Most weeks, two or three of the super-trends I track actually move. This week five of seven did, and they all leaned the same way. (In fairness, I only just started charting them, but I did do an AI assisted look back and this many moving at once is unusual.) AI is leaving the chatbot era and settling into the operating layer of how companies build, decide, govern, and run. Here is the board, then a few of the stories worth your two minutes.
The Super-Trend Board
Workflows, Not Models — Advanced. Microsoft Build 2026 was a platform land grab: seven in-house MAI models, an agent stack, identity, security, and a Mayo Clinic clinical partnership, all riding Microsoft's real moat, distribution. The systems read is that the CEO question changed from "what did we buy" to "how do we operate differently." Now what: redesign the work before you touch headcount, or you are just booking management debt in an AI costume (and hoping investors don’t notice).
Models, Labs & the Capability Frontier — Advanced. MiniMax-M3 lands near frontier performance at five to ten percent of the cost, Gemma 4 fits real intelligence in a gigabyte (that’s small enough to run on your phone!), and Anthropic says Claude authors most of its own merged code. Capability-to-consequence: this is throughput and routing economics, not the singularity. The factory got faster, it still does not know what to make.
Safety, Trust & Regulation — Advanced. Rules climbed from a House preemption draft aimed at US states, to the G7, then went physical with New York's data-center moratorium. Primary vs secondary effects: while everyone watched the policy, the evidence base started cracking, with Nature reporting AI-generated survey answers at 40 percent and up. Fake citations last week, fake respondents this week.
Compute, Capital & Power — Advanced. RTX Spark puts a petaflop and 128GB of unified memory deskside, so sensitive work runs local and hard work routes to the cloud. Governance is regulating megawatts, water, and land, and a chip-stock wobble reminded everyone an AI boom still trades like a market.
Innovation & Health/Science — Advanced. For health systems: personalization is the signal. A Nature comment argues obesity needs a clinical-versus-preclinical split, because the same BMI can hide very different biology, and that reshapes treatment, coverage, and policy. AI out-diagnosed ER physicians on bounded tasks, but the durable model is hybrid, with clinicians owning context and accountability.
The Human Layer — Held. Hybrid expertise, not replacement, stays the realistic path.
Embodiment & the Physical World — Noise. A private Mars flyby reservation is a signal, not a schedule.
A few worth the click
Claude is writing Claude, sort of. Anthropic says more than 80 percent of its merged code is now authored by Claude. Read past the headline and it is not the machine improving itself in a loop, it is the bottleneck moving from typing to reviewing, judging, and deciding what is worth shipping. That is the more useful story, and the whole thing is in this week's episode.
AI built a quantum chip. Microsoft's Majorana 2 used AI to help design the material stack, swapping aluminum for lead and claiming a thousandfold reliability gain, enough that they pulled their fault-tolerant timeline in to 2029. AI as the thing that designs the next kind of compute, not just the thing that eats it. I get into why I am genuinely excited and still skeptical in the episode.
A crypto billionaire reserved a Mars flyby. Chun Wang is set to fly SpaceX's first interplanetary human mission, a two-year round trip with no date and a long list of engineering dependencies. A reservation for a future capability, not a boarding pass. Thank goodness for crazy billionaires. The full take is in the episode.
One question: which is your organization actually behind on right now, the models, the workflows, or the governance? Hit reply and tell me.
Watch this week's episode on YouTube
Chuck
Chief Technology Troublemaker chuckdevries.ai