The week the excuses fell away - Gateway 2, MCP, and the measure of impact

A quieter week after DCW, with a different shape. The stories that landed all push in the same direction: the things that used to be the reason a UK construction AI conversation stalled have stopped being the reason.

The biggest of them is the Building Safety Regulator. For two years the honest answer to "why is the higher-risk-building scheme stuck?" was the BSR's inbox. Gateway 2 was taking up to 48 weeks in London at its worst, and around 43 nationally. Programmes were being written around it. Some weren't being written at all. That story has changed. Andy Roe has put the current wait at "13, 14, 15 weeks" and "often" the statutory 12, with firms now told within a week whether their application has been validated. Trade coverage on 4 June reported Gateway 2 approvals rising alongside a marked increase in new-build applications - the clearest sign yet that the pipeline is moving, not just the backlog shrinking. The catch is that the constraint is shifting forward. Gateway 3 - the sign-off you need before anyone can occupy the building - is shaping up to be 2026's gridlock. And the golden thread has turned, as one trade editor put it this month, into a "golden burden". When the regulator was the bottleneck the quality of your evidence pack was a secondary concern. With approvals turning in three months, whether your documentation is complete, consistent and navigable on first submission is what decides whether your scheme moves.

The integration layer for construction AI also stopped being a debate. Autodesk's announcements at DevCon 2026 closed out the story the brief has been tracking since 19 May. Forma Assistant is being positioned as Autodesk's AI orchestration layer, with public Revit, Fusion and Fusion Data MCPs in tech preview and the Design and Make Marketplace updated so certified third-party agents can be called by Assistant directly. Read with Procore's CDE relaunch on Datagrid agents (1 June) and Bluebeam Max running Claude via MCP (19 May), three of the dominant AEC software vendors are building on the same orchestration protocol within a six-week window. MCP support is now the sensible procurement filter for anything AI-adjacent. The honest caveat MCP itself has been carrying since spring - that it's an interface protocol being asked to carry a governance load it wasn't built for - means the filter is "ships MCP and has a credible answer on credentials, audit logging, data residency and approval gates." Both halves matter.

On the policy side, the Treasury and DSIT launched the AI Economics Institute on 8 June, chaired by Simon Johnson - Nobel-winning economist, former IMF chief economist, co-author of Power and Progress, a book whose central argument is that the gains from new technology don't automatically reach workers. That's a pointed choice of chair. The four big labs - Anthropic, OpenAI, Google, Microsoft - have signed a Joint Statement of Collaboration, and DSIT is asking firms to share anonymised role, skills and outcome data through a proposed "AI adoption insights agreement". Construction is among the lowest-productivity and most labour-short sectors in the economy, so when this evidence base gets built, the figures attached to our trade will help decide skills funding, apprenticeship policy and public-sector procurement. Get ahead of it by deciding what your firm's task-level AI impact metric is going to be before a national average lands on the desk in front of you.

On the model side, Anthropic released Claude Fable 5 on 9 June - the first generally available model from its restricted "Mythos" tier. Skip the leaderboard theatre and look at where the gains actually land, because for our world it's unusually relevant. Anthropic's own figures put Fable 5 ahead of Opus 4.8 specifically on document-heavy knowledge work: the top score on Hebbia's finance benchmark, with the headline gains in document-based reasoning, chart and table interpretation, and visual-document understanding. On GDPpdf - reasoning over visual documents without tools - Anthropic reports 29.8% against Opus 4.8's 22.5%. That's the capability that maps onto a Gateway evidence pack, an O&M manual, a clause-heavy contract or a dense set of structural calcs. The catch is price - $10/$50 per million tokens, roughly double Opus 4.8. Reserve it for the hard reads and route routine work to a cheaper model that already clears your bar.

Underneath, the "AI under your control" architecture filled out across the stack. Rodic Consultants unveiled a sovereign, air-gapped AI ecosystem for infrastructure delivery on 8 June, launched in the post-DCW news cycle - roads, railways, bridges, urban systems and public infrastructure with critical project data never leaving the customer's perimeter. Google DeepMind released Gemma 4 12B on 3 June, an encoder-free multimodal model that handles text, images, audio and video natively, runs on 16GB of laptop RAM, and ships under Apache 2.0. With Anthropic's Claude Platform on AWS and NVIDIA's Nemotron 3 Ultra both landing in the past fortnight, the data-residency objection that PI insurers and information-governance teams have been raising for the last six months has visibly weakened. ServiceTitan's 2026 industry report puts the share of contractors with measurable AI business impact at 38%, up from 17% twelve months ago. The investment question is settled. The impact question is the one that wins or loses budgets now.

A couple of other things worth holding onto. Provision - co-founded by a quantity surveyor - is the preconstruction-AI play worth knowing about, claiming to compress 30-40 hours of manual scope review per bid into under an hour, with every requirement linked back to its source document. With RICS reporting 31% of UK quantity-surveying practices having unfilled vacancies, the labour-shortage case for AI estimating is structural, not cyclical. Bluebeam acquired drawing-review startup mbue on 9 June - phase-to-phase drawing comparison and priority-based issue reporting going straight into Bluebeam Max. And the floating data centres story from Posidonia 2026 (Samsung Heavy, Capital Clean Energy Carriers, Lloyd's Register) put a new construction typology onto the UK marine and offshore-engineering pipeline.

Pull all of that together and the test for the rest of June is sharper than it has been. The regulator has stopped being the reason. The model has stopped being the limit. What's left is whether your evidence is complete and citable, whether your impact is measurable, and whether your procurement filter is doing its job. Pick the workflow. Set the metric. Run it. Write it up in your own voice.