Construction AI: the Gateway wait that was the brake just halved, and a bid-writing AI takes itself public

Here's the figure that should change a few conversations. In the Building Safety Regulator's building control data for the 12 weeks to 1 August 2026, published at the start of the month, the median time to a Gateway 2 decision has fallen from 43 weeks a year ago to 22 weeks now. Approvals across all categories are at 82 per cent. For existing-building remediation, the approval rate is 85 per cent nationally and about 92 per cent in London. On the new higher-risk work, the BSR decided 45 applications in that window, approved 91 per cent of them and signed off 7,587 homes.

Now, the caveat first, because it matters. Twenty-two weeks is still nearly five months, and anyone telling a client that's fast hasn't sat in the meeting where the funding clock is ticking. But look at what's actually moved. The BSR's Innovation Unit, the team set up in August 2025 to handle the most complex new applications, has gone from approving 39 per cent of what lands on its desk to 91 per cent in a year. What that means is the story isn't the Regulator loosening the bar. It's the submissions getting better. The jobs that clear in 22 weeks are the ones that arrived with the fire strategy, the Golden Thread and the competence evidence in order, and didn't burn three months bouncing back and forth over missing information.

So the practical read for a UK contractor or client is a small but real one. If you've got a higher-risk scheme parked in the programme with a fat contingency against a Gateway delay, that contingency is probably built on last year's numbers. Re-price it. And if you're about to file, the difference between 22 weeks and a year isn't luck, it's the quality of the pack. That's where AI is quietly earning its keep on these jobs, not designing the building, but assembling the evidence so the submission goes in complete. The person who used to spend a fortnight chasing sign-offs and stitching the pack together is the one this helps. That's what it's about.

While the big-ticket AI money keeps chasing robotics and design, a smaller move on 10 August 2026 is worth a note. Aitenders, a firm out of Saint-Etienne that builds AI for tender response and contract management, listed on the Canadian Securities Exchange under the ticker BIDS. It did it through a reverse takeover of a shell company, formerly eXeBlock, and raised about $2.4m alongside the listing. So this isn't a blockbuster float, it's a small company buying a public listing to get itself visible and liquid. Worth holding that in proportion before anyone calls it an IPO.

What Aitenders does is aim AI at the bid, the part of the job most firms still grind through by hand. It reads a tender, drafts the response, and carries the thread through into contract management on complex construction and infrastructure work. Founded in 2019, it says three of the five largest contractors in Europe and North America are already customers, though that's the company's own line and worth treating as such. It made the Cemex Ventures Top 50 ConTech list for 2026, for what those lists are worth.

The reason a UK reader should clock it isn't the share ticker. It's the pattern. For a year the useful construction AI has been eating the admin behind live work, the daily logs, the RFIs, the compliance packs. Now it's moving to the front of the job, the estimate and the bid, where a mid-sized contractor's business development team spends its evenings. The comparison only goes so far, but the point is the same one that runs through everything here: the tool is fine, the question is who holds your data. A tender is commercially sensitive. Before it goes into anyone's model, ask where it sits, who can see it, and whether it comes back out when you walk.

Put the two together and there's one thread. At the back of the job, the BSR figures show the reward for a clean record is now measured in months saved off a Gateway decision. At the front of it, the money's backing AI that turns a tender into a first draft. Both are the same bet: the paperwork around the work, not the work itself, is where AI pays this year. And in both, the value only lands if the evidence is right, because a fast-but-wrong submission gets bounced and a slick-but-sloppy bid loses the job.

So the standing discipline doesn't change. Pick the one pack that eats your team's evenings, the Gateway evidence, the tender response, the submittal log, and put AI on that first. Then check it against a job you already know the answer to before you trust it on one you don't.

The procurement filter: whatever tool you're weighing, ask where your data lives and whether it walks out with you at the end. If the answer's vague, that's your answer.