Construction AI: the plumbing under every AI tool you own gets rewritten next Tuesday, and the regulator admits it was refusing two-thirds of applications

A week today, on 28 July 2026, the Model Context Protocol publishes its final specification. It's called the 2026-07-28 revision, which is about as glamorous as software naming gets, and the maintainers describe it as the largest revision since the protocol launched in November 2024. Beta SDKs for Python, TypeScript, Go and C# went out on 29 June so implementers could test against real workloads first.

Most people in construction have never heard of MCP and don't need to. But you're almost certainly using it. It's the standard that lets an AI model reach into a tool and do something. Bluebeam ships Revu with Anthropic's Claude connected over MCP. Procore, Autodesk and the rest are all wiring agent access the same way. So, when the protocol underneath changes shape, the tools your QSs and document controllers use every day are quietly built on top of it.

What changes is that MCP goes stateless. The initialize handshake disappears, the protocol-level session id disappears, and any request can hit any server instance. In plain terms, you can put an MCP server behind a bog-standard round-robin load balancer and stop worrying about sticky sessions. There's also authorisation hardening, iss validation per RFC 9207, credential binding to the issuing authorisation server, and a formal deprecation policy so this doesn't have to happen again without warning. The maintainers are explicit that nothing breaks on 28 July, and that the date is when the normative text is published, not a switch-off. I believe them. I'd still want it tested rather than trusted.

Here's the bit that matters for a construction business. The authorisation changes are the ones your IT people should read, because they tighten who an agent is allowed to be when it talks to your systems. If you've been letting agents into a CDE with credentials that were vaguely scoped, the direction of travel is that vagueness stops being tolerated. That's a good thing, and it's also work.

Practical bit: If anyone in your organisation runs an MCP server, ask them this week whether they've run the beta against real traffic and pinned their versions. If the answer is "our vendor handles that", get it in writing with a date on it.

On 16 July, Nemetschek Arabia announced regional availability of Bluebeam Max across the Middle East. Bluebeam Max went global on 20 May, so this is a distribution story rather than a product one, and I'd normally leave it alone. It's worth a paragraph for one reason: it's the same premium tier that carries Smart Review, Smart Overlay, Magic Markups, Stitching, and the Revu-to-Revit connected Studio Sessions, and it's the tier where Revu talks to Claude over MCP. So the item above isn't abstract. It's the wiring inside a tool your drawing reviewers may already have on the desktop.

Nemetschek says more than 2,000 beta users reported significant productivity gains in preconstruction, design coordination and construction management. That's vendor-reported and unquantified, so treat it as marketing until you've measured your own team. Usman Shuja, who sits on Nemetschek's executive board and runs Bluebeam, framed it as complexity becoming manageable and data becoming actionable. Fair enough as a pitch. What I'd want to know is whether Smart Review catches the scope gap that costs you £40k on a fit-out package, and nobody has published that number yet.

The procurement filter: Before you upgrade a floor of licences, run Smart Overlay across one live revision set you've already reviewed manually. Count what it found that your team missed, and what it flagged that wasn't real. That ratio is your business case, not the press release.

This one is a fortnight old rather than a few days, and I'm flagging that up front, but it hasn't had the attention it deserves. In a written ministerial statement on 9 July 2026 (HCWS209), the government set out changes to how the Building Safety Regulator handles building assessment certificates for occupied higher-risk buildings.

The number underneath it: since April 2024 the BSR has directed nearly 2,000 buildings to apply for a certificate, and 66% of those applications have been refused. Two-thirds. Read that again, because it isn't a story about incompetent applicants. When a regulator refuses two-thirds of the submissions it specifically asked for, the honest reading is that the standard wasn't legible from the outside.

The response is a more proportionate, risk-based approach. The BSR will focus more closely on organisations managing multiple higher-risk buildings, and will publish updated resources for small, resident-led management companies from September 2026. That last point is the humane bit. A resident-led right-to-manage company with three retired directors and no in-house surveyor was never going to produce a safety case to the same standard as a national managing agent, and asking them to was always going to end in refusal. Separately, the government confirmed funding to extend the Cladding Safety Scheme to buildings under 11 metres with serious life-critical cladding defects, with applications expected to open in August and the worst-affected buildings prioritised.

So, where does AI come into it? Nowhere directly, and that's the point. The golden thread problem the BSR keeps running into is a document problem: evidence scattered across handovers, no traceable chain from design decision to as-built, and a safety case that reads as assertion rather than proof. That's exactly the work document intelligence is good at, and exactly the work most organisations are still doing by hand at eleven at night.

For your board pack: If you hold or manage buildings in scope, pull your last refused or pending BAC and map every claim in the safety case back to a source document. Where you can't, that's your gap, and it's the same gap the revised approach will still find.

Turner & Townsend published its global construction market intelligence on 9 July, and one figure in it should change how your bid desk works. Two-thirds of the markets surveyed, 66%, said AI capability has become slightly or much more important in tendering and client discussions over the past twelve months. Not in innovation workshops. In tender conversations, where the money is.

The rest of the report is the backdrop that makes it bite. 87% of markets report shortages in MEP trades, the exact trades that data centres consume. 71% report labour shortages generally. London is now the fifth most expensive construction market in the world at $6,032 per square metre, behind only New York, San Francisco, Geneva and Zurich. UK construction inflation is forecast at 3.7% this year, rising to 4.2% in 2027. Stephanie Marshall, who runs real estate cost management at Turner & Townsend, put it plainly: there's a very real risk the pool of skilled labour won't keep up with demand.

What that means on a bid is straightforward. When a client asks how you'll deliver a programme with a labour market that tight, "we'll resource it" is no longer an answer. And when they ask what AI you use, the answer they're scoring isn't a tool list. It's whether you can say what it does, who checks it, and what happens when it's wrong. Most contractors can name the tools. Far fewer can name the person who reviews the output before it reaches a client.

That's what it's about.

Today's action: Write down, on one page, the three AI tools your bid team actually uses, the named person accountable for each, and the check that happens before anything goes out. If you can't fill that page in twenty minutes, your next PQQ will find the same hole.