Where to put the human in the loop
Humans bottleneck agents when HITL means inspecting outputs instead of managing outcomes.
Notes RSS feed
-
-
The Tokenization of Labor
The AI Labor Playbook by Jules White is one of the clearest frameworks I’ve seen for how orgs should think about AI.
Some highlights:
AI is labor, not software. It doesn’t “run”, it works. And like any workforce, it needs to be led, trained, and deployed strategically.
Prompts = tasks. Tokens = wages. The “labor-to-token exchange” model is a powerful shift in thinking. You’re not using a tool, you’re hiring a temp.
Leading AI labor is a skill. It’s not just prompt engineering. It’s part comms, part systems thinking, part org design. Most people aren’t trained for it, but to stay competitive they must be. That’s the leverage we possess in the modern economy.
-
Natural language is the new OS
Language models are a trojan horse for a new type of software.
Natural language is becoming the new operating system, humanity’s native interface. Generative AI redefines software from a fixed tool to a dynamic collaborator.
Why it matters: The long-term shift isn’t AI replacing humans, it’s replacing traditional software. AI UX is the most significant design shift since the GUI. Rigid UX conventions like brittle menu structures will feel outdated. When interfaces adapt to us, those that don’t will be left behind.
-
The Real Bottleneck to Progress
The real bottleneck to progress is coordination, not knowledge.
We are not held back by a lack of good ideas. Humanity has had blueprints for energy abundance, clean water, nutrition, and education for decades. The deeper constraint is coordinated action. Incentive misalignment, local optima, and entrenched institutional power structures prevent execution. Most problems persist not because we don’t know how to solve them, but because we can’t agree to.
Why it matters: Breakthroughs will come less from novel inventions and more from novel coordination protocols, whether tech, legal, or cultural.
-
Confidently wrong
LLMs can be confidently wrong. That isn’t a bug — it’s a mirror.
They’re trained on human language, processed through neural networks modeled after the human brain. Of course they share our flaws — they’re made to communicate like us. Don’t try to use them as a truth machine. The leverage comes from the conversation - the space to think, reflect, and understand.
-
Knowledge half-life
When the half-life of knowledge is short, speed and adaptability are your leverage.
Go deep on what lasts - it’s worth the study.
Treat the rest like tools: learn what you need, use it to build, then move on. -
Global, timeless interconnectedness
Thoughts, feelings, and shared experiences—transcends time and space.
Thoughts and emotions that create bonds of attachment between us have no difficulty in crossing seas and decades, sometimes even centuries, tied to thin sheets of paper or dancing between the microchips of a computer. We are part of a network that goes far beyond the few days of our lives and the few square meters that we tread.
— The Order of Time, Carlo Rovelli
-
Everything is still possible
The past can be remembered or regretted, the future is full of possibility.
We cannot change the past; we can have regrets, remorse, memories. The future instead is uncertainty, desire, anxiety, open space, destiny, perhaps. We can live toward it, shape it, because it does not yet exist. Everything is still possible… Time is not a line with two equal directions: it is an arrow with different extremities.
— The Order of Time, Carlo Rovelli
-
Knowledge requires sacrifice
Wisdom is the perseverance to make the right sacrifices over time for a richer, more fulfilling life.
Perhaps the deepest insight that comes from thinking about later life as a chance to exploit knowledge acquired over decades is this: life should get better over time. What an explorer trades off for knowledge is pleasure.
— Algorithms to Life By, Brian Christian, Tom Griffiths
-
The illusory safety of inaction
The illusory safety of inaction leads to fragility; trying, learning leads to growth and adaptation.
To try and fail is at least to learn; to fail to try is to suffer the inestimable loss of what might have been.
― Chester Barnard