AI This Week: A Nobel Laureate Defects, Construction Insolvencies Drop, and the Talent War Heats Up

Key Takeaways

• Nobel laureate John Jumper is leaving Google DeepMind to join Anthropic, signalling that the AI talent war is intensifying at the very top of the field.
• Construction insolvencies in England and Wales fell by nearly a third in May 2026, with 281 firms collapsing compared to the previous month's figures, according to the Insolvency Service.
• Glenigan's Summer 2026 Construction Forecast projects sector recovery in 2027, though the firm warns the start of 2026 has been painful.
• The modular construction segment continues to face its own structural difficulties, with Elements Europe's administration extended to June 2027.

John Jumper Is Leaving DeepMind for Anthropic

This is the macro story of the week. John Jumper, the researcher who shared the 2024 Nobel Prize in Chemistry for his work on AlphaFold, is leaving Google DeepMind to join Anthropic. And he's reportedly not alone. According to TechCrunch AI, he's one of several prominent names departing DeepMind at the same time.

To understand why this matters beyond the headlines, you have to appreciate what AlphaFold actually did. It solved a 50-year-old protein-folding problem, one that had stumped structural biologists across the world. DeepMind was built on moments like that, and Jumper was central to one of its biggest. The fact that someone at that level is moving to a direct competitor says something about how the competitive dynamics inside frontier AI are shifting.

Anthropic has positioned itself as the "safety-first" lab, the more considered alternative to OpenAI's speed-at-all-costs culture. But that positioning hasn't stopped it from being aggressive in attracting talent. The company is growing fast, it has significant backing, and it's clearly offering researchers something DeepMind no longer can, whether that's autonomy, compensation, a particular research direction, or some combination of all three.

For UK businesses, the direct implication of this story isn't about DeepMind specifically. It's about what happens to the models you're using when the people who built them start leaving. The AI tools your business depends on, whether that's Claude, Gemini, GPT-4, or anything else, are only as good as the ongoing research teams maintaining and improving them. Talent volatility at the top of these labs is worth watching, because model quality isn't static and the lead can change hands faster than most people expect.

The broader lesson is that locking your business operations into any single AI platform without a strategy to adapt is a risk. The custom AI systems we build at Aucta AI are designed with this in mind, built to work with whichever models and tools best suit the job, rather than betting everything on one provider's continued dominance.

Construction Insolvencies Fell Sharply in May 2026

The headline number is good news. According to the Insolvency Service, 281 construction firms collapsed in England and Wales in May 2026, a 30 per cent month-on-month decrease from April. After a prolonged period of brutal attrition across the sector, any meaningful reduction in insolvency figures is worth noting.

But 281 firms in a single month is still a significant number. A 30 per cent drop feels substantial until you remember the baseline it's dropping from. The construction sector has been shedding businesses at a genuinely alarming rate since interest rates spiked, materials costs stayed elevated, and labour costs climbed with minimum wage increases. One better month doesn't reverse that trend, it just suggests the rate of deterioration may be slowing.

What it likely reflects is a market in the process of clearing. The firms that couldn't hold on through 2024 and 2025 have largely gone. What remains is a leaner, more cautious pool of businesses operating with tighter margins and more selective project pipelines. The surviving contractors tend to be the ones who got serious about operational discipline, watching cashflow closely, cutting the administrative overhead where possible, and following up on every viable lead rather than assuming work would come in.

This is exactly where workflow and admin automation earns its keep for trades and construction firms. When margins are tight and headcount is flat, the businesses that make it are the ones who don't lose a job because a quote took three days to go out, or a follow-up call never happened. Operational leakage, the slow bleed of missed enquiries and uncontacted leads, is what kills firms in lean markets, not the dramatic collapses but the quiet failures to convert.

For firms operating in construction right now, the May figures are a cautious green shoot. But the structural pressures that drove 18 months of elevated insolvencies haven't gone away. Margins remain thin. Payment terms remain punishing. And the firms that survive the next 18 months will be the ones who've stopped absorbing those pressures manually.

Glenigan Says Recovery Is Coming, But Not Yet

Glenigan's Summer 2026 Construction Forecast is projecting a sector recovery, with the outlook pointing toward 2027 as the point where conditions meaningfully improve. The forecast notes a painful start to 2026, which maps with everything the insolvency data has been showing for the past year.

This kind of forward-looking data matters for planning decisions. If recovery is tracking for 2027, then the businesses that invest now in the infrastructure to handle increased volume, faster quoting, better enquiry handling, smoother project admin, are the ones that will actually be positioned to benefit when the pipeline opens up.

There's a pattern that repeats in sector recoveries. Businesses that spend the down period cutting capacity struggle to scale when demand returns. The smarter move is to use the quieter period to fix the operational problems that were always there but masked by busyness. An enquiry handling system that routes, qualifies, and follows up on incoming leads automatically doesn't just help during a boom. It means you're not losing a recovery lead because someone was on site and didn't pick up the phone.

Glenigan's forecast also fits with the modular sector picture painted by the Elements Europe administration extension. That firm's process has now been extended to June 2027, suggesting the cleanup in that corner of construction is a longer, slower job than initially anticipated. The modular segment had specific dynamics around fixed-price contracts and supply chain risk that hit it particularly hard, and the administration timeline reflects that.