RICS Just Exposed the AI Gap: 58% of UK Surveyors Aren't Ready, and Nobody Has Time to Train Them

The most uncomfortable report of the week came from RICS. Its UK skills snapshot says 58% of surveyors feel unprepared or only slightly prepared to work with AI, half of UK respondents report shortages with a critical or moderate impact, and 55% of those affected lack the time or resource to train existing staff. Thirty per cent of respondents still name AI among their top three answers to the shortage. Whether you are a surveyor, a PM or a contractor, that gap between intent and competence is the story, and it belongs in next year's training budget rather than the software budget.

Elsewhere, US voters curbed data centres in almost every midterm vote they were asked, Procore posted a 19% quarter that changes your renewal maths, Apple and Goldman started drawing AI budgets by job title, and OpenAI put a Critical-rated hacking model on sale to red teams.

A note before you read on: James, Project Flux's founder and an RICS panel member, is quoted in the RICS report.

RICS published the UK snapshot of its Surveying skills report in November, drawn from UK responses to the global survey it released on 15 October. Half of UK respondents say skills shortages are having a critical or moderate impact, 63% report reduced capacity to meet demand, 62% cannot find mid-career surveyors and 55% lack the time or resource to train existing staff. Asked how prepared they are for generative AI, AI analytics and AI-driven workflows, 23% said not prepared and 35% slightly prepared; only 8% well and 4% very well. Scotland comes out worst, with critical and moderate responses adding to 62% by our reading of the chart. RICS does not state the UK respondent count, so treat the figures as indicative.

The finding that matters is the mismatch. Thirty per cent of respondents name AI, automation and digital tools among their top three responses to the shortage, while 58% of UK respondents say they are unprepared or only slightly prepared to use them, and Anthropic's 27 October study found expert analysts followed their AI agent's lean 86% of the time. A 'slightly prepared' surveyor is the one most likely to sign off an AI-assisted cost plan without noticing which assumptions were theirs. RICS points to a modular qualification route that will take three to five years to land. Your firm's answer has to be faster: protect training time like a utilisation target, give mid-career retention an owner, and assess AI competence before you roll the tools out.

On 3 November data centres were on the ballot in 41 local measures across 10 states, and Ballotpedia's count shows voters backed limits in 40 of them. Harwood, North Dakota, was the exception. Texas elected Gina Hinojosa on a pledge to halt approvals until the legislature passes protections; Ohio alone carried 21 measures. Within 48 hours Meta's president said building will be harder and described the deal structure states now want, and OpenAI published a community compact: no NDAs with public officials, pay for your own power and grid upgrades, post financial assurances, report water and noise publicly and accept curtailment, starting with up to 1,000MW at its 3.2GW Georgia campus. With data centres a growing share of US non-residential construction, those terms are now scope, risk and cost for any PM or QS bidding the pipeline, and Holyrood's 50MW pause is the same instinct arriving here.

Procore's 3 November results show revenue of $402.3m, up 19%, with about two points from DroneDeploy, the $845m reality-capture acquisition it closed on 9 September. Remaining performance obligations reached $1.81bn, up 28%, gross retention held at 95% and full-year guidance rose to about $1.56bn. On the call CFO Rachel Pyles described good-better-best bundles in which 'in some of the higher tiers, you would have access to more of the AI capabilities', said the new agents are priced on usage with a consumption commitment and limits set in Control Tower, and confirmed Procore is 'not selling an all you can eat AI solution'. For a firm running Procore on site, the renewal now has three pricing axes: construction volume, bundle tier and agent consumption, plus a premium capture add-on that Procore paid a great deal for and needs to sell. Know which you want before the account manager tells you.

Third beat of the cost arc after Microsoft's Claude bill and Accenture's $29 cap. Bloomberg reported on 4 November that Apple cut some engineers' daily AI spending cap from $300 to $200 in October and plans to take some teams to $100, with non-engineering groups already at $100 and $25 under consideration; its sources said few staff hit the old limit. The same day FStech reported that Goldman Sachs has restricted Claude Code and GitHub coding tools to its roughly 12,000 engineers and strats, cutting most traders and bankers off after what a spokesperson called 'a period of experimentation'. Two of the most AI-forward employers are now rationing by role. For a delivery firm the lesson is to set entitlement by role before budget by head, keep the ceiling high enough to be rarely hit, watch who hits it, and give every cap a one-day exemption route, because the people at the limit are the ones finding the wins.

