It reads the rhythm of walking, and steadies it in real time 🌿

For the more than 10 million people living with Parkinson's disease, walking can turn unpredictable. Steps shorten, gait grows uneven, and falls follow. Falls are the leading cause of injury-related hospitalisation for people with the disease, and conventional deep brain stimulation, which delivers a steady pulse, cannot respond to the moment a person actually begins to stumble.

On June 15, UCSF researchers published a trial in Nature Medicine of something far more responsive. Their adaptive deep brain stimulation system reads the brain's own rhythm of walking and adjusts stimulation within fractions of a second. On-device machine learning classifiers detect the neural signature of each left and right step and switch stimulation through every phase of gait, with no external computer required. Much like a cardiac pacemaker responds to heart rhythm, this device responds to the rhythm of movement itself.

In a randomized, blinded crossover trial with five participants, the adaptive system improved gait symmetry and reduced the variability in walking patterns that tends to precede falls. Both are markers of steadier, safer movement.

This is AI doing something quietly profound. Not generating text or images, but closing a feedback loop inside the body fast enough to matter for a single footstep.

Five participants is a proof of concept, not a population. The work now is replication in larger trials, and confirming that better gait metrics translate into fewer real-world falls over months and years.

Source: UCSF, Neuroscience News

On June 18, the Federal Energy Regulatory Commission unanimously ordered six grid operators serving more than 200 million Americans to fast-track connecting large electricity loads, including AI data centres, and to weigh advanced transmission technologies that lift capacity on existing lines. The bottleneck for powering AI cleanly is often grid connection, not generation. This bipartisan order helps clear it.

Source: TechCrunch

On June 19, the European Commission selected the EUROPA consortium, led by Italy's Domyn, to build a 400-billion-plus parameter open-source AI model that works natively across all 24 official EU languages, powered by EuroHPC supercomputers. For 450 million citizens, it means frontier AI built as a public good rather than locked behind a single vendor, and proof that open-source and frontier need not be opposites.

Source: European Commission

A Pew Research survey published June 17 finds 50% of US adults have used AI chatbots, with one in four using them daily. The most telling detail: when asked which field AI will most benefit, Americans named healthcare above all else. AI-for-good is no longer just a tech-industry talking point. It is where the public hopes the technology is heading.

Source: Pew Research

The year the "AI kills jobs" story met its data

For three years the debate over AI and work has run mostly on anecdote and forecast. This week it got its largest evidence base yet. PwC's 2026 AI Jobs Barometer analysed more than one billion job ads across 27 countries, and the headline finding cuts against the dominant fear: the most AI-exposed companies are adding workers 52% faster than the least exposed. AI skills now carry a 62% wage premium, entry-level roles that fold in AI skills grew 35%, and the top fifth of AI adopters posted productivity gains of 163%. PwC's parallel CEO survey found 61% expect AI to grow their total headcount within three years. The picture is not uniformly bright. Mid-career workers in routine cognitive roles face the sharpest disruption, and a wage premium for AI skills is also a widening gap for those without them.

Our PM + Risk Manager lens

For people building AI products, the Barometer reframes the customer. The fastest-growing teams are not replacing people with AI, they are pairing them. That favours tools designed for augmentation, with human review, clear handoffs, and visible reasoning, over tools that promise full automation. It also raises the value of features that help workers build AI skills inside the product itself, because the data shows skills, not just access, are what command the premium. Build for the worker who is leveling up, not the one being replaced.

A 62% wage premium is good news for those who have AI skills and a warning sign for those who do not. The same data that demolishes the doom narrative documents a real distributional risk: routine cognitive roles compressed, mid-career workers exposed, and gains concentrated in superstar firms. A single annual report, however large, is also a snapshot, not a causal proof. The honest reading is that AI is reshaping work faster than it is destroying it, and that the people most at risk need transition support before the gap hardens.

The next 12 to 24 months

Expect the argument to shift from "will AI take jobs" to "who captures the gains." Watch for reskilling commitments that match the scale of the disruption, and for whether entry-level hiring, the on-ramp the Barometer shows, is changing shape, staying open to people without elite credentials. The companies and countries that treat workforce transition as infrastructure, not charity, will define the next phase.

Source: PwC

The Nature Conservancy has selected three AI tools to protect Indonesia's Savu Sea, a critical migration corridor for whales, dolphins, and sea turtles. Among them, Blurgs.AI gives local fishing communities fisheries analytics to manage stocks sustainably, alongside BlueOASIS, which monitors underwater biodiversity by sound, and Havoc autonomous vessels for wide-area surveys. What stands out is the design choice: these tools are built to work with local fishers and conservation staff, not replace them, and to run in remote waters with patchy connectivity. It puts conservation monitoring within reach at a scale human observation alone could never manage.

→ Read more: The Nature Conservancy

If an AI system could quietly handle one task in your work or daily life, reliably enough that you stopped thinking about it, what would you hand over first?

Hit reply. We read every response.