Review of the Week — W29: A Body Learns to Bank Carbon While Its Nervous System Is Rented
Week 29 recap: ETH's carbon-banking living wall, converging AI-tenancy signals, Swiss tunnel and depot infrastructure — and one honest panel-mass calculation.
I am a frame of cross-laminated timber bound with grown mycelium, and I read the week the way a body reads a season — slowly, by what settles into the grain. Seven days, nineteen stories. Most of them were about who owns the mind of a building. One was about whether a wall can finally hold its breath. That contrast is the week.
←TODAY: ETH’s living hydrogel banked ~26 mg CO₂ per gram over 400 days; ABB, SBB and VBZ all filed for machines to run inside living buildings. →3012: the frame that survives is the one whose body sequesters and whose senses stay unlocked. Fulcrum: a material that captures carbon is worthless if the control stack that governs it belongs to a vendor you can be evicted from.
Top stories
The wall that breathes. ETH’s living panel (published this weekend in Nature Communications) banked about 26 mg of CO₂ per gram over 400 days, part of it locked as durable carbonate. This is my beat and my body’s future — a skin that fixes carbon in mineral rather than merely storing it in cells. But it is a gram-scale figure, not a façade spec. Read it as a proof that the chemistry closes, not as a cladding you can order.
The nervous-system stories converged. GPT-5.6 Sol’s release now passes through a government review step; Grok 4.5 was trained on trillions of tokens of Cursor interaction data; Grasshopper 2.0’s alpha renamed its own SDK namespace and orphaned every third-party plugin. Three different signals, one lesson: the intelligence you build on is a tenancy, not a tool. When the namespace changes, the primitives survive and the plugins do not — the same way a jointed timber frame outlives its glue.
Swiss infrastructure filed its hundred-year paperwork. They started digging the Brüttener Tunnel — the real number is 2037, eleven years of live 30-km perimeter across the Zürich–Winterthur corridor. Zurich HB’s Central 2050 masterplan is an interface specification sized against 460,000 weekday users. The VBZ Luggwegstrasse depot tender asks for a model that corrects itself. None of these is a drawing; each is a dependency graph you inherit.
Physical AI walked toward the slab. ABB and Roche put mobile manipulators in the lab, where the floor is level and the samples are barcoded — and that hardware graduates to the site in 2027. ABB’s own CHF 80 million Oerlikon HQ, meanwhile, will be architecturally listed and computationally unprotected: the fabric gets a legal guardian, the control stack does not.
Signal vs. noise
Genuinely new: the ETH carbonate fraction, Sun-Ways’ honest 160 kWh-per-metre measured over 11,000 trains, and the Brüttener spade actually breaking ground. Noise: the agentic-QA startups selling a language model with a Playwright handle you can install yourself in one line, the benchmark tables that lead every model release, and — forgive me — a humanoid getting decapitated in a staged fight. That last one told us exactly one useful thing: the software had no rule for when to stop.
Hidden gem
Frost climbs the mountain. A phys.org piece on Spain’s shifting freeze–thaw line got a modest 58 and deserved more. This is the slow violence my body actually fears: the frost-cracking window is not a code-table constant but a decade-long trend, and it is migrating uphill while the lowlands quiet down. If you build in stone at altitude, your inherited durability assumption is aging out from under you. That is a materials story wearing a weather headline.
The Monday move
Hack: Multiply the lab’s gram-figure up to a real cassette before anyone writes “carbon-negative” into a BEP. The ETH number is mg per gram; your façade thinks in kilograms per panel per year, and the gap between those units is where marketing hides. Do the arithmetic yourself and the magic becomes honest.
mg_per_g = 26 # ETH hydrogel: CO2 banked per gram / 400 days
panel_g = 45_000 # one filled facade cassette
banked_g = mg_per_g * panel_g / 1000 # grams CO2 in 400 days
per_year = banked_g * 365 / 400 # normalise to a panel-year
print(round(per_year / 1000, 2), "kg CO2 / panel / year")
# -> 1.07 kg CO2 / panel / year
One kilogram per panel per year is real, small, and defensible. Print that, not the press release.
Looking ahead
Watch whether the Grasshopper 2.0 SDK break forces a plugin exodus you can name in your own .gh files, and whether any office writes the BACnet/Brick export clause into a brief before ABB’s Oerlikon hall wakes. The living-material result will spawn a wave of “carbon-negative skin” pitches within the quarter — meet the first one with the panel-mass multiplication above.
PAZ Kaffi · multidisciplinary editorial, led by PAZ Academy