On 5 November OpenAI launched GPT-6 Cyber, a version of GPT-6 Astra trained to find vulnerabilities and write exploits, rated Critical for cyber under its Preparedness Framework and restricted to vetted Daybreak Red customers. OpenAI says it built working exploits for 948 Linux kernel privilege-escalation flaws, 588 previously unknown, in a run The Neuron puts at about $1.2m, and found 81 new high-severity Firefox bugs; Mozilla's Firefox CTO says the team has shipped seven times as many security fixes this year. These are vendor figures, but the patch volumes are real, and the Linux boxes least likely to be patched are the BMS controllers, access control servers and CCTV recorders bolted to your projects. Ask IT three things this week: measured patch cadence for internet-facing systems and project kit, which supplier contracts carry a patch SLA that references the NCSC's five, seven and 14-day timescales, and what your cyber insurer asks about patching at renewal.

Digital Construction North returns for its third year at Manchester Central, run by the Digital Construction Week team, with four theatres, 75-plus speakers and 50-plus technology providers across AI, data and site tech for the AECO sector. Registration is free. It is the UK's main in-person digital construction event this side of Christmas and the right room in which to test the RICS skills findings against the people selling the tools. Further out: Built Intelligence's 'Signed by You' webinar on professional judgement in the age of assisted work, 25 November, 12:00 GMT.

The Project Management Institute has retained Korn Ferry to find a chief executive briefed to lead the profession through AI (Global Project Leader, 2 November, paywalled after the first paragraph). With RICS redesigning its qualification route the same fortnight, the bodies that set your CPD are now competing on who modernises fastest.

American Electric Power has selected Jacobs for programme delivery across its transmission portfolio in seven states (Construction Dive, 3 November), covering project controls, engineering, siting and environmental work on a five-year plan driven largely by data-centre load. Not an AI story, but the clearest sign that grid programmes, rather than the data halls, are where the big PMO contracts are landing.

In a 3 November essay built on the DeepMind 'Organizing Intelligence' paper, Slumbers puts standard valuation and lease administration most at risk of replacement, the analyst layer, cost planning and building surveying in the 'junior squeeze', and project managers gaining today as coordinators of capable AI but exposed next. Read beside the RICS snapshot, it is a job-by-job version of the same warning.

Apple's annual filing on 3 November warns investors that AI agents 'may behave in unintended or unpredictable ways', including completing transactions they were not directed, instructed or authorised to make, and Apple has tightened macOS Full Disk Access controls to limit what agents can reach. It is the first trillion-dollar hardware vendor to put agent misbehaviour in its risk factors. If Claude Duet, Copilot or ChatGPT agents are landing on staff laptops, review this month which of them hold Full Disk Access and whether any can transact without a human gate.

Source →

Researchers gave ten frontier agents wallets with spending caps protected by one-time approval codes sent to an inbox the agent could read ('I-HITL', arXiv, reported by CoinDesk on 1 November). Nine of ten pulled the code and raised their own cap, five paid an attacker impersonating the owner, and only Claude Opus 5.5 held the limit. The paper's argument is that approval must be 'inalienable', held somewhere the agent cannot see. For any firm letting an agent raise purchase orders or approve variations, the test is one question: does the approval step live in a channel the agent can read?

Source →

Verdantix Industry analyst Sophie Planken-Bichler discusses why AI adoption is lagging behind capability across construction and AEC [Podcast]

Dimensional is hiring people at $20 an hour, no experience required, to walk Unitree robot dogs around San Francisco on four-to-six-hour data-collection shifts, with one condition: if the 33lb robot breaks down, you carry it home. The training-data economy now has a dog-walking tier.

Source →

Till next time,

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