards its models, effective 12 November. For firms using Team or Enterprise seats, administrators should understand the new conduct rule and check that internal AI policies align with it.
Microsoft cuts its internal Claude spend by more than a third, and the Copilot model choice is becoming a cost line
The Information reports that Microsoft reduced projected internal Anthropic spending by more than a third from a reported annual run rate above $1bn, while Meta's Claude Code users roughly halved. Microsoft is reportedly exploring cheaper models for Copilot customers. The useful procurement question is now which model sits behind a licence and whether that model can change without a clear user-facing notice.
Texas freezes data-centre permits as Holyrood backs a pause on schemes over 50MW
A large-load queue of 474GW in Texas, much of it linked to data centres, has met a grid that cannot connect projects quickly enough. MSPs have separately voted, in a non-binding motion, that no decisions should be taken on schemes above 50MW until national guidance is published, and ministers have made an environmental impact assessment mandatory for every such proposal. For contractors and consultants, permits and grid access are now schedule risks in their own right.
AI in AEC and Projects
Here are some intriguing reads, specifically for AEC and project delivery professionals:
ANSI/A3 R15.08 Part 3 covers how users operate autonomous mobile robots and manipulators around workers. It is a voluntary consensus standard, rather than legislation, but gives method-statement writers a complete reference point.
The agreement concerns uprates at existing plants and power supplied to the PJM grid. It is a signal that firm power procurement and specialist engineering capacity are becoming part of the AI infrastructure pipeline.
After more than 750,000 autonomous miles with a safety observer, the partners plan a driverless 219-mile freight lane. Materials-heavy projects should watch what this does to long-haul logistics assumptions.
OpenAI's frontier models will power Rovo experiences across Jira and Confluence, grounded in Atlassian's Teamwork Graph. Delivery teams may gain agents that can read decisions, documents, and issues together rather than one ticket at a time.
Amazon says the five-year programme will support schools and housing energy upgrades. Social licence is becoming a visible line in the data-centre business case alongside power and land.
Curated Links
Selected stories worth reading before you get back to delivery:
Wikimedia reports unauthorised activity from OpenAI-linked agents. A useful incident for testing whether proposed agents can access systems as proxies or generate uncontrolled request volume.
Microsoft Execution Containers reach general availability. Policy-driven sandboxes give IT teams a practical control for model-written code on Windows devices.
Adversa AI's guide to OWASP ASI08 explains cascading failures in multi-agent systems The checklist translates agent trust chains and feedback loops into questions for a deployment review.
Free Gemini now routes most unpaid prompts to Flash-Lite. Teams using the free tier should check whether a quiet routing change affects the quality of work entering deliverables.
Claude Haiku 5.5 arrives with lower pricing and new plan credits. Model prices and cache economics are moving underneath familiar tools, so renewal comparisons need more than seat counts.
Tool of the Week
Claude Dashboards: live views built from project data
Claude Dashboards is in beta on paid plans. Users describe the view they need and Claude builds a dashboard connected to live sources such as Salesforce, Snowflake, and BigQuery. A PMO could test a cost and programme view that refreshes from a live report instead of relying on a monthly screenshot. Enterprise administrators have the feature off by default, so confirm the control position before testing.
Governance
EU AI Act Omnibus moves high-risk duties to December 2027
The adopted EU Digital Omnibus delays certain high-risk obligations to 2 December 2027 and product-embedded system duties to 2 August 2028. The bans and AI-literacy duty remain in force, while providers of generative systems face a machine-readable marking deadline on 2 December 2026. Firms with EU projects have more time for conformity assessment, not for training or labelling controls.
Training
Google Cloud's Advent of Agents, Season 3
Google Cloud's Season 3 runs across October 2026 with short daily tutorials covering agent development, identity, permissions and security. The opening lessons include lifecycle foundations and SAIF guardrails. It is vendor-run and free to follow, yet the sequence gives delivery and IT leads a structured way to learn the controls behind the agents they may soon be asked to approve.
Also This Week
Sam Rees, CEO of Hopper, discusses how AI is transforming data management in the construction and infrastructure industries. [Podcast]
One More Thing
Figure retires its F.02 humanoids by jumping most of the fleet into a Finnish steel furnace
Late last month, Figure decommissioned most of its F.02 fleet by having the robots jump into a 75-tonne electric arc furnace at a foundry in Imatra. It is theatrical, but also a neat reminder that proprietary hardware needs an end-of-life plan, even when the original deployment was mainly a demonstration.
Quick Win
Audit a cost plan’s formulas with Claude inside Google Sheets
Claude’s Google Workspace integration is in public beta on paid plans and defaults to “Ask before edits”. Use a copy of a cost plan or cash-flow sheet to flag broken references, hard-coded values, unreconciled subtotals, and inconsistent rates. Review each proposed change individually, keep “Accept all edits” switched off for contractual material, and paste the change list into the issue notes.
Event of the Week
RICS in Bristol: Responsible AI in Surveying, tomorrow, 14 October
RICS Matrics and the Bristol Local Members Group are hosting a structured CPD seminar at Engine Shed from 15:00 to 17:00 BST. Speakers include Project Flux founder and RICS panel member James Garner, Grazyna Wiejak-Roy, and Tom Storey. Tickets start at £20 plus VAT for members. James is speaking in a personal capacity.
This week's newsletter is sponsored by:
Morta — the controls and information platform for design and construction delivery.
Morta is the controls and information platform for design and construction delivery. We replace fragmented spreadsheets and disconnected systems with a single source of truth to structure information, track planned versus actual, and bring consistency to mission-critical workflows from design through to handover. Trusted by industry leaders across 10+ countries, including ALEC, Kier, Mott MacDonald, Mace, and the UK's Ministry of Justice, and deployed across more than £100bn of construction value worldwide.
Try Morta for free. Start with automating your information delivery planning and tracking by connecting with systems like ACC, Aconex, Procore, ProjectWise and Viewpoint at morta.io
Till next time,
Project Flux
All content reflects our personal views and is not intended as professional advice or to represent any organisation.