Links
25 links — things worth reading, with a note on why.
saanyaojha.substack.com · via Saanya Ojha
Meta cut staff, force-reassigned thousands into roles they didn't choose, then tried to patch morale with perks. Ojha's point: AI transitions are won or lost on whether people are treated as participants or disposable resources — the agency argument applied to the adoption itself, not just the tooling.
linkedin.com · via Mille Bojer
Eight practitioner principles from a 3-day transformative scenarios workshop with 36 participants. AI handled synthesis and pattern-finding in the background; humans kept full ownership, presence, and creative source-hood. The key move: AI outputs are inputs to human judgment, never final products — and because the work isn't attributed to a person, participants hold it more lightly and push back more freely.
cutlefish.substack.com · via John Cutler
Cutler on the question every team is now living: how to adopt AI for the company's benefit without quietly trading away the agency that made the team work. The operating-model companion to the agency argument this site keeps circling.
linkedin.com · via Jazz Rasool
Where AI-assisted coaching meets regulation — the governance layer forming around human-development work done with machines.
every.to · via Every
A coaching-shaped use of the model: reflection partner, not answer machine. The centaur move applied to inner work.
wired.com · via Wired
The judgment, relationship and timing work doesn't automate away — it becomes the job. Pairs with the glue-work argument.
science.org · via Costello et al, Science
The optimistic counter-case: a well-aimed AI dialogue measurably shifted entrenched beliefs and it held. Agency designed in, not eroded — AI as interlocutor rather than oracle.
linkedin.com · via John Cutler
Sycophancy as the everyday face of cognitive surrender — when the model mirrors you, the verification step quietly disappears. A practical tell for when you've stopped thinking.
microsoft.com · via Microsoft Research
319 knowledge workers — the more they trusted the AI, the less critical thinking they did. The empirical anchor for why agency has to be designed in, not assumed.
oneusefulthing.org · via Ethan Mollick
The two ways to stay in charge of AI work — alternating (centaur) or interweaving (cyborg). The framing this whole site keeps coming back to.
hbs.edu · via HBS / BCG
Field study: people who kept judgment over both what and how (centaurs) were most accurate and upskilled; the self-automators who abdicated stalled.
roughtype.com · via Nicholas Carr
The sharpest critic's case: automation dependence quietly erodes the very competence it was meant to support. Worth steelmanning.
arxiv.org · via Margaret-Anne Storey
AI pays down code debt while running up two new kinds — eroded shared understanding (cognitive) and lost rationale (intent). Agency loss as a measurable org-health property.
AI-only planning caught 0% of novel risks; human+AI review caught 86.7%. Concrete evidence that collaborative verification beats solo AI.
oneusefulthing.org · via Ethan Mollick
A P&G field experiment where AI behaves like a teammate, not a tool — individuals with AI matched whole teams and silos softened. Multiplayer AI, measured.
a16z.com · via Andreessen Horowitz
The thesis that 2026 is when agents leave single-player mode and the collaboration layer becomes the moat.
sequoiacap.com · via Sequoia
If AI takes over hierarchy's information-routing job, what is the org chart even for? A provocation about rebuilding the company as an intelligence.
sloanreview.mit.edu · via Thomas Malone
Malone's collective-intelligence frame: groups plus machines as "superminds" that can out-think either alone — the optimistic bookend to the surrender literature.
cutlefish.substack.com · via John Cutler
Cutler's sharpest take on the theme: AI's value depends on whether it supports multiplayer collaboration or accelerates single-player isolation. Uses 4E cognition and stigmergy — shared context and trace-based coordination compounding into collective capability.
cutlefish.substack.com · via John Cutler
AI can reproduce the artifacts of glue work while quietly removing the connective human labour that made a team actually function. A warning about automating away the multiplayer layer.
cutlefish.substack.com · via John Cutler
Context is not a transferable input you merge into understanding — teams enact it through interaction. The single-player fantasy of "just give the AI all the context" misses distributed cognition.
ignorance.ai · via Charlie Guo
Nearly every AI tool is single-player: one human, one conversation, no shared context. Maps what changes when agents join the group instead of sitting in a private chat.
The technical reasons AI apps default to single-player — and what persistent, shared state would take to make them genuinely collaborative.
newsletter.aroundthebonfire.com · via Kevan Lee
A crisp framing of the single-player to multiplayer shift in how teams actually adopt AI together.
When every post carries tags, the tag index stops being navigation and becomes a mirror — you see where your attention actually went. A small consistent act accumulates into judgment you couldn't have designed up front.