Construction AI: the government starts measuring AI's economic impact, and Bluebeam buys its way deeper into drawing review

On 8 June the Treasury and the Department for Science, Innovation and Technology launched the AI Economics Institute - a joint research body whose entire reason for existing is to replace anecdote with evidence on how AI is changing productivity, the labour market and trade. It's chaired by Simon Johnson, the Nobel-winning economist, former IMF chief economist, and co-author of Power and Progress, a book whose central argument is that the gains from new technology don't automatically reach workers - they have to be steered there. That's a pointed choice of chair. This isn't a cheerleading outfit; it's being built by someone professionally sceptical of the "it'll all work out" story.

Two details matter for our sector. First, the institute has signed a Joint Statement of Collaboration with Anthropic, OpenAI, Google and Microsoft - the four labs whose models most of us are already running, directly or through a vendor. Having them inside the tent shaping the measurement framework is sensible and slightly uncomfortable at the same time, and worth watching. Second, and more concrete: DSIT is asking firms to contribute anonymised data on roles, skills and workplace outcomes through a proposed "AI adoption insights agreement" - Sage is named among the early contributors. The stated aim is to see where AI automates tasks, where it creates new work, and where it forces retraining, rather than guessing from headcount announcements. Construction is one of the lowest-productivity sectors in the UK economy and among the most exposed to labour shortage; when this evidence base gets built, the figures attached to our trade will help decide skills funding, apprenticeship policy and how the public sector procures.

Here's the honest read. The institute is genuinely useful - better policy needs better data, and the sector has been flying on vibes and vendor decks for two years. But two cautions. One, the published averages will be blunt; "construction" spans a self-employed bricklayer and a tier-one digital delivery team, and a sector-wide productivity number can be weaponised in a pay or procurement negotiation in ways that don't reflect your firm. Two, the data-sharing is voluntary and aggregated, which is right for privacy but means the early evidence will lean toward larger, better-instrumented firms - exactly the ones already ahead on AI. The SME reality may show up late and understated.

For your board pack: Decide now what AI-and-jobs data you'd be willing to share if asked, and - more importantly - start measuring your own task-level AI impact (hours saved on takeoff, RFI turnaround, document review) so you're arguing from your own numbers rather than a national average someone else applies to you.

On 9 June Bluebeam, part of the Nemetschek group, announced it had acquired mbue, a startup focused on preconstruction and document workflows, alongside a set of new Bluebeam Max capabilities. AEC Magazine framed it accurately as a "talent and technology" deal rather than a big-platform land grab - Bluebeam is buying specialised expertise in drawings, submittals and construction-specific AI and folding it into a product UK estimators, design managers and site teams already have open most of the day.

The substance is in what mbue does. It brings AI-driven, phase-to-phase comparison of drawing sets, and priority-based issue reporting, directly into Bluebeam's Overlay and Compare workflows. In plain terms: point it at two revisions, or at the architectural, structural and MEP sheets for the same area, and it surfaces the cross-discipline and cross-phase mismatches that otherwise get found on site as an RFI, a clash, or a variation. Anyone who's spent an afternoon manually overlaying a P3 against a P2 set knows exactly how much time and how many missed conflicts that represents.

Some context worth being straight about. We covered Bluebeam Max putting Claude inside Revu back in late May, and this is the next move in the same strategy - Bluebeam has been acquiring its way into AI for a while (Firmus AI last autumn, mbue now). The figures and capability claims here are vendor-reported, and "native AI in private beta, broader availability over the summer" is doing some work in the announcement - the drawing-comparison features aren't all generally available the day you read this. So treat it as a direction-of-travel signal, not a tool you can switch on this afternoon. But the direction is the right one, and it's aimed at coordination - the genuinely expensive, genuinely tedious work - rather than another chatbot.

The procurement filter: If drawing comparison or automated clash-spotting is on your buy list this year, check what's landing in Bluebeam Max before you sign for a separate point solution - you may already be paying for the seat.

Two data points landed within four days of each other. On 8 June, Construction News ran a long-read - "Future site: AI in action" - arguing that 2026 is the year AI crosses from the innovation stand to live project delivery on the UK's big sites: document intelligence first (interrogating specs, RAMS, NEC clauses, O&M data and Gateway 2 evidence faster than manual review), then estimating, safety monitoring and scheduling. Four days earlier, on 4 June, the US contractor McCarthy signed a multi-year, multi-million-dollar deal with Palantir to build a connected "AI operating system" across its operations, from early design to field execution. Different scale, same direction: AI is being wired into core delivery, not bolted on as a pilot.

It's worth holding both at arm's length, though. The McCarthy-Palantir announcement is a partnership and an intention, not a published outcome - there's no independent number yet for what it delivers on a job. And the broader "AI on live sites" framing leans on vendor case studies and the same handful of adoption surveys doing the rounds. The honest position is that the deployment is real and accelerating, the labour-shortage and Building Safety Act pressure pushing it is real, but the independent evidence of margin or programme impact is still thinner than the enthusiasm. That gap is exactly what the AI Economics Institute up top is meant to start closing.

So the standing discipline holds, and it's the same one every week: pick a workflow you can measure, run it against your own baseline, keep a human approval gate on anything that touches a contract or a safety case, and demand the independent number before you believe the vendor's. The tools are good enough now that the bottleneck is no longer capability - it's whether you can prove the value on your own projects.

A practical step: Choose one document-heavy task this month - takeoff, submittal review, or Gateway 2 pack assembly - time it manually once, then time it with AI. That single before/after is worth more to your board than any survey.