AI This Week: UK Construction Is Contracting, and the Tools You're Using at Work May Already Be Flagged as High-Risk

Key Takeaways

• UK construction saw a 37% decline in main contract awards and a 51% drop in detailed planning approvals year-on-year in May 2026, according to Construction News.
• Alibaba has reportedly classified Anthropic's Claude Code as "high-risk" software and banned employees from using it, citing data security concerns.
• AI tools used informally by your staff carry real data leakage risks that most SME owners have not yet mapped or addressed.
• A tightening construction pipeline means faster, more accurate quoting and follow-up is no longer optional for firms competing on fewer available contracts.
• Yan LeCun's public comments at Meta reinforce that current AI, including the tools widely deployed today, has significant cognitive limitations worth understanding before you build your business processes around them.

What Is Actually Happening to UK Construction Activity Right Now?

UK construction is having a rough spring. May 2026 data published by Construction News shows a 51% decline in detailed planning approvals against the previous year, a 37% drop in main contract awards, and a 10% decrease in project starts. Industrial specifically is worse: a 19% drop in project starts year-on-year, and a 64% collapse in main contract awards to £534 million.

To be clear about what this means at the ground level: there are fewer contracts being awarded, fewer projects clearing planning, and a thinner pipeline for contractors, subcontractors, and the trades working beneath them. If your business development has been coasting on inbound enquiries from a busy market, this is the data that should prompt a rethink.

The firms that will hold margin in a contracting market are the ones that respond to enquiries faster, follow up on quotes more consistently, and lose fewer leads to slow admin. That is not opinion. When there are fewer jobs available, the businesses that convert at a higher rate win. Right now, a significant number of UK trades and construction firms are still relying on manual quote follow-up, spreadsheets to track enquiries, and their memory to chase warm leads. That works fine when the phone does not stop ringing. It stops working the moment the pipeline tightens.

In the systems we build for construction businesses, one of the most common problems we find is what you might call invisible leakage: enquiries that came in during a busy week, got a rough response, and were never followed up when the job board cleared. The business owner often has no idea how many of those became someone else's contract. When the market is buoyant, that leakage is annoying but survivable. In a 37% contract award decline environment, it becomes an existential problem.

The practical takeaway is not complicated. If your firm cannot answer definitively how many enquiries you received last month, how many received a quote, and how many of those quotes were followed up within 72 hours, you are operating blind at exactly the wrong moment. A properly configured enquiry handling system gives you that visibility, and it closes the gap between a lead arriving and someone responding before the prospect calls the next firm on the list. This is not about automation for its own sake. It is about not losing work you should have won.

For businesses in renewables and solar, the picture is slightly different given the ECO4 pipeline, but the underlying operational point is the same. More competition chasing a more selective set of available jobs demands faster, more disciplined follow-up.

Why Alibaba Banning Claude Code Should Make UK Business Owners Think About Their Own Tool Stack

Alibaba has reportedly banned employees from using Claude Code, Anthropic's AI coding assistant, classifying it as high-risk software. The reasoning, as reported by TechCrunch, centres on data security concerns.

This story is not really about Alibaba. It is about the pattern it represents, and what that pattern means for any UK business that has staff using AI tools informally across their working day.

Most SME owners would be surprised, if they sat down and audited it honestly, at exactly which AI tools their team members are already using. Someone on the sales side is probably drafting emails in ChatGPT. Someone in the office is summarising documents in Claude. Someone on site might be photographing drawings and feeding them into an image model to get a quick take on dimensions. None of these things have been signed off. None of them have been reviewed for what data is being transmitted to a third-party server. And in most cases, no one has read the terms of service to understand how that data is handled or retained.

Under GDPR, the obligation on UK businesses is clear: if you are processing personal data using a third-party tool, you need a lawful basis for doing so, you need to understand where that data goes, and in many cases you need a Data Processing Agreement in place with the tool provider. Most informal AI tool usage in SMEs satisfies none of those requirements.

The Alibaba story is useful precisely because it comes from a technically sophisticated organisation that clearly thought carefully about this and still concluded the risk was too high for internal use. That should prompt a question: if a technology company with a dedicated security team decided to classify a widely used AI coding tool as high-risk, what does that say about the informal AI usage happening across your ten-person plumbing firm or your solar installation business?

The practical response is not to ban everything. That approach fails because people will simply use the tools on their personal phones instead, and then you have no visibility at all. The better approach is to run a proper audit of which AI tools your staff are currently using, map what data those tools are touching, and then either formalise the usage with appropriate controls or replace ad-hoc habits with a system you have actually designed and secured. In the systems we build for clients, we make this explicit from the start: every data flow is documented, every third-party integration is reviewed, and the business owner knows exactly what is being sent where. That is a much safer position than hoping nobody is doing anything risky.

