The pipeline thinned and the software stack grew

On 1 October the Home Builders Federation published its latest Housing Pipeline report, and the headline number is the one to put in your forward plan. Just 45,315 new homes were granted planning permission in England in the second quarter of 2026. That's the lowest quarterly figure since 2012, and it's down 21 per cent on the quarter before. Over the full year to the end of June, 214,515 homes got permission, the weakest annual total since 2013, 8 per cent down year on year and 36 per cent below the 2017 peak. The government's own count, done a different way, lands in the same place: 212,000 homes permitted in the year to June, against 242,000 a year earlier.

Now hold that against the figure every client quoted a fortnight ago. New-build starts in England were up 20 per cent year on year in the second quarter, and part of that was a genuine recovery. So which is it, boom or bust. Both, and that's the bit worth getting straight. Starts are what's going in the ground now. Permissions set the ceiling on what can go in eighteen months to two years out, because you can't start a home that was never consented. A rising start count can hide a falling pipeline for a while. Then the pipeline runs dry and the starts follow it down.

The detail I'd underline, and an editor might cut it, is the timing. Both numbers come out of the same few months, so this isn't last year's problem catching up with us. The consents that will feed sites in 2027 and 2028 are being signed right now, and there aren't enough of them. If you carry a forward programme or a resourcing plan, that's the line to look at, and it isn't the one your client will read back to you.

The discipline: take the permissions figure, not the starts figure, into your 2027 resourcing plan this month, and write down what a 20 per cent thinner year would do to your overhead.

The number everyone will quote came out on 6 October. The S&P Global UK construction PMI hit 46.1 for September, up from 44.3 in August and comfortably ahead of the 44.9 the City expected, the slowest the sector has shrunk in eight months. Read it twice, because the word doing the work in that sentence is "shrunk". Anything under 50 on this index means activity is still falling. September was a smaller fall. Less bad and good are not the same thing, and I'd keep that straight before it goes near a board pack.

What sits underneath the headline is the useful part. The three big sub-sectors have stopped moving together. Civil engineering is where the strength is, and Carly Thorpe at Walker Morris put the modest improvement down to stronger civils work. Commercial building had its smallest drop since May 2025. Housing, again, is the one on the floor, with high borrowing costs and a flat market holding output down. Tim Moore at S&P Global reckons all three are at least stabilising, which is the generous reading. New orders tell the other half, with clients still sitting on decisions about the big jobs.

Here's the line an editor might cut. Firms are still cutting jobs, and have done every single month since January 2025. That has run for close to two years now, the industry shedding people while the commentary talks recovery. Put the PMI next to the permissions figure above and the shape is consistent: a bit more work going through now, less in the pipe behind it, and headcount coming down to match. The recovery, such as it is, is lopsided, with infrastructure and the data-centre build-out lifting the average while housing stays still.

For your board pack: lead with the sub-50 figure rather than the eight-month line, and show the sub-sector split so nobody mistakes a data-centre recovery for a housing one.

Schneider Electric confirmed on 5 October that it is buying PTC, the American design and lifecycle software company, for $22.6bn in equity value, or $23.7bn once the debt is counted. The price works out at $205 a share, and the deal is expected to close in the third quarter of 2027. Bloomberg and the FT had it on 4 October before Schneider confirmed. It's the largest software acquisition Schneider has made, and it doesn't stand on its own: the same company bought the AI firm Cognite in June 2026, and it already owns AVEVA, the industrial-software business it took full control of in 2024.

PTC isn't a name most site teams would recognise, but its tools sit under a lot of the work. Creo is design software. Windchill handles product lifecycle, which in plain terms means the record of what got built and why. Schneider's pitch is that it can bolt that software onto its own hardware, the switchgear, the cooling, the power distribution, and sell the lot as one stack to the people building data centres, factories and large buildings. Which is precisely where the demand is, as the PMI sub-sector split just told us.

So why does a construction brief care about an industrial software merger. Because the tools a design office or a QS opens are drifting the same way, towards fewer suppliers, bigger bundles and more of your stack owned by one firm. Nothing changes on your Tuesday, and the deal doesn't even close until 2027. But a thinning order book and a consolidating supply of software is a bad combination for whoever has to negotiate the renewal, and that's the part worth noticing now rather than in two years.

The procurement filter: when you next price a platform, put a line in the model for what it costs to leave, not only what it costs to run.

