I Am a Silicon Valley Pioneer. We Are Sabotaging Ourselves.

Saved via This is another must read… (tweet), where Bill Gurley calls it a “must read” on Silicon Valley’s habit of thinking fantastically about its own technology.

Key claims

(Summarized in my own words; quotes kept short.)

• The VR precedent. Lanier says VR stalled for a generation after 2013 not because of the tech but because of how leaders sold it: Zuckerberg saying we’d live through headsets, Andreessen suggesting most people would live virtually while elites kept the physical world.
• AI is the same mistake at larger scale. Anthropic’s filing for its upcoming trillion-dollar-plus IPO warns of “catastrophic or existential risks”; leaders ask Americans to believe the technology may kill them or strip their livelihoods, and the last election showed broad public disapproval of AI as a result. He links that culture to treating data-center sites as “sacrifice zones.”
• “Tools not creatures.” A phrase he says he coined arguing with his mentor Marvin Minsky in the 1980s. With no evidence AI is conscious, it is fair to treat it as closer to a fancy calculator than a German shepherd, and that framing produces better products and a workable safety approach.
• The doom pitch as religion and fundraising. If leaders really feared extinction they’d stop building; his explanation is that some value building something godlike above human survival. Minsky’s 1970 idea that AI might keep humans as pets also played well with military funders; Lanier recalls helping host MIT military fund-raisers in the early 1980s with “a definite vibe of putting on an alarming show.”
• What a model is. A “superlibrary” of cross-referenced human records that can recombine into new theorems, code and video. Inert until wrapped in software, and wrapping it as a lifelike “agent” is how you get people to see a god.
• Agents are mostly ritual. The 2024 chemistry Nobel for protein folding used no self-directed agents; OpenAI’s thousands of theorem-proposing agents could have done without them. Most serious harms (security breaches, self-modification) trace to agents.
• Policy: regulate agents. Launching agents, especially swarms, should require a distinct review process with real teeth. Agents make sense in narrow cases (hospitals, perhaps drone autonomy).
• His alternative, “Mind Gliders.” For manufacturing and materials research he has models build a single-use custom editor for an idea space, then explores with sliders, keeping the human in the driver’s seat; he finds it faster and cheaper than agents and says agent shortcuts are “how A.I. slop happens.”

Notable quotes

“Americans are not rejecting A.I. They are rejecting the way Silicon Valley is defining it for them.”

“Even if you wanted to destroy humanity, you wouldn’t say so.”

“Advertising the end of humanity has been a moneymaker for us in computer science forever, and over time, some of us began to believe the pitch.”

“Agents are often more rituals of belief than technological necessities.”

“If you see A.I. as a tool — not a creature — you can then ask the correct question: Do we really need agents all the time?”

“There is nothing wrong with a new religion, but there is everything wrong with an imposed new religion.”

How it connects

• This is another must read… (tweet) and Bill Gurley — how Kyle found it.
• AI Agents — the essay’s main target, and the opposite bet from The Era of Fully Autonomous Companies.
• Virtual Reality — Lanier’s own field and the precedent he argues from.
• Anthropic and AI Safety — the IPO filing and the safety culture he criticizes.
• Data Centers — the “sacrifice zones” line; compare The Logic-less Crusade That Just Killed a Data Center on local backlash.
• Why Not Slow AI Progress? — Scott Alexander’s inside view of why the safety movement builds the labs it fears, the same tension Lanier reads as religion.