If you want a starting point, our AI automation checklist walks through the questions you should be asking about your current setup before a data incident forces the conversation.

What Yan LeCun's Criticism of Current AI Actually Means for Businesses Deploying It Now

Yan LeCun, one of the most respected researchers in the field and VP of AI at Meta, made headlines this week via the BBC with a characteristically blunt assessment: current AI systems are "not smart." His argument, which he has made consistently for some time, is that large language models lack the kind of flexible, world-model-based reasoning that would make them genuinely reliable for open-ended tasks. He is working on alternative approaches through his own research, but has been transparent that nothing is ready to replace what exists today.

This is worth taking seriously, not because it means you should stop using AI tools, but because it clarifies exactly where the failure modes are.

Current AI models are excellent at pattern matching across text. They produce fluent, plausible-sounding outputs very quickly. They are genuinely useful for drafting, summarising, classifying, and routing. Where they fall apart is in any task that requires consistent logical reasoning across multiple steps, reliable retrieval of specific facts without hallucinating, or genuine understanding of cause and effect in novel situations. LeCun's point is not that these tools are useless. His point is that the gap between what they appear to be capable of and what they are actually capable of is still very large, and that the AI industry has been, at times, loose with how it characterises that gap.

For a UK SME deploying AI, this translates directly into a design question. The tasks where AI adds clear, low-risk value are the ones with well-defined inputs and outputs: routing an inbound enquiry to the right person, drafting a follow-up email from a template, extracting data from a structured document, sending a reminder at a set interval. The tasks where AI will let you down, sometimes embarrassingly, are the ones that require contextual judgement, nuanced interpretation of unusual circumstances, or any kind of factual precision without a verified data source behind it.

This is why the systems we build are always designed around defined workflows rather than open-ended AI autonomy. A workflow automation system that triggers a quote follow-up 48 hours after sending is reliable and measurable. An AI that "handles your admin" in some vague, general sense is neither. LeCun's broader critique of the industry is really a useful framing tool for any business owner trying to separate the genuinely useful applications from the hype that surrounds them. Build around the things current AI does well. Keep humans in the loop for the things it does not. That is not a limitation to apologise for; it is just good system design.

The firms that will get real value from AI in the next two to three years are not the ones that automate everything blindly. They are the ones that identify the specific operational bottlenecks, the places where time and money are leaking, and apply the right tool to exactly that problem.

How the NFB Academy Connects to a Wider Skills and Compliance Gap in UK Construction

The National Federation of Builders launched the NFB Academy this week, described as a comprehensive training and professional development hub aimed at construction businesses across the UK. The focus areas are leadership, compliance, and sustainability, which is a fairly precise read of where smaller construction firms are feeling the pressure right now.

This is worth covering because it signals something broader about where the construction industry's operational stress points are sitting in 2026. Compliance burden in construction has grown considerably over the past few years. The Building Safety Act has raised the bar on documentation and accountability. CHAS accreditation requirements are tightening. CDM regulations demand more structured site management. And the push toward sustainability reporting, driven partly by procurement requirements on larger contracts, is creating paperwork obligations that most smaller firms did not have even three years ago.

The NFB Academy is a direct response to that pressure. The idea is to give construction business owners and their management teams a more structured path through the compliance landscape, rather than leaving them to piece it together from scattered guidance documents and the occasional trade association newsletter.

Where this intersects with what we do at Aucta AI is in the administrative load that compliance creates. Every RAMS document that needs updating, every training record that needs logging, every certification that needs renewing represents time that someone in the business spends on process rather than production. For a small firm, that someone is often the owner. The irony is that a lot of this compliance documentation follows highly repeatable patterns: the same data entered in slightly different formats, the same checklists completed on a recurring schedule, the same records retrieved when an auditor asks for them.

That is exactly the kind of work that a properly designed admin automation system can take off your plate. Not by cutting corners on compliance, but by making sure the compliant thing happens automatically and on time, without someone having to remember to do it. A groundworks firm working under CHAS accreditation, for example, has a predictable set of recurring documentation tasks. There is no reason the majority of that should be a manual process in 2026.

The NFB Academy is a good initiative. Training and compliance literacy matter. But for most construction SMEs, the knowledge of what needs doing is not the bottleneck. The time to actually do it, consistently, without it falling through the cracks during a busy site period, is the real problem. That is where the operational answer sits.

---

If any of the stories this week have made you think about where your own business is leaking time or revenue, the AI automation checklist is the right place to start. It is free, it takes fifteen minutes, and it gives you a clear picture of where the gaps are before you spend anything.