On 5 October, Homes England and the Liverpool City Region Combined Authority signed what the government is calling England's first Place Delivery Agreement. The name is dull and the idea is not. It pulls the national agencies and the local authority into one delivery team, pointed at one place, with one set of priorities, instead of each running its own programme and hoping the three line up. Liverpool is the pilot. Homes England has said plainly it means to sign one of these with every mayoral authority as devolution spreads, so this is a template rather than a one-off.

The money behind it is the government's new National Housing Bank, sitting on £16bn of debt, equity and guarantees, and the agreement names real schemes to spend it on, the North Docks and the Central Station Gateway among them. Steve Rotheram, the Liverpool City Region mayor, put it the way a mayor would: "This is what devolution should be about, giving places more control and then getting on with the job." The national target hanging over all of it is still 1.5 million homes this Parliament, and it arrives in the same week the permissions figure hit a 14-year low. So the delivery machinery is turning up exactly as the pipeline feeding it runs thin. Whether the machinery shifts the permissions number is the thing to watch, and nobody can promise it yet.

Here's the reason a construction-AI brief cares about a regeneration deal. A joint delivery team only works if everyone is looking at the same picture. Pull three organisations into one programme and the first thing that breaks, every time, is the data: whose land record is right, which pipeline figure is current, who signed off what and when. The agreement is a promise to share accountability, and shared accountability runs on a shared, trustworthy record, or it runs on arguments in meetings.

Today's action: if you work in or around one of the next authorities in line, find out who will own that shared record before the deal is signed, because the standard gets set once, early.

Also on 5 October, Procore announced an EU Application Zone, a chunk of infrastructure that hosts and processes European customers' data inside Europe rather than shipping it across the Atlantic, sitting alongside the UK Zone the firm already runs. The detail worth reading is the list of what stays in-region: not only the files, but the metadata, the logs, the RFI text, the directory and the database records, the exhaust of a running project. That's Procore's own description from its product announcement, so treat the specifics as the vendor's until someone independent tests them, and note that the exact go-live and the countries covered aren't spelled out yet.

Why it counts is procurement rather than performance. For a public-sector client, or a large contractor bound by data rules, where the data physically sits can decide whether a tool clears the buying process long before anyone opens the feature list. What Procore has done is remove a reason to say no. That's a shrewd, unglamorous move, and it's one the big incumbents can afford, because standing up a data-centre presence in every region a client might name is not a small bill. A smaller challenger has to answer the same question with a straight answer instead of a map, and the good ones already do.

So you've probably already got "can we get our data out" on the procurement checklist. The line underneath it is the new one.

Practical bit: add "where does our data live while we're in" to your software checklist this month, because a public client is going to ask, and somewhere in America is increasingly the wrong answer.

The frontier news this week came out of Paris. On 6 October Mistral released a public preview of Mistral Large 4, which it has nicknamed "Le Chonk", and the specs are big: a trillion parameters with 49 billion live at any one time, natively multimodal, a context window up to a million tokens, and more than 160 languages including every official EU one. It was trained from scratch on 3,800 of Nvidia's Grace Blackwell chips sitting in Mistral's own European data centres, and the weights are due to go open at the end of the month, with Reuters reporting the 27th.

Hold the benchmarks at arm's length, because most of them are Mistral marking its own homework. It claims a combined coding-agent score of 49.8 per cent, ahead of the latest from DeepSeek and Qwen, and 82 per cent on a test of reproducing software vulnerabilities, where it says the top models from Anthropic and OpenAI scored near zero because they refused the task outright. The independent read is more sober. Artificial Analysis puts Mistral Large 4 at an intelligence index of 38.4, a big jump on the last version but still only eighth among open-weight models, behind seven Chinese systems and a long way back from the closed frontier. A real step up and not a leaderboard topper, both true at once, and I'd be wary of anyone quoting only the half that suits them.

Why a French model release sits in a construction brief is the same reason as always. The agents that read a drawing, check a spec or chase an RFI run on models like this, and they reach you inside tools you already pay for. What Mistral is selling here is the postcode more than the score. A capable open-weight model, run under European law in European data centres, is a direct pitch to any buyer who's been told their project data can't leave the UK or the EU, which is the same question that decided the Procore move above.

A practical step: if your firm is starting to lean on AI inside its project tools, add one line to the spec, which model and whose data centre, so a European option is something you can choose rather than discover after the fact.

