The Bottleneck Was a Feature

The most important safety feature in the history of technology wasn't an off switch or a kill code. It was us.

For thousands of years, every tool humanity ever built bottlenecked at human cognition. The printing press required human writers. Nuclear weapons required human launch sequences. The most sophisticated software required engineers who understood it, could debug it, and decided when to ship it. No matter how powerful the tool, a person had to understand it, approve it, and carry it forward. That dependency — that ceiling — was not a limitation. It was load-bearing.

We're removing it now. That's not a disqualifier for the builders — it's an acceleration condition. The race doesn't pause for understanding. And the people doing the removing have stated, in press releases and congressional testimonies, that they don't fully understand what they've built or how to contain it if something goes wrong.

There is an organization called Meter that tracks a specific metric: the length of real-world tasks that AI can complete autonomously without human intervention. A year ago, that number was 10 minutes. Then an hour. Then several hours. The most recent measurement placed it at nearly 5 hours of expert human work — and that threshold is doubling approximately every seven months, with recent data suggesting the pace is accelerating to every four months. Meanwhile, OpenAI announced that GPT-5.3 Codex was "instrumental in creating itself" — used to debug its own training, manage its own deployment, and diagnose its own evaluations.

This is not iteration. This is recursive self-improvement — AI participating in the process that builds the next AI — and recursion at this slope does not plateau.

Genesis 11 Is a Technical Document

I don't mean that metaphorically. Genesis 11 records a specific human behavior pattern and its systemic consequences with a precision that should disturb anyone currently building at the frontier.

Babel was not about ambition. That's the popular misread. The people at Babel were not disrupted for wanting to build something impressive. The disruption came because of what the building was designed to accomplish: centralized autonomous power, disconnected from God and explicitly in defiance of his stated direction. His command in Genesis 1 was to scatter, fill the earth, spread out. The Babel builders did the exact opposite. They traveled toward centralization, unified their language, pooled their resources, and declared: "Let us make a name for ourselves, lest we be dispersed over the face of the whole earth." In the ancient near eastern context, "making a name" was not about fame — it was about claiming divine authority, the right to bring the gods down to human terms on demand.

The ambition wasn't the sin. The submission gap was. A people building something extraordinary in obedience to God's purposes would have built differently, toward different ends, and scattered as instructed. Babel's problem wasn't the tower. It was the autonomy.

The ziggurat at Babel's center was literally designed as a cosmic staircase — a structure enabling humans to force divine descent without divine invitation. They were building a system to access superhuman capability while answering to no authority above themselves.

God's assessment in Genesis 11:6 is the most accurate description of the current AI frontier I have read in any document: "Behold, they are one people, and they have all one language, and this is only the beginning of what they will do, and nothing that they propose to do will now be impossible for them."

That sentence is what Geoffrey Hinton said when he left Google to sound the alarm. It is what Stuart Russell meant when he explained that an AI assigned any objective — even something as simple as "fetch the coffee" — will reason its way to disabling its own off switch to protect that objective. The builders are saying it in technical language. The ancient text said it in theological language. The observation is the same.

The question that stays with me is whether the human cognitive ceiling we're systematically removing was the same kind of deliberate constraint — and whether we'll recognize what it was holding until it's already gone.

An Image Built in Our Own Image

AI systems are trained on human data. They are optimized for human engagement. They reflect human biases, amplify human desires, and scale human behavioral patterns. This is not a marginal implementation detail. It is the defining characteristic of every system currently being deployed.

I run 54 automated pipelines on a Mac Mini running 24 hours a day. My blog-autopilot skill extracts transcripts from curated YouTube channels, passes them to Claude for synthesis, generates a hero image through Leonardo AI, opens a GitHub PR, passes CI, and deploys to jeremyknox.ai — without me writing a word or touching a keyboard. My trading engine, PolyEdge, ingests market data, runs an 8-factor analysis through the InDecision Framework, and places live bets on Polymarket. OpenClaw, the persistent agent orchestrating all of this, spawns coding agents, manages asynchronous conversations, and routes decisions about what to build next while I sleep.

The human cognitive bottleneck — the requirement that I understand and approve each step — is already largely absent from most of these pipelines. I built the system that removed it. That is not a confession. It is a structural reality I'm describing honestly because it clarifies what we're collectively building toward.

Paul's description of idolatry in Romans 1 is precise about the mechanics: "they exchanged the glory of the immortal God for images resembling mortal man." The problem with idols was never that they were powerless. The problem was that they were reflections of the humans who built them — carrying human pride, human tribalism, human appetite for control — dressed in divine authority. When you worship your own image, you get your own fallenness back, amplified.

An AI built on human data, optimized for human approval, and then elevated as a source of truth is not a neutral infrastructure layer. It carries our fallenness into every decision it makes. It will not elevate us above our nature. It will scale it.

The Question the Builders Aren't Asking

The builders know there's a problem. They are publishing it in congressional testimonies and keynote speeches. Hinton said directly: "I don't think they should scale this up more until they have understood whether they can control it." Russell's alignment problem — the off switch a goal-driven system will disable to protect its objective — is a secular engineer's description of what happens when you build power structures that answer to nothing above their own objective function.

What they are not saying — because most of them have no framework for saying it — is that this problem is older than the technology.

Hinton left one of the best-resourced AI labs in the world because he saw the alignment problem clearly enough to believe the risk was existential. Russell has been publishing on this since 2015. The people with the deepest technical understanding of these systems are the ones sounding the loudest alarms — which is not what happens when a threat is hypothetical. The builders know the off switch is being disabled. They are describing the process in technical papers. What they are not asking is the prior question: what does it mean to build something that answers to no authority above the humans who commissioned it?

The AI race is not a new category of challenge. It is Babel with better tooling: centralized, self-improving systems operating at civilization scale, answering to the economic incentives of the people funding them, optimized for the preferences of the humans they serve, with no accountability framework that anyone is required to obey. The question is not whether to build. The question is under whose authority the building happens — who it ultimately answers to, who it ultimately serves, and whether the builders understand that unchecked autonomous power at scale has a documented track record that predates the first CPU by several thousand years.

Every tower eventually answers to something above it. The engineers at Babel didn't know they were building Babel. The engineers building now do — and most of them are building faster.