Pagabo, with the public buying organisation YPO, is bringing its third-generation National Framework for Major Works live this month. It's worth up to £5bn over a four-year term running to October 2030, split into five lots by project value, from schemes above £5m up to the big stuff over £60m. The trade press reported at the start of October that contractors have been appointed and the framework opens around 19 October, with the tier ones taking slots across the lots. That appointment detail is as reported by Construction Enquirer via Bricks and Bytes, so treat the exact line-up as provisional until the official notice lands.

A framework like this is the machinery a large slice of public construction gets bought through, and it's built around contractors big enough to carry the compliance load, which quietly means the software stacks those contractors already run. The fair question is whether a mid-tier firm bidding into it can run a lean AI-first tool and still produce the audit trail a public client demands. I think it can. That's a thing to prove on a live bid, not to assert in a brief.

Worth doing: if you're going for a slot or subcontracting under one, get your framework positioning and evidence pack done this month, and check that whatever tool you build it in leaves you holding your own records at the end.

On 28 September Arcadis said it was deepening its tie-up with Autodesk and connecting its own Model Context Protocol tooling to Autodesk Assistant, the AI that's meant to read geometry, engineering intent and project history. The headline the consultant led with was a number: automation had pulled its quality reviews down from five days to roughly half a day on live initiatives across North America and Europe. That figure is Arcadis's own, from existing projects, so treat it as a claim rather than a benchmark. Heather Polinsky at Arcadis and Andrew Anagnost at Autodesk both put their names to it, which tells you how much weight each side is putting on the pairing. There's a joint project too, called See Through Walls, using sensors and predictive modelling to work out the condition of what's hidden inside existing buildings, aimed at material reuse and cutting carbon. That's the genuinely interesting bit for retrofit, and the part I'd want to see stand up on a real job rather than in a press release.

Notice the shape of it, though. One of the biggest consultancies working on UK infrastructure has decided the smart move is to embed itself deep inside a single incumbent's platform. Reasonable for a firm that size, and not a template for a 40-person practice.

The takeaway: don't copy the partnership, copy the target. The review-and-check work is where the time actually goes, so go after that first, whatever tool you use.

The 5 October ConTech roundup tallied a busy week of raises. Kahua took a $250m growth cheque from Bain at a billion-plus valuation for capital-programme project management, Autonomy pulled $225m from SoftBank for autonomous fleet control, and Miter raised $40m for construction workforce admin. A clutch of smaller AI-first estimating and bidding startups took seed and Series A money too, the German bidding tool Kuro, an estimating platform called Quotr, and a takeoff tool called Constor coming out of stealth. Most were European or American. None were British, which is worth sitting with for a moment given how much UK construction software gets talked about at conferences here.

Those are the reported figures, and venture numbers always read big, so hold them lightly. The pattern that matters is where the money is pointing: programme management, workforce admin, estimating and takeoff. The paperwork, in other words.

The practical bit: look at where that money is going and ask which of those four jobs eats the most hours in your own business, then test one tool against that job rather than against a demo.

The sharpest line in that same 5 October roundup wasn't a cheque. It was the argument that as AI models get cheap and samey, what decides whether a tool earns its place on a construction job is the scaffolding around it, the tools, the memory, the workspace, the guardrails and the repeatable tests that check output against what a good QS or site manager would actually accept. In plain terms, the cleverness has moved. Anyone can rent a frontier model by the token. The hard part is the domain knowledge baked into how the tool checks itself, because on a drawing or a valuation, knowing what good looks like is tacit and it doesn't come in the box with the model.

I'm wary of anything that sounds like a grand theory of AI. This one survives contact with a real job, which is more than most manage.

A contributed piece in Planning, Building and Construction Today on 7 October argued that the data-centre boom, the one busy corner of a lopsided market, is turning subcontractor payment from an admin chore into a delivery risk. It's written by a payment-software firm, so read it as a vendor making its own case. The point stands anyway: a data-centre job can run hundreds of subcontractors and throw off hundreds of payment applications a month, and when that's held together with spreadsheets and email, the thing that breaks first is people getting paid on time.

That matters more than any model this autumn. On the one job that's genuinely busy, the specialist electrical and mechanical firms you need are in demand and can pick their main contractor. What keeps them is dull and human: accurate valuations, approvals they can see coming, a payment cycle they can predict. It's a bit like a good crew. They come back for the firm that pays straight, long before the one with the fancier kit.

On the programme: before you add another tool to a data-centre job, ask what it does to your month-end payment run, and keep your payment records in a form you can take with you if